Home · Technology
ADIP — the A2go Decision Intelligence Platform
One governed decision layer over the systems you already run.
ADIP is made up of a data orchestration engine that feeds AI-ready data to supply chain specific AI agents, which combine with the Judgment Layer to shape every recommendation. Human oversight runs across all of it, and each approval or rejection is captured and used to make the next recommendation sharper.
Signals flow up from the systems you already run. Governed actions flow back down into them. Every approved decision is recorded and shapes the next one.
Layer by layer
Read it bottom to top.
Five layers, one system. Your stack sits at the bottom and does not move. Everything A2go adds sits above it, and approved decisions come back down into your systems of record.
Layer 01
Your existing systems — unchanged.
ERP, WMS, OMS, planning tools, commerce, MES and OT signals from the floor, supplier feeds, and the spreadsheets that hold the parts nobody automated. Often more than one of each, accumulated across acquisitions.
- Nothing is replaced. ADIP layers over your systems of record. They stay the systems of record.
- Nothing is migrated. Your data stays where it lives, in open formats, read through governed interfaces.
- Add or retire freely. Bring on a new system or decommission an old one without interrupting the agents above.
Layer 01 of ADIP
Layer 02 · ADOE
Your data, made AI-ready where it already lives.
ADOE — the A2go Data Orchestration Engine — is the foundation every agent runs on. It reads from your source systems and continuously reconciles, enriches, and streams exactly the data each agent needs, in the shape it needs, at the moment it needs it. Built natively on Databricks.
Real-time, micro-batch, or scheduled — ADOE adapts to each source rather than the reverse.
Entities reconciled, hierarchies aligned, codes standardized, time series put on one timeline.
Open formats on the Lakehouse, versioned so historical states stay traceable and auditable.
A tailored data product per consumer, not one flattened table everything has to share.
Agents, models, and dashboards draw what they need through governed interfaces.
Anomalies, schema drift, and upstream breaks caught before decisions run on bad inputs.
If you already have a lake, a warehouse, or a fabric, ADOE reads from it rather than replacing it. Static stores become live, agent-ready feeds instead of extracts somebody refreshes on a schedule. And ADOE does not wait for a finished master-data program — it reconciles across the definitions in your systems today and supports that work over time rather than blocking the business until it lands.
Layer 02 of ADIP
Layer 03
Supply chain AI agents that talk to each other.
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.
- Organized in four pillars. Forecasting & Planning, Operational Planning, Supply & Inventory Optimization, OTIF Optimization.
- They reason on governed data. Each draws only what it needs from ADOE, so the answer rests on current, reconciled inputs.
- They get sharper with use. Every approval and override feeds the Judgment Layer above them.
Layer 03 of ADIP
Layer 04
The Judgment Layer — the part that never made it into software.
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. The Judgment Layer encodes it, and every approval and override makes it sharper.
Traces
Every decision your planners make is captured — including the ones they reject, override, escalate, and handle as exceptions. The why behind every yes and no.
Decision memory
Priorities under load, supplier and allocation rules, lead-time and SLA guardrails, escalation policy — accumulated into a memory specific to your company.
Governed decisions
The next recommendation reflects how you actually decide. Approved overrides become reusable playbooks, available across the next disruption.
Layer 04 of ADIP
Layer 05
Human oversight and governance, across every layer.
Oversight is not a step at the end. It runs across the whole platform: what the agents may see, what they may act on, who approves, and what gets written back. Every approval and every rejection is recorded and returns to the Judgment Layer as a training signal.
- Approval-ready, not autonomous. Recommendations reach a person with the reasoning attached. Nothing writes back until someone says so.
- Limits set by your team. Agents operate inside the guardrails you define, and cannot act outside them.
- Recorded either way. A rejection teaches the system as much as an approval does.
Layer 05 of ADIP
What you do not have to do first
No heavy migration. Not at the start, not later.
Your data and your workflows stay where they are. A2go works above the systems you already run, which means the first decision domain can go live in weeks — and the tenth one runs on exactly the same footing as the first.
Your data
Stays where it lives.
ADOE reads from your source systems in open formats and reconciles against the definitions you have today. No lake to consolidate into first, no master data program to finish before value starts.
Your workflows
Stay where they are.
There is no new tool for your team to operate. Planners work in the systems they already know, and approved decisions are written back into the ERP, the schedule, and the order book.
Your architecture
Stays the same as you expand.
Adding a decision domain means adding agents on the same foundation, reading the same systems, feeding the same Judgment Layer. The architecture in month thirty is the architecture in week six.
How it runs
Nothing runs unsupervised.
Judgment stays with your planners. In audited and regulated environments that is a requirement, not a compromise — and it is how trust in the recommendations gets built in the first place.
Approval-ready
Every recommendation arrives as a decision package: what triggered it, the alternatives considered, the tradeoffs, and the expected impact. The planner approves, edits, or rejects.
Auditable end to end
Action, approver, reasoning, timestamp — with lineage and permissions running through governed controls into the systems of record.
Automatic on your timing
Promote a call to automatic once your team has watched it be right, decision class by decision class. That schedule belongs to you.
Guardrails
Your business rules apply at every step. Agents see only what they are permitted to see and cannot act outside the limits your team sets.
The approval
The planner sees the ranked options, the constraint that applied, and what each choice costs. Override authority stays with the person, not the system.
Write-back and record
The approved action returns to the ERP, WMS, OMS, and planning systems, and the decision is logged with its full audit trail.
The question is not whether AI can decide without you. It is whether you can show, a year later, why it decided what it did.
Where the compounding lands
Your decision memory is yours.
Every approval, every override, every rejection and the reasoning behind it accumulates into a Judgment Layer that reflects how your company decides. It accumulates in your tenant.
Not in a harmonized copy of your enterprise held somewhere else. Not folded into product content shared with anyone. The Judgment Layer sits on your own foundation, in open formats, inside your environment — and it is never shared across customers.
Where it lives
Your environment, your foundation, open formats.
Who it is shared with
No one. Your decision memory is specific to your company and stays that way.
If you ever stop working with us, what your planners taught the system stays with your planners.
Start here
Bring one decision. We will map it against the stack.
Pick a decision your team loses time or margin on, and we will walk it layer by layer — what ADOE would read, which agents would reason on it, and what would write back.