The data arrived on time. The decision didn’t.
Your ERP, planning system and warehouse each hold part of the answer. Putting them together still takes meetings, spreadsheets and days you don’t have. ADIP — the A2go Decision Intelligence Platform — coordinates the decision across all of them, on the systems you already run.
Supplier confirms a 9-day slip on a resin lot. ERP has the PO. MES has the run. Neither has raised anything.
Three options costed against the schedule that will actually run. Protecting the contract account costs 4 days of a lower-tier order.
Planner approves option 2. New promise date and resequenced build write back to the ERP and the MES.
The rule is now yours: this account is protected against this supplier in this season. The next call starts there.
Clear to Build. A work order releases Friday. One BOM line out of 214 is short, and a $438,000 shipment is at risk. Equivalent stock exists under another customer’s part number — if the ontology can prove it. Open the worked decision → Opens in a new tab, so this page stays where it is.
Databricks published how a coordinated A2go agent system took a regional demand and pricing forecast cycle from 28 hours to under one hour, running on the customer’s own lakehouse with no migration. The wider coordinated agent system at that customer carries a $200M+ [figure cleared 8/30 — final wording aligns to the slide at port] annual profit impact. Read the A2go–Databricks story →
Numbers from live deployments, not a projection.
Regional demand and pricing forecast cycle at a major Brazilian beef producer — 65 plants, 17 sales channels, 150,000+ customers.
Published by DatabricksAnnual profit impact at the same producer, from the coordinated agent system across demand, production, S&OP and pricing — 60,000+ SKUs simulated in minutes.
A2go–Databricks story →Master production scheduling at a ~$500M industrial manufacturer and distributor. 24 spreadsheets replaced by one coordinated run.
Customer deploymentSee how each was measured Find the operation that looks like yours
Four things have to happen. Most stacks only do the first.
Your ERP, WMS and planning tool each hold part of the answer and none of them is responsible for the whole one. ADIP is the layer that carries a decision from signal to action to memory — on the systems you already run.
Your data, where it already lives
ADIP reads from the ERP, WMS, OMS, planning tool, MES and supplier feeds, reconciles the definitions you have today, and streams each agent exactly what it needs.
Agents that reconcile, not compete
Demand informs the plan, the plan informs inventory, inventory informs the promise, and the promise pushes back as a constraint. One recommendation reaches your planner, not four.
Approved, then written back
Every recommendation carries the trigger, the alternatives, the constraint that applied and the expected impact. A person approves, edits or rejects. Approved actions write back to the system of record.
Your judgment, kept
Which customer you protect when two orders compete. Which supplier’s lead time to distrust in August. When the rule gets broken. None of that is in your ERP — it is in people, and it retires.
Why agentic AI.
Three things are sold as AI. Only one of them makes a decision, and that difference is the whole argument.
“We already have AI.”
Probably true. Almost certainly not this. Three different things are sold under one word, and only one of them closes a decision.
It predicts
Statistical forecasting, demand sensing, anomaly detection. Often already inside your ERP or planning tool. Produces a number.
It explains
Chat assistants and productivity AI. Summarize the report, draft the email, answer a question about the data. Produces language.
It decides and acts
Reasons across data, rules, constraints and tradeoffs from several systems at once, and returns a ranked, costed recommendation ready to approve.
The full comparison, with what each one cannot do See a decision closed end to end
Nothing migrates. Nothing gets replaced.
ADIP sits above the stack, never inside it. Your ERP is still the system of record, still transacting, still doing the job you bought it for. What crosses the boundary is a copy of the data an agent needs at that moment — and an approved decision on the way back.
- Multi-ERP is normal, not an exception. Reconciled against the definitions in your systems today, with no requirement to standardize the estate first.
- Your lake or warehouse gets more valuable. Snowflake, Synapse, Redshift, BigQuery, Delta — read through governed interfaces, not replaced.
- First domain live in 8–12 weeks. One environment, one decision, one measured result before you commit to a second.
- Remove ADIP and nothing breaks. Your systems keep running exactly as they did. You lose the coordination, not the business.
You have the same volatility, but less room to absorb it
A $20B enterprise can survive a two-year data program and a consulting engagement measured in years. A $500M manufacturer has the same tariffs, the same weather, the same three inherited ERPs from the last three acquisitions — and considerably less room to absorb a bad quarter.
The incremental path is not the compromise version for smaller budgets. It is the better path at any size, and at some sizes it is the only one available.
The shortage was never insight. It was everything that happens after it. Mike Romeri, CEO — 40 years in supply chain operations
The profile
Manufacturers and distributors moving physical goods, roughly $400M to $2B in revenue, where planning, inventory, procurement, scheduling and fulfillment decisions carry real financial consequence.
The complexity
Multi-site and multi-ERP. Systems accumulated through acquisition. High SKU counts, constrained capacity, and regulatory obligations that don’t bend.
The signals you already recognize
- Planners spend their time reconciling rather than deciding
- Stockouts and excess inventory at the same time
- OTIF below 90%
- An S&OP cycle nobody believes by week two
Mike Romeri
CEO & Founder
- 40 years in supply chain
- 20+ years Partner, PRTM (now PwC)
- Co-founded OPS Rules with MIT’s David Simchi-Levi — acquired by Accenture
Cesar Oliveira
Chief Operating Officer
- 20+ years ERP transformation
- 25+ enterprise go-lives: SAP, Oracle, Epicor
- Architect of ADIP
Stephen Hutson
CTO & Co-Founder
- 9 years Principal, PwC-PRTM
- Nominated Distinguished Engineer, IBM
- Ran Watson AI & Commerce Analytics, APAC
A2go
Founded 2018
- Built natively on Databricks
- Governed through Unity Catalog
- Delivery team in the US and Brazil
Bring the decision that costs you the most.
Not a platform evaluation. One 30-minute session on where a single decision stalls across your systems, what it is worth, and what a first deployment against it would look like. You leave with a scoped first domain and a timeline.
No migration proposal. No platform commitment.
One governed decision layer over the systems you already run.
ADIP has four jobs. It reads your data where it lives, coordinates a set of purpose-built agents across it, routes the result to a person for approval, and keeps how your company decides so the next recommendation is better than the last.
Your data, made AI-ready where it already lives.
The data orchestration engine reads from your source systems, reconciles the definitions that exist today, and streams each agent exactly what it needs — in the shape it needs, at the moment it needs it. No migration and no new destination.
- Ingestion
- Reconciliation
- Streaming
- Storage
- Governance
- Monitoring
Snowflake, Azure Synapse, AWS Redshift, BigQuery, Cloudera, Delta, Iceberg, a data fabric, or a mix accumulated across acquisitions — ADIP draws from it through standard governed interfaces. Your existing investment stays in place and usually gets more valuable: static stores become live, agent-ready feeds instead of extracts somebody refreshes on a schedule. It also doesn’t wait for a finished master-data program — it starts delivering against the definitions you have and supports that work over time rather than blocking the business until it lands.
Coordinated, not centralized.
Each agent is built for a specific supply chain decision, and none of them works alone. One agent’s answer becomes context for the next, continuously, so what reaches your planner is a single reconciled recommendation rather than four competing ones.
- Purpose-built, not general purpose. Every agent is tied to a specific decision and a measurable financial metric.
- They orchestrate continuously. Demand reaches planning, planning reaches inventory and order promising, without waiting for a meeting.
- They reason on governed data. The demand number in the schedule is the demand number in the promise. Disagreements between agents are about tradeoffs, never about whose data is right.
- Conflicts resolve against your rules. When inventory and service pull against each other, the Judgment Layer decides how the conflict is weighed.
- What triggered it, and when
- The alternatives that were considered
- The constraint that applied
- The expected financial impact of each option
- A ranking, and the tradeoff behind the ranking
Approval-ready — not another alert.
A person approves. Then it writes back.
Routine decisions inside limits you define can run automatically. Everything else arrives in front of the person who owns the call, with the projected outcome attached, while the order can still be re-promised, the shipment re-routed, the buy adjusted.
Your ERP remains the ERP. ADIP does not become a second place where the truth lives.
Nothing acts outside the limits you set
Every deployment defines three things before an agent runs: which decisions may be automated, which must be approved, and what the agent may never do regardless of confidence. Those limits are configuration, not a policy document — the system cannot exceed them.
Every decision is logged with who approved it, when and why, and the log is exportable for audit.
See a write-back in a worked decision What a decision package contains Security and governance in full
The Judgment Layer.
Your best planner knows which customer you protect when two orders compete, which supplier’s lead time to distrust in August, and when the rule gets broken. None of that is in your ERP. It is in people, and it walks out at retirement.
Your Judgment Layer carries your company’s own decision logic — the approvals, the rejections and the reasoning behind them. It is never shared across customers and never used to train a model any other customer touches. Without it, AI is automation. With it, expertise compounds instead of retiring.
First domain live in 8–12 weeks, then at your pace.
- Assessment · 1–2 weeks
- Connect · 2–4 weeks
- First domain · 4–6 weeks
- Write-back · on your signal
- Expansion · your roadmap
Coordination is the product, and every pain point has to be tied to a business outcome before anyone builds anything.Cesar Oliveira, COO — 25+ enterprise ERP go-lives
Start with the decision that hurts most.
One session, your people, your systems. You leave with a scoped first domain and a timeline, not a platform proposal.
Controlling your AI starts with the ontology.
A model trained on the internet knows language. It does not know that a “case” in your business means a pallet, that account 4021 is never short-shipped, or that your Ohio supplier’s August lead times are optimistic and everyone plans around it. The ontology is where all of that gets written down. It is what makes an agent reason like your company instead of like a general-purpose model.
Nouns, verbs, and the rules that govern both.
An ontology is a working model of how your operation is put together. It has three parts, and an agent needs all three before it can produce a recommendation anyone would approve.
What exists in your business
SKUs, lots, plants, lines, orders, promises, suppliers, BOM levels, channels — with your definitions, not generic ones. Three ERPs can each hold something called a customer and mean three different things. The ontology settles which one is the customer.
What may actually be done
Resequence a build. Reallocate a short lot. Re-promise an order. Release a purchase order early. Each one is a real action with real consequences, defined as something an agent can propose and a person can approve, edit or reject.
When, by whom, within what limits
Which verbs apply to which nouns, under which constraints, requiring whose approval. Allocation policy, lead-time guardrails, SLA commitments, escalation thresholds, and the actions an agent may never take regardless of how confident it is.
Nouns alone give you a data model. Nouns and verbs give you a workflow. Only when the rules are attached does a recommendation arrive with the constraint that applied and the alternatives that were ruled out — which is the difference between an answer and a decision someone can defend to a customer.
One work order. 214 BOM lines. One of them says no.
A work order for 300 units releases Friday against a customer promise the following week. Two hundred and twelve lines are covered. One capacitor line is short 1,160 after a purchase order slipped nine days, and a $438,000 shipment is at risk. Equivalent stock exists — enough of it — but it sits under another customer’s part number, on another program.
Recommending that transfer is not a data question. It is an ontology question, and here is everything that had to be joined before an agent could propose it.
Two part numbers, one part
Approved on both AVLs
What is actually yours to move
Real movement, not assumed FIFO
Who else is counting on it
What it may never propose
Today that join is a five-to-six-step hunt across the ERP, the warehouse system, quality and the customer master, done by whoever knows where to look. In the ontology it is one governed read — and the resulting recommendation carries the proof of why the transfer is safe, which is what makes it approvable and auditable.
Open the full worked decision More decision examples What the decision package contains
Agents raise the price of ambiguity.
A planner absorbs ambiguity without noticing. The Tuesday number is the reliable one. The Mexico plant reports a week behind. None of it is written down anywhere.
An agent has no such instinct. It reasons confidently from whichever definition it found first, and a wrong answer looks exactly as well-formed as a right one.
Ontology-first platforms are usually developer platforms where your team models the enterprise before anything produces a number. A2go arrives with a supply chain ontology already modelled — the entities, events and metrics that recur across forecasting, planning, inventory and order promising, because they are the same in every manufacturer and distributor. What is specific to you is your rules, your priorities and your guardrails. Ontology makes your systems operate by them.
Your first domain is live in weeks rather than after a modelling program, because the general structure is already there and only your part of it has to be written. Every decision added after that extends the same model instead of starting a new one.
We start from decisions, not from your schema.
Modelling an entire enterprise before anything works is how two-year programs happen. A2go works backward from the decisions that cost you money and models only what those decisions need: the rules that govern them, the entities they touch, and the fields they actually draw on.
Your business rules are held in the ontology, not hard-coded separately inside every agent. Change an allocation policy once and every agent that touches allocation obeys the new version on the next cycle. Rules written into individual agents drift apart the moment the business changes, and nobody finds out until two agents recommend opposite things.
The ontology holds what you can state. The Judgment Layer catches the rest.
Allocation policy, lead-time guardrails and escalation thresholds can be written down in a workshop, and they go into the ontology during deployment. The overrides, rejections and edits your planners make afterward cannot be, and those accumulate into decision memory specific to your company. Both stay in your tenant. Neither is shared across customers.
The ontology governs. The agents reason.
The model of your business
- Entities, relationships and events
- Business rules, policies and constraints
- Which data each decision may draw on
- Roles, permissions and approval thresholds
- Explainability and the audit trail
- Coordination across agents
The reasoning for one decision
- Reason across the data that decision requires
- Cost the alternatives against real constraints
- Run scenarios and comparisons
- Assemble the recommendation and its rationale
- Carry the approved action back to your systems
Bring one decision and we will model it with you.
Thirty minutes on a single decision in your operation: the entities it touches, the rules that already govern it, and what an agent would need to know before it could recommend anything you would sign.
See one closed, end to end.
Not a feature tour. Each of these pins a single decision and shows the whole anatomy of it — the signal that opened it, the systems the agent consulted, the recommendation it made, and what each alternative would have cost. Open one and you will know what a closed decision looks like, whether or not the scenario is yours.
Every one has the same five parts.
This is the structure that transfers. The scenario changes; the shape of a closed decision does not.
The event that made this a decision — a slip, a shortfall, a drift past a control band.
Which of your systems held each piece of the answer, and what the agent read from each.
One reconciled action, with the constraint that applied and the rule that governed it.
The options not taken, priced. This is the part that makes a recommendation defensible.
The approved, edited or rejected outcome returning to the system of record with its audit trail.
Two decisions, closed in full.
ATP short-ship
A priority customer is about to be short-shipped. Here is who ships, who waits, and what it costs.
A port strike halts inbound flow and committed orders exceed available-to-promise by 3,800 units. Four allocation strategies ship the same 8,600 units — with very different penalty exposure.
Open the worked decision →
SCOR Source / Transform · Material Availability agentClear to Build
The work order releases Friday. One BOM line out of 214 says no. What unlocks it, and at what cost?
A supplier slip leaves one capacitor line short while 212 lines keep arriving to plan. Equivalent stock exists under another customer’s part number — if the ontology can prove it. Transfer, expedite, or wait, priced as one decision.
Open the worked decision →
That example turns on whether equivalent stock under a different part number can be proven equivalent. That is an ontology question, not a data question. How the ontology settles it.
Five more being worked now.
Each one pins a different decision family. If the one you need is not here, that is the useful thing to tell us.
Supplier OTIF drift
A key supplier’s on-time-in-full is drifting. When does a trend become a decision?
In preparation
Forecast deviation
Four weeks of MAPE creep on one family. Re-plan now, or ride the noise?
In preparation
Capacity spillover
Line 2 is over commitment for the week. What spills, where, and at what cost?
In preparation
Inbound reroute
A port closes overnight. Which inbound lanes move, and what does each day of dwell cost?
In preparation
Safety-stock rebalance
Service targets moved. Where does the buffer belong now — and what frees up?
In preparation
The survey takes about ten minutes and asks how your forecast, schedule, inventory and promise decisions actually get made today. Send it back and we will work your decision as an example, and walk it with you in the session. Take the survey.
Your scenario is different. The shape is not.
The families are shared
Every example is pinned to a SCOR process, so a Source decision reads as a Source decision whether the shortage is a capacitor or a cut of beef. The framework is ASCM’s, not ours.
The hard part is always the tradeoff
Short-shipping a priority customer and spilling capacity off line 2 are the same problem underneath: two commitments, one constraint, and a number attached to whichever one you break.
What reaches the planner is identical
One reconciled recommendation with the trigger, the alternatives, the constraint and the expected impact attached — ready to approve, edit or reject. That does not change by industry.
Bring the decision you wish were on this page.
Tell us where your decisions stall and we will work yours the same way — the signal, the systems, the recommendation and what the alternatives cost — against your own operation.
Any ERP. Any planning system. Usually several at once.
The question we get is rarely “do you support X”. It is “we have three ERPs from three acquisitions, a planning tool nobody likes and a warehouse system from 2009”. That is the normal case, and it is what ADIP was built for.
Systems of record
- SAP (ECC and S/4HANA)
- Oracle
- Microsoft Dynamics
- Epicor
- Infor
- NetSuite
- QAD, IFS and industry-specific ERPs
- Homegrown and legacy systems via database or API
Operational systems
- Warehouse management (WMS)
- Order management (OMS)
- Manufacturing execution (MES)
- OT and plant floor historians
- Transportation management (TMS)
- CRM and commerce platforms
- Supplier portals and EDI feeds
Data platforms
- Databricks (native)
- Snowflake
- Azure Synapse
- AWS Redshift
- Google BigQuery
- Cloudera
- Delta Lake and Apache Iceberg
- Existing data fabrics and legacy warehouses
ADIP draws from it rather than replacing it. Your existing investment usually gets more valuable: static stores become live, agent-ready feeds instead of extracts somebody refreshes on a schedule.
Read access and a technical contact.
| Step | What we need from you | Typical time |
|---|---|---|
| Field mapping | A walkthrough of the systems that hold the decision, with someone who knows the data. We document the exact fields required. | 3–5 days |
| Connection | Read-only credentials or a governed view. No schema changes, no agent installed in your ERP. | 1–2 weeks |
| Reconciliation | Nothing — we reconcile entities, hierarchies and time series against the definitions you already use. | 1–2 weeks |
| Validation | A planner to confirm the numbers match what they see in the source system. | 2–3 days |
No master-data programme is required first. ADIP works against the definitions that exist in your systems today and supports that work over time rather than blocking on it.
Tell us what you run.
Bring the actual list, including the systems nobody wants to admit to. We will tell you in one session what connecting to them involves.
Four pillars. Every agent sits in one.
Supply chain decisions cluster into four families, and they are not independent: what demand is really doing changes what the plants should run, which changes what you should hold, which changes what you can promise. The agents are organized the same way.
Forecasting & Planning
Ahead of the report that says so. Internal and external demand sensing, supplier and distributor forecasting, SKU-level forecasting, and full-horizon forecast and planning.
Moves forecast accuracy and the cost of reactive replanning- Internal Demand Sensing
- External Demand Sensing
- SKU-Level Forecasting
- Supplier Forecasting
- Distributor Forecasting
- Forecast & Planning (full horizon)
Operational Planning
Master production scheduling and S&OP optimization, purchase order excellence, lead-time and safety-stock optimization, and scenario analysis across both.
Moves schedule attainment and planning cycle time- Master Production Scheduling
- S&OP Optimization
- Purchase Order Excellence
- Lead-Time Optimization
- Safety-Stock Optimization
- Scenario Analysis
Supply & Inventory Optimization
Classification, slow-moving inventory, multi-echelon inventory optimization, vendor-managed inventory opportunity, and supplier reliability.
Moves working capital and inventory health- Inventory Classification
- Slow-Moving Inventory
- Multi-Echelon Inventory Optimization
- Vendor-Managed Inventory Opportunity
- Supplier Reliability
OTIF Optimization
Capable-to-promise, promise-date jeopardy, unexpected customer orders, customer promise intelligence, and revenue and OTIF optimization.
Moves on-time in-full and revenue at risk- Capable-to-Promise
- Promise-Date Jeopardy
- Unexpected Customer Orders
- Customer Promise Intelligence
- Revenue & OTIF Optimization
Context passes forward. Constraints pass back.
A single agent answering one question well is useful. A set of agents answering across the chain, against one another, is what changes the decision.
Demand informs the plan
A shift in sensed demand does not stop at the forecast. It reaches master scheduling as a change in what the plants should be building, before the monthly cycle would have surfaced it.
The plan informs inventory
A resequenced build changes what needs to be on hand and where. Safety stock, replenishment and slow-moving positions are re-evaluated against the schedule that will actually run.
Inventory informs the promise
What is genuinely available — across plants and DCs, net of what the schedule has committed — is what capable-to-promise answers with. Not a static availability figure.
The promise pushes back
A promise that cannot be held without breaking a higher-priority commitment returns as a constraint on the plan. The loop closes rather than escalating to a person.
Left alone, each pillar optimizes itself: inventory carries less, service carries more, the plant runs the longest campaigns it can. Each answer is locally correct and the combination is bad. The coordination is what stops four right answers from producing one wrong outcome.
Every recommendation arrives as a decision package.
Not an alert, and not a number on a dashboard that someone still has to explain. A decision package is the whole case for one action — explained in full, saved, and auditable. It is what makes a recommendation something a planner can approve in minutes and defend to a customer a month later.
What opened it
The signal that made this a decision, and the constraint it collided with. A nine-day supplier slip. Committed orders past available-to-promise.
Which systems answered
Every system the agents read for this decision and what each one contributed — ERP, WMS, OMS, planning, MES, supplier feeds.
One reconciled action
Not four competing suggestions from four agents. One action, with the rule from your ontology that governed it stated alongside.
Priced, not listed
What the options not taken would have cost. This is the part that turns a recommendation into a decision someone is willing to sign.
The metric it moves
Which financial or service measure this action changes, and by how much — stated before the approval, not reconstructed afterward.
Kept and exportable
Who approved, edited or rejected it, when, and why. Retained with the package, exportable for audit, and fed back into your Judgment Layer.
A recommendation nobody can interrogate does not get approved, and one nobody can reconstruct six months later does not survive an audit. The decision package is what makes an agent’s output usable by a business rather than interesting to a data team. Every package is logged, and every approval and override strengthens the decision memory the next one draws on.
See decision packages worked in full The rules that govern them
Every agent exists because a decision was costing someone money.
These agents were not adapted from general-purpose building blocks and pointed at supply chain afterward. Each starts from a decision manufacturers and distributors actually lose money on.
Multi-plant master scheduling
Sequencing across sites that share supply, capacity and customers — the decision our first production deployment was built on.
Multi-level bills of material
One purchased component moving through three BOM levels into four finished goods, traced without a planner rebuilding it in a spreadsheet.
Capacity and constraint reasoning
What the line can actually run this week, against what the plan assumes it can.
Batch variability and yield
Process environments where the input is not uniform and the output cannot be assumed — including perishability and shelf-life constraints.
MES and OT signals
What the floor knows, in the same decision as what the ERP recorded — rather than in a separate report nobody reads in time.
Multiple ERPs
Reconciled against the definitions that exist today, with no requirement to standardize the estate first.
Which decision would you start with?
Bring the one your team loses the most time or margin on. We will show which agents would touch it, what they would reason on, and what would reach your planner.
Three things are sold as AI. One of them decides.
Most midmarket manufacturers already have AI somewhere — a forecast engine in the ERP, an assistant in the productivity suite, a model a data scientist built two years ago. All of it is real. None of it closes the gap between knowing and acting.
Traditional AI predicts
Statistical and machine-learning models: demand forecasting, anomaly detection, propensity scoring, demand sensing. Frequently already embedded in your ERP or planning suite.
What it produces: a number, and usually a confidence interval.
Where it stops: it has no view of capacity, no view of the order book, and no authority. A better forecast that reaches a planner who still has to reconcile it against four other systems has not shortened the decision.
Generative AI explains
Large language models and chat assistants: summarize the variance report, draft the supplier email, answer a question about last quarter in plain English.
What it produces: language, from data someone already assembled.
Where it stops: it reasons about text, not about constrained tradeoffs. Ask it what to do when two orders compete for the same lot and it will describe the considerations rather than cost the options against your actual capacity.
Agentic AI decides and acts
Purpose-built agents that reason across data, business rules, constraints and tradeoffs from several systems at once, coordinate with each other, and return a ranked recommendation with the reasoning attached.
What it produces: a decision package — the trigger, the alternatives considered, the constraint that applied, the expected financial impact.
Where it goes: to the person who owns the call, while the order can still be re-promised. On approval it writes back to the system of record.
What each generation can and cannot do
| Capability | Traditional | Generative | Agentic |
|---|---|---|---|
| Predict what demand will do | Yes | No | Yes |
| Explain a result in plain language | No | Yes | Yes |
| Reason across ERP, WMS and planning at once | No | No | Yes |
| Cost several options against real capacity | No | No | Yes |
| Settle a tradeoff between two functions | No | No | Yes |
| Produce an approval-ready recommendation | No | No | Yes |
| Write an approved action back to the ERP | No | No | Yes |
| Retain how your company decides, and reuse it | No | No | Yes |
Ask any AI you already run a single question: a supplier just slipped nine days on a lot that three finished goods depend on — which customer do I disappoint, and what does it cost? Traditional AI will not have the order book. Generative AI will describe the considerations. Only an agentic system can cost the alternatives and hand you one to approve.
Two things arrived in the same short window.
Agentic AI became genuinely capable
Not a chat window — systems that reason across data, rules, constraints and tradeoffs and return an action a planner can defend in front of a customer.
Supply chains stopped behaving
Tariffs, conflict, weather, freight, demand that no longer resembles last year’s. Nobody is waiting this one out.
The model is the commodity. What you build around it is not. Stephen Hutson, CTO — formerly IBM Watson AI, APAC
Every supply chain organization will eventually have access to the same models at roughly the same cost. The advantage was never going to live there. It lives in the data a specific decision requires, drawn from wherever it already sits — and in what accumulates around that decision afterward.
Bring us a pain point your current AI can’t solve.
Thirty minutes on one pain point in your operation. We map where the decision stalls today, which of your systems hold each piece of the answer, and what an agentic recommendation against it would put in front of your planner — the trigger, the options costed, and the expected impact.
Everything, in one list.
Including the background reading that does not sit in the main navigation.
Platform
Background reading
- Why agentic AI · traditional vs generative vs agentic
White papers and articles.
White papers
- AI Decision Intelligence for Modern Supply Chain Operations
- Why SCOR Still Matters in the Agentic Era
- Agentic AI Needs a Governance Layer, Not Better Labels
- Predictive Supply Chain Intelligence is the New Competitive Advantage
- From Buzzword to Business-Critical Capability
- Hybrid Agentic AI: The Future of AI-Activation
Articles
- Decision intelligence: an operating capability, not a project
- Seven S&OP challenges that outlive your ERP
- The Ontology of Execution
- Stop Watching Your Supply Chain. Start Running It.
- AI Orchestration: The Missing Link in Smart Factory Operations
- When Operations Break Down, Stop Looking for Someone to Blame
- Nobody Wants Governance Until They Need It
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Built from the pain points in, not the workflow diagram down.
The agents are organized around the decisions that cost money in a specific operation, rather than around an industry-standard process map. These four are where our team came from and where the deployments are.
Food, protein and perishables
Beef and protein processing · Dairy · Bakery · Produce and fresh · Prepared foods · Beverage
See the three agents →
Industrial manufacturingIndustrial and discrete manufacturing
Industrial equipment · Automotive and vehicle components · Electrical and electronics · Building products · Metals and fabrication · Plastics and packaging
See the three agents →
DistributionDistribution and wholesale
Industrial distribution · Building and construction supply · Electrical and MRO · Food service distribution · Healthcare and lab supply · Aftermarket parts
See the three agents →
AgribusinessAgribusiness and agri-processing
Grain and oilseed · Sugar and ethanol · Coffee and cocoa · Animal nutrition and feed · Fresh produce and packing · Fertilizer and crop inputs
See the three agents →
The qualifying question isn’t your industry.
It is whether decisions in your operation span systems that were never built to reason across one another. If the following describes you, the industry label matters less than the shape of the problem.
Physical goods
Manufacturing or distributing something real, where a scheduling or allocation call has a number attached by end of week.
$400M to $2B
Big enough for the complexity, small enough that a two-year data programme is not survivable.
Accumulated systems
Multi-site, often multi-ERP, frequently inherited through acquisition. Nobody designed this stack; it arrived.
Bring your operation, not your industry code.
One session on the decision that costs you the most. If it isn't a fit, we will say so in that session rather than three meetings later.
Shelf life doesn’t wait for the S&OP cycle.
Yield you cannot assume, an input that is not uniform, a clock running on every lot, and a customer base that reorders weekly. The planning horizon that matters is days, and most planning systems were built for months.
Beef and protein processing · Dairy · Bakery · Produce and fresh · Prepared foods · Beverage
Where the decision stalls.
Yield variance breaks the plan the day it is made
Carcass yield, batch strength and moisture all move. A schedule built on standard yields is wrong by the first shift, and the correction happens in a spreadsheet.
Shelf life turns excess into write-off, not carrying cost
In most industries slow-moving inventory costs you working capital. Here it becomes markdown or waste on a fixed date, so the allocation decision has a deadline attached.
Price and demand move together, weekly
Regional demand, competitor pricing, feed and input costs and weather all move inside the cycle. By the time a static price list reaches the sales team it is stale.
Multi-plant, multi-channel, huge SKU counts
Dozens of facilities, many sales channels, tens of thousands of SKUs. No planner can hold the tradeoff between a channel commitment and a plant constraint in their head.
One story: allocating a short lot across channels
A yield miss leaves you short on a high-demand cut on a Tuesday. Three channels have standing commitments, one of them contractual. The decision has to be made before the lot ages, and it has to be defensible to the account you disappoint.
External Demand Sensing
Reads regional demand, competitor movement, weather and input cost signals so the short position is sized against what demand is actually doing this week, not last month.
Multi-Echelon Inventory Optimization
Costs the reallocation across plants and DCs net of shelf life, so the option that protects service does not create a write-off somewhere else.
Customer Promise Intelligence
Ranks the channels against your own commitment rules and returns which customer gets protected, what it costs, and what the alternative would have cost.
One reconciled recommendation, ranked, with the trigger, the alternatives considered, the constraint that applied and the expected impact attached — ready to approve, edit or reject. Not three separate alerts from three separate systems.
A major Brazilian beef producer — 65 plants, 17 sales channels, 150,000+ customers — cut its regional demand and pricing forecast cycle from 28 hours to under one hour, and now simulates pricing across 60,000+ SKUs in minutes.
The coordinated agent system across demand, production, S&OP and pricing carries a $200M+ [figure cleared 8/30 — final wording aligns to the slide at port] annual profit impact. The deployment is documented publicly by Databricks.
Start with your version of this decision.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
The crop decides the calendar. You decide everything else.
Biological supply you do not control, a commodity price that moves while you plan against it, harvest windows that will not wait, and a customer base that reorders on a weekly cycle. The planning horizon that matters is days, and the systems were built for months.
Grain and oilseed · Sugar and ethanol · Coffee and cocoa · Animal nutrition and feed · Fresh produce and packing · Fertilizer and crop inputs
Where the decision stalls.
Supply is biological, and the plan is not
Yield, moisture, grade and maturity all move against forecast. A plan built on standard assumptions is wrong by the first delivery, and the correction happens in a spreadsheet.
Price and margin move inside the cycle
Commodity indices, currency, freight and input costs shift weekly. By the time a position reaches the commercial team the number it was built on has already moved.
Origin-to-market spans systems nobody joined
Field or farm-gate intake, storage, processing, and the customer commitment each live in a different system. The tradeoff between them lives in someone’s head.
Harvest windows do not negotiate
Capacity, drying, storage and logistics all compete inside a window that closes on its own schedule. Every hour of indecision has a cost attached to it.
One story: a grade miss against a contracted delivery
Intake comes in below contracted specification on a lot already committed. Blending could cover it, but the blend consumes inventory promised to a second contract that prices higher. The decision has to be made before the material moves, and it has to be defensible to whichever customer absorbs the shortfall.
External Demand Sensing
Reads regional demand, commodity movement, weather and input-cost signals so the short position is sized against what the market is doing this week rather than last month.
Multi-Echelon Inventory Optimization
Costs the blend and reallocation options across storage and processing sites, net of quality and shelf-life constraints, so protecting one contract does not create a write-off elsewhere.
Customer Promise Intelligence
Ranks the contracts against your own commitment rules and returns which one gets protected, what it costs, and what the alternative would have cost.
One reconciled recommendation, ranked, with the trigger, the alternatives considered, the constraint that applied and the expected impact attached — ready to approve, edit or reject. Not three separate alerts from three separate systems.
Our agribusiness depth is not theoretical. It is where a large share of the team works every day.
A2go’s Brazil operation runs delivery and product engineering across food, protein, agriculture, pricing and demand planning, in complex multi-plant SAP environments. That is the environment the agent library was hardened in.
Start with your version of this decision.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
The schedule is right until the first supplier moves.
Multi-level BOMs, shared capacity across sites, long-lead purchased components and a master schedule that takes most of a day to rebuild. So it gets rebuilt once, and then defended for a week it should not have survived.
Industrial equipment · Automotive and vehicle components · Electrical and electronics · Building products · Metals and fabrication · Plastics and packaging
Where the decision stalls.
Master scheduling is a spreadsheet exercise
Twenty-odd sheets maintained by different people, reconciled by hand. Any material change after the run means starting over, so in practice the schedule freezes early.
One component, three BOM levels, four finished goods
A single purchased part slipping affects products nobody traced back to it until the promise date was already missed.
Capacity the plan assumes versus capacity the line has
The plan runs on routings. The floor runs on the machine that went down Thursday. MES knows; the planning system finds out in the variance report.
Sites share supply and customers but not a schedule
Each plant optimizes locally. The combination overbuilds in one place and starves another, and nobody owns the tradeoff between them.
One story: a nine-day supplier slip on a shared component
A supplier confirms a nine-day slip on a component that feeds three finished goods across two plants. Two of those goods have firm promise dates this month. The question is not whether you are late — it is who you are late to, and by how much, and whether the plan can absorb it instead.
Master Production Scheduling
Resequences across both plants against real capacity and shared supply, and returns the sequences that hold rather than the one that looks best on paper.
Lead-Time and Safety-Stock Optimization
Re-evaluates what has to be on hand, and where, against the schedule that will actually run — not the one that was frozen last week.
Promise-Date Jeopardy
Flags which commitments are now at risk, costs the alternatives, and returns the commitment that cannot be held as a constraint on the plan rather than an escalation.
One reconciled recommendation, ranked, with the trigger, the alternatives considered, the constraint that applied and the expected impact attached — ready to approve, edit or reject. Not three separate alerts from three separate systems.
A ~$500M industrial manufacturer and distributor took its master production scheduling cycle from 18 hours across 24 spreadsheets to a single 15-minute run, with [FIGURES HELD — the inventory and cash-conversion pair is in neither 8/30 slide; strips at swap unless given a home].
Start with your version of this decision.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
You don’t make it. You still have to promise it.
Thin margins, high SKU counts, supplier lead times you do not control, and customers who judge you entirely on whether the order arrived complete and on time. Every point of OTIF is worth real money and every point of excess inventory costs it back.
Industrial distribution · Building and construction supply · Electrical and MRO · Food service distribution · Healthcare and lab supply · Aftermarket parts
Where the decision stalls.
Stockouts and excess at the same time
The classic distribution signature. Working capital tied up in the wrong SKUs while the fast movers go short, because replenishment runs on rules set two years ago.
Supplier reliability is known but not modelled
Every buyer knows which vendors slip. That knowledge lives in their head and in a personal spreadsheet, and it does not reach the replenishment calculation.
Branch-level demand is invisible at the centre
Regional demand shifts show up in the aggregate weeks after the branch felt them, so transfers happen late and in the wrong direction.
Order promising is a phone call
Available-to-promise is a static number that ignores what is already committed and what is inbound, so the promise gets made and then negotiated afterward.
One story: a fast mover going short across three branches
Demand for a high-volume SKU shifts regionally. Two branches are heading for a stockout inside ten days; a third is sitting on nine weeks of cover. The replenishment buy has a lead time longer than the gap, so the answer is a transfer — and the question is which branch gives it up.
Internal Demand Sensing
Detects the regional shift at branch level as it happens, rather than after it clears the aggregate forecast.
Multi-Echelon Inventory Optimization
Costs the transfer options across the network against service targets and freight, and returns the reposition that protects the most revenue per dollar moved.
Capable-to-Promise
Re-answers what can actually be committed, net of the transfer and what is already promised, so the sales team quotes a date the network can hold.
One reconciled recommendation, ranked, with the trigger, the alternatives considered, the constraint that applied and the expected impact attached — ready to approve, edit or reject. Not three separate alerts from three separate systems.
Distribution deployments start with the decision that carries a number — usually replenishment or order promising — and the first domain is live in 8 to 12 weeks without touching the ERP.
Start with your version of this decision.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
What has actually changed, and how it was measured.
Two deployments, both live. Every figure below is a before-and-after on a named process, with the measurement method stated. Where a customer is not named, it is because the name is not yet cleared for external use — not because the deployment is hypothetical.
Regional demand and pricing forecast cycle
Beef producerAnnual profit impact from the coordinated agent system
Beef producerSKUs price-simulated across markets in minutes
Beef producerMaster production scheduling cycle
Industrial manufacturerInventory carrying cost reduction
Industrial manufacturerA forecast cycle that took a day and a half now takes under an hour.
The company
One of Brazil’s largest beef producers. Operations spanning multiple countries, 65 production facilities, 17 sales channels, more than 150,000 customers, and tens of thousands of SKUs against rising volatility in demand, pricing and logistics.
Before
Pricing analysts spent hours manually extracting data into Excel, running calculations offline, and distributing static recommendations that were frequently outdated by the time they reached the sales teams. Reacting to a change in the data could take days.
What A2go deployed
Agentic applications for price optimization, demand forecasting and inventory allocation, governed by the customer’s own business rules and running on data integrated from warehouse management, ERP and CRM systems plus external feeds — weather, market movements, feed costs. Built on the Databricks Data Intelligence Platform, governed through Unity Catalog.
After
- Regional demand and pricing forecasts: 28 hours to under one hour
- Pricing scenarios across 60,000+ SKUs simulated in minutes rather than days
- 25+ users [published wording applies at port: "25 users … daily and concurrently"] across market intelligence now run complex simulations daily and concurrently, in a governed environment, without data engineering support
- A centralized, batch-driven process became a self-service capability across distributed teams
- $200M+ [figure cleared 8/30 — final wording aligns to the slide at port] annual profit impact from the coordinated agent system running across demand, production, S&OP and pricing
Read the A2go–Databricks customer story →
Measurement: elapsed wall-clock time for the named forecast cycle, before and after, on the customer’s own instrumentation. Published as a customer story by Databricks, December 2025.
| Production facilities | 65 |
| Sales channels | 17 |
| Customers | 150,000+ |
| SKUs modelled | 60,000+ |
| Daily concurrent users | 25+ [published wording applies at port: "25 users … daily and concurrently"] |
| Source systems | WMS, ERP, CRM, external feeds |
This deployment is documented publicly by Databricks. Read their write-up.
Master scheduling: 18 hours across 24 spreadsheets, to one 15-minute run.
The company
A roughly $500M industrial manufacturer and distributor. Multi-plant, shared supply and shared customers across sites, with a master production schedule rebuilt by hand every cycle.
Before
Twenty-four spreadsheets, maintained by different people, reconciled into a single master schedule over roughly 18 hours of planner time per cycle. Any material change after the run meant starting again, so in practice the schedule was frozen well before it should have been.
After
- Master production scheduling cycle: 18 hours to 15 minutes
- Inventory carrying cost: [held — no slide source]
- Cash conversion cycle: [held — no slide source]
- Rescheduling became something the team does when conditions change, rather than once a cycle
Measurement: planner hours logged against the scheduling cycle before and after; inventory and cash figures from the customer’s own finance reporting over the deployment period.
This customer’s name is pending clearance for external use. Reference calls can be arranged under NDA — ask during a working session.
Three agents carried this decision, in the order they run.
- Master Production Scheduling — sequences across sites that share supply, capacity and customers
- Lead-Time and Safety-Stock Optimization — re-evaluates what has to be on hand against the schedule that will actually run
- Promise-Date Jeopardy — returns a commitment that cannot be held as a constraint on the plan, rather than as an escalation to a person
Neither one started with a migration.
Existing systems stayed
No ERP replacement, no data warehouse consolidation, no master-data program completed first. ADIP read from what was already there.
One decision first
Each began with a single decision the business could name a number against, not a platform rollout. Expansion followed the result.
A person kept the call
Recommendations arrived as decision packages — the trigger, the alternatives priced, the constraint that applied and the expected impact attached. Approvals and overrides both fed back into the decision memory.
Ask for the number that matters to you.
Bring your own process and we will tell you, in one session, whether it looks like these two and what a measured first deployment would target.
Decision intelligence, in depth.
White papers and articles on agentic AI, governance, and turning supply chain data into decisions that move the business. Written by the team that builds the platform.
AI Decision Intelligence for Modern Supply Chain Operations
Most midmarket manufacturers and distributors are sitting on more data than ever, yet the distance between what they can see and what they can act on is where margin leaks out.
Decision Intelligence
Worked decisions, not write-ups.
Each one pins a single decision and shows the whole anatomy of it — the signal, the systems consulted, the recommendation, and what every alternative would have cost. Two are worked in full and clickable end to end; five more are in preparation.
Clear to Build
The work order releases Friday. One BOM line out of 214 says no, and a $438,000 shipment is at risk. Equivalent stock exists under another customer’s part number — if the ontology can prove it. Change the unlock path and watch the cost move.
Open the clickable example →
Worked in full · SCOR P2 Plan / OrderATP short-ship
A priority customer is about to be short-shipped. Four allocation strategies ship the same 8,600 units with very different penalty exposure. Here is who ships, who waits, and what it costs.
Open the clickable example →
Supplier OTIF drift
A key supplier’s on-time-in-full is drifting. When does a trend become a decision?
In preparation
Forecast deviation
Four weeks of MAPE creep on one family. Re-plan now, or ride the noise?
In preparation
Capacity spillover
Line 2 is over commitment for the week. What spills, where, and at what cost?
In preparation
Inbound reroute
A port closes overnight. Which inbound lanes move, and what does each day of dwell cost?
In preparation
Safety-stock rebalance
Service targets moved. Where does the buffer belong now — and what frees up?
In preparation
The long-form arguments.
Where the thinking is worked through in full — architecture, governance and the economics of deciding faster.
Why SCOR Still Matters in the Agentic Era
AI agents can read anything, so why would a process framework born in 1996 matter more now? Because agents raise the price of ambiguity.
SCOR · Agentic AI · Decision Intelligence
Read it →
White paperAgentic AI Needs a Governance Layer, Not Better Labels
The debate over what qualifies as an agent misses the point. The real question is which actions may be automated, who owns the outcome, and when human authority intervenes.
Agentic AI · Governance
Read it →
White paperPredictive Supply Chain Intelligence is the New Competitive Advantage
Financial engineering, cost elimination and incremental operational improvement are no longer enough. Predictive supply chain intelligence is where durable advantage now comes from.
Supply Chain · Decision Intelligence
Read it →
White paperFrom Buzzword to Business-Critical Capability
Most organizations sit on mountains of data, yet only a fraction of it informs the decisions that drive performance. How decision intelligence becomes a repeatable capability.
Decision Intelligence
Read it →
White paperHybrid Agentic AI: The Future of AI-Activation
Maximize impact, minimize change. If you are dealing with excess inventory, slow order-to-cash cycles and demand-planning guesswork, hybrid agentic AI layers over the systems you already run.
Agentic AI · Decision Intelligence
Read it →
Shorter pieces, same argument.
Decision intelligence for supply chains: an operating capability, not a project
A continuous lifecycle of designing, executing and improving decisions, run as a performance program with KPIs and owners, rather than a one-off modeling exercise.
Decision Intelligence · Supply Chain · Agentic AI
Read it →
ArticleSeven S&OP challenges that outlive your ERP
ERP systems record what happened; S&OP is about deciding what to do next. Seven persistent planning challenges that live in that gap.
S&OP · Agentic AI · Supply Chain
Read it →
ArticleThe Ontology of Execution
The next advantage comes not from more dashboards or models but from systems that turn business intent into coordinated action, built on an explicit ontology of how decisions connect.
Agentic AI · Governance · Supply Chain
Read it →
ArticleStop Watching Your Supply Chain. Start Running It.
Most companies are paying for AI that tells them what is wrong. Very few have AI that actually fixes it.
Agentic AI · Supply Chain · Forecasting
Read it →
ArticleAI Orchestration: The Missing Link in Smart Factory Operations
The sales-operations divide is a decision architecture problem. Edge AI, orchestrated agents and decision memory together close it.
Agentic AI · S&OP · Decision Intelligence
Read it →
ArticleWhen Operations Break Down, Stop Looking for Someone to Blame
Most operational failures trace back to broken systems, not individual error. Why blame culture and decision debt keep operations from improving.
Decision Intelligence · Supply Chain · Governance
Read it →
ArticleNobody Wants Governance Until They Need It
Decision governance is the accountability framework behind trustworthy, explainable AI decisions, and the organizational pain that only surfaces after something has gone wrong.
Governance · Decision Intelligence · Agentic AI
Read it →
Written up by someone other than us.
A2go on Databricks
Databricks' own write-up of the apps A2go runs on its platform, and the deployment behind the 28-hour to under-one-hour forecast cycle.
Read it on databricks.com →
AssessmentAI readiness survey
Ten minutes and twenty questions across the four pillars. A written read on where your decisions stall, whether or not you ever talk to us.
Take the survey →
Read enough. Bring us the decision.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Two things happened at the same time.
A2go was built to bring them together, in a way that creates ongoing agility and resilience in a supply chain now and into the future.
Agentic AI became genuinely capable
Not a chat window, but systems that can reason across data, rules, constraints and tradeoffs, and come back with an action a planner can defend.
Supply chains stopped behaving
Tariffs, conflict, weather, freight, demand that no longer resembles last year’s. Nobody is waiting this one out.
Founded
Decision pillars of purpose-built supply chain agents
Operating centres — United States and Brazil
Weeks to first domain live
Three vantage points on the same problem.
A2go’s executive team came at supply chain from three directions and kept arriving at the same conclusion: the technology was finally ready, and the way it was being applied was going to fail most companies.
Mike Romeri
Chief Executive Officer & Founder
Mike spent his career inside supply chain operations, and he kept finding the same gap in every one of them. The insight side kept improving — better forecasts, better analytics, better dashboards, reports arriving sooner every year. The decision side did not move. A disruption would surface on a Monday and still take days to resolve, and the resolution came out of a conference call, a planner’s instinct and a spreadsheet rebuilt by hand. Afterward, nobody could reconstruct why that call was made, what got traded away, or whether it was the right one.
The shortage was never insight. It was everything that happens after it.
Speed up the reporting into a decision process that is still manual, still partly guesswork and still not auditable, and all you have bought is an earlier view of a problem you cannot act on any faster.
- 40 years in supply chain · Harvard MBA
- 20+ years Partner at PRTM (now PwC)
- Contributed to SCOR as a practitioner at PRTM
- Co-founded OPS Rules with MIT’s David Simchi-Levi — acquired by Accenture, 2016
Cesar Oliveira
Chief Operating Officer
Cesar has spent years deploying ERP and enterprise platforms across manufacturing, distribution, food and industrial operations. Systems of record execute and track reliably; they were never built to reason across one another. And no company arrives at its stack by design — it accumulates, one platform at a time, plus whole environments inherited through acquisition. What companies need is not another system. It is AI that connects not just the siloed data but the decisions that span it: the tradeoffs, the guardrails, the priorities that determine which order gets protected.
Coordination is the product, and every pain point has to be tied to a business outcome before anyone builds anything.
He designed the A2go AI Assessment around exactly that — start from the decision that is costing money, name the outcome it should move, then coordinate across forecasting, planning, inventory and allocation to move it.
- 20+ years ERP transformation
- MBA, University of Colorado Boulder
- Led 25+ enterprise ERP go-lives: SAP, Oracle, Epicor
- Architect of ADIP, built natively on Databricks
- Leads the 32-person Brazil team
Stephen Hutson
Chief Technology Officer & Co-Founder
Stephen saw the commoditization coming before most of the market did. The models themselves were going to become a utility — every supply chain organization would eventually have access to the same ones, at roughly the same cost, with roughly the same capability. Whatever advantage existed was never going to live there. It was going to live in what surrounds the model: the data a specific decision requires, drawn from wherever it already sits, and what accumulates around that decision afterward.
The model is the commodity. What you build around it is not.
Routine decisions get automated outright. The ones that need a person arrive as explainable recommendations with the projected outcome attached, in front of a knowledge worker while there is still time to act. Every decision is logged and auditable — and becomes part of the record of how your company actually decides, which is what makes the next recommendation better than the last. That part cannot be commoditized, because no vendor has it.
- 9 years Principal, PwC-PRTM
- Contributed to SCOR while at PRTM
- Nominated Distinguished Engineer at IBM
- Ran Watson AI and Commerce Analytics Services for APAC
- Designed A2go’s data orchestration layer
- Architected the composable agent framework
Guilherme Yoshimura
Head of Data Science
- 12+ years in IT, a decade of it in data science and advanced analytics
- MSc Data Science, University of São Paulo; academic residency, University of Alabama
- Data-driven transformation work with Nestlé, iFood, Thoughtworks, Honda and Itaú Unibanco
- Leads agent design and deployment across the four pillars
Guilherme owns A2go’s data science strategy — how the agents are built, evaluated and put into production. His background spans manufacturing, financial services and high-volume digital platforms, which is the range the agent library has to hold up across. He builds and leads the teams that turn machine learning into decisions a planner will actually approve.
Matt Calamusa
US Director of AI Sales
- 16+ years in enterprise technology sales and GTM leadership
- Graduate thesis-in-residence, The Walt Disney Company Global Corporate Alliances
- Senior commercial roles at Syren Cloud, Nutanix and Arista Networks
- Enterprise deals across healthcare, financial services, aviation and manufacturing
Matt leads A2go’s US revenue strategy and the executive conversations that start it — where a company’s highest-cost decision sits, and what it would take to move it. He has spent his career translating technical capability into deployments with a number attached, across cloud, data and security. He is a player-coach: he builds the team and carries a bag.
Brazil is where a large part of the platform gets built.
Not an offshore delivery function. A2go’s Brazil operation is a 32-person engineering and delivery centre that designs agents, hardens them against real operations and takes them into production alongside customers.
Where the hardest environments are
Food, protein and agriculture at national scale — multi-plant, multi-channel, high-SKU, with biological supply variability and commodity price exposure that most planning systems were never built for.
Deep SAP and multi-ERP practice
Complex SAP estates, plus the mix of platforms that accumulates through acquisition. Reconciling across them without a standardization program first is routine work, not a special case.
Pricing and demand planning at speed
Regional demand, pricing and allocation decisions that move weekly rather than monthly. This is where the forecast and OTIF pillars were pressure-tested.
The agent library was not built in a lab against clean data. It was built against operations that break the assumptions in most planning software, which is why it holds up in a multi-ERP estate assembled through three acquisitions.
Agribusiness and agri-processing Food, protein and perishables
Inside ADIP
The three layers these three arguments produced, and how they fit together.
Learn more →
Point of viewThe Coordination Gap
The thesis in full: why the cost sits between your systems, not inside them.
Learn more →
ReferenceWorked decisions
A decision closed end to end, with the alternatives priced.
Learn more →
The platform has three layers because the team has three arguments.
Each layer answers one of them. Take any one away and the argument above it stops holding.
A data foundation
That reads from where your systems already are.
Coordinated supply chain agents
That reason across them.
A judgment layer
That holds the part no vendor can supply.
Why we built ADIP the way we did.
Three routes are on offer today. Two ask you to change how you work. The other starts with how you already work.
Buy a prebuilt platform and conform to it
Its data model, its workflows, your teams re-engineered around it — all before any of it reaches the numbers.
Fund a long consulting engagement
Measured in years, against an outcome nobody can name at the start.
Take what you already run and make it work better
Your pain points set the sequence, not a vendor’s roadmap — and the first one is live in weeks.
ADIP orchestrates only the data a decision actually needs — from the ERP, the planning tool, the warehouse system, the supplier feed, wherever it sits today — and puts interoperable agents on top of it that coordinate across forecasting, planning, inventory and order promising. The recommendation reaches the person who owns the call with the tradeoffs and the projected outcome attached, while the order can still be re-promised, the shipment re-routed, the buy adjusted.
Built from the pain points in.
Our team came out of supply chain, with depth in industrial manufacturing, agriculture, and food and protein processing — multi-plant, multi-ERP, high-SKU, thin-margin environments where a scheduling decision has a number attached to it by end of week.
That is why the agents are organized around pain points rather than industry-standard workflow diagrams. We built for the four pillars of supply chain and the pain points within them.
Forecasting & Planning
Operational Planning
Supply & Inventory Optimization
OTIF Optimization
Adopting AI shouldn’t put the business on hold.
Every company with a physical supply chain is going to build AI capability. That part is settled. What is still open is whether building it has to mean a consulting engagement measured in years, a data program measured in millions, and a return that arrives after the people who approved it have moved on.
You have the same volatility, but less room to absorb it
For a $20B enterprise, that path is survivable. For a $500M manufacturer it is not — and that company has the same volatility, the same fragmented systems from the last three acquisitions, and considerably less room to absorb a bad quarter.
Not the compromise version
The incremental path is not the compromise version for companies with smaller budgets. It is the better path at any size, and it happens to be the only one available at some of them.
Who we serve
The profile, the complexity and the qualifying signals.
Learn more →
Deep diveWhat deployment looks like
Rapid assessment, first domain live, then expansion at a pace you set.
Learn more →
AssessmentAI readiness survey
Ten minutes, and it tells you where your decisions stall before we ever talk.
Learn more →
Your stack records. The agents reason. Your people decide.
A2go is supply chain decision intelligence — one governed layer over the systems you already run.
What we touch, what we don’t, and who can prove it.
You are being asked to let an AI system read from your systems of record and, once you approve, write back to them. That deserves a straight answer rather than a badge. Here is what ADIP actually accesses, what it cannot do, and what you keep control of.
Read-only by default. Write-back only where you scope it.
What ADIP reads
- Only the fields the agents in your scope actually reason on — not a full database copy
- Through standard governed interfaces: APIs, database views, existing lake or warehouse tables
- No schema changes on your side, no agent installed inside the ERP
- The field list is documented at assessment and reviewed before go-live
What ADIP writes
- Nothing at all until you enable it — first domain runs recommend-only
- Field-level scope agreed in writing, not blanket system access
- Every write carries the decision ID, the approver identity and the timestamp
- Prior state retained and restorable
- Recommend-only mode remains available permanently if that is your policy
Disconnect ADIP and your business keeps running exactly as it did. Your ERP is still the system of record, your data never moved, and no process depends on us to transact. You lose the coordination and the decision memory — you do not lose the operation.
Auditable end to end.
Lineage and permissions
Data lineage, access control and audit run through Unity Catalog on Databricks. Every field an agent reads is traceable to its source system and its permission grant.
Decision logging
Every recommendation and every human response is logged: the trigger, the alternatives considered, the constraint that applied, the expected impact, the approver and the timestamp. Exportable for audit.
Operating limits
Each deployment defines which decisions may be automated, which require approval, and what an agent may never do regardless of confidence. These are enforced in configuration, not stated in a policy.
Model isolation
Your Judgment Layer — your approvals, overrides and the reasoning behind them — is never shared across customers and is never used to train anything another customer touches.
Data residency
ADIP runs in your Databricks environment or a dedicated one, in the cloud and region you specify. Data does not transit to a shared A2go tenancy.
Open formats
Storage in open formats on the lakehouse, versioned. Portable by design — you are not locked into a proprietary store you cannot read without us.
What your security team will ask for.
| Document | Contents |
|---|---|
| Architecture and data flow | Every connection, direction of flow, field-level scope, and where processing occurs. |
| Access control model | Authentication, role definitions, permission inheritance and separation of duties between agent and approver. |
| Audit and retention | What is logged, for how long, in what format, and how it is exported. |
| Write-back scope template | The document you sign before any action reaches a system of record. |
| Subprocessors and hosting | Cloud provider, region, Databricks configuration, and any third party in the path. |
| Incident and continuity | Notification commitments, escalation path, and what happens to your operation if ADIP is unavailable. |
Compliance certifications and their current status are provided on request — ask during a working session and we will send the current pack rather than a claim on a webpage.
Bring your security team to the first call.
It is a better use of the session than a second one later. We will walk the data flow, the write-back scope and the audit model against your actual environment.
Find out where your decisions stall.
Twenty questions across the four pillars. You get a written read on which decisions in your operation are costing the most time and margin, which systems are holding the pieces, and which one is worth starting with — whether or not you ever talk to us.
What it asks
How your forecast, schedule, inventory and promise decisions get made today: who touches them, how long they take, and where the spreadsheets are.
What you get
A ranked read on where the decision latency and margin leakage sit, with the pillar each one belongs to.
What it costs
Ten minutes. No call is scheduled unless you ask for one.
Start here
The survey opens once you tell us where to send the result.
Bring the decision that costs you the most.
This is not a platform evaluation and not a demo of features you did not ask about. It is one working session on a single decision: where it stalls across your systems, what it is worth, and what a first deployment against it would look like.
What the session looks like
- You bring the decision, and the people who actually make it
- We map where it stalls across your systems, live
- You leave with a scoped first domain and a timeline
- You don’t leave with a migration proposal or a platform commitment
We will say so in that session rather than three meetings later. Not every operation needs this, and the ones that don’t are usually obvious within twenty minutes.
Not ready for a call?
The readiness survey takes about ten minutes and tells you where your decisions stall before we ever speak. Take the survey instead.
Request a session
We reply within one business day.
AI Decision Intelligence for Modern Supply Chain Operations
Most midmarket manufacturers and distributors are sitting on more data than ever, yet the distance between what they can see and what they can act on is where margin leaks out. A look at how decision intelligence closes that gap.
Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Why SCOR Still Matters in the Agentic Era
AI agents can read anything, so why would a process framework born in 1996 matter more now, not less? Because agents raise the price of ambiguity. A founder’s argument for SCOR DS as the shared language between humans, systems, and the agents that now sit between them.
SCOR · Agentic AI · Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Agentic AI Needs a Governance Layer, Not Better Labels
The debate over what qualifies as an agent misses the point. The real challenge is deciding which actions may be automated, who owns the outcome, what evidence is required, and when human authority must intervene.
Agentic AI · Governance
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Predictive Supply Chain Intelligence is the New Competitive Advantage
Strategic value creation has entered a new operating reality. Financial engineering, cost elimination and incremental operational improvement are no longer enough. Predictive supply chain intelligence is where durable advantage now comes from.
Supply Chain · Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
From Buzzword to Business-Critical Capability
Make better decisions, consistently, at scale. Most organizations sit on mountains of data, yet only a small fraction of it informs the decisions that drive performance. How decision intelligence becomes a repeatable capability.
Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Hybrid Agentic AI: The Future of AI-Activation
Maximize impact, minimize change. If you are dealing with excess inventory, slow order-to-cash cycles and demand-planning guesswork, hybrid agentic AI offers a path that layers over the systems you already run.
Agentic AI · Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Decision intelligence for supply chains: an operating capability, not a project
Decision intelligence is how supply chains turn data into consistently good decisions: a continuous lifecycle of designing, executing and improving decisions, run as a performance program with KPIs and owners rather than a one-off modeling exercise.
Decision Intelligence · Supply Chain · Agentic AI
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Seven S&OP challenges that outlive your ERP
ERP systems record what happened; S&OP is about deciding what to do next. Seven persistent planning challenges that live in that gap, and how AI-assisted decision-making addresses each one.
S&OP · Agentic AI · Supply Chain
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
The Ontology of Execution
The next advantage in enterprise operations comes not from more dashboards or models but from systems that turn business intent into coordinated action, built on an explicit ontology of how people, decisions, workflows and outcomes connect.
Agentic AI · Governance · Supply Chain
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Stop Watching Your Supply Chain. Start Running It.
Most companies are paying for AI that tells them what is wrong. Very few have AI that actually fixes it. The shift from AI as a dashboard to AI as an operator is what separates supply chain leaders from the rest.
Agentic AI · Supply Chain · Forecasting
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
AI Orchestration: The Missing Link in Smart Factory Operations
A synthesis of three IndustryWeek analyses arguing that the sales-operations divide is a decision architecture problem, and that edge AI, orchestrated agents and decision memory together close it.
Agentic AI · S&OP · Decision Intelligence
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
When Operations Break Down, Stop Looking for Someone to Blame
Most operational failures trace back to broken systems, not individual error. Why blame culture, decision debt and the midmarket intelligence gap keep supply chain operations from improving.
Decision Intelligence · Supply Chain · Governance
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.
Nobody Wants Governance Until They Need It
Decision governance is the accountability framework behind trustworthy, explainable AI decisions, and the organizational pain that only surfaces after something has gone wrong.
Governance · Decision Intelligence · Agentic AI
The body of this piece goes here.
This landing page carries the title, summary and topics so the Library links stay inside this site instead of handing the reader to a different design. The article or paper text drops into this section — it is not in this preview build.
More from the Library.
All white papers and articles
Everything A2go has published on decision intelligence, agentic AI and governance.
PlatformHow ADIP works
The layer that carries a decision from signal to action to memory, above the systems you already run.
ProofResults in production
Two live deployments, with the before-and-after figures and how each was measured.
Bring us a pain point your current AI can’t solve.
Thirty minutes on the decision that costs your operation the most, mapped across the systems you actually run.