Home · How ADIP works
How ADIP works, in more detail.
Five things the homepage states and this page explains: why nothing has to move, what coordination means when the alternative is centralization, where your business rules actually live, what a planner receives, and why the whole thing improves without another project.
Nothing migrates. Nothing gets replaced.
Read it bottom to top. Your stack stays exactly where it is; everything A2go adds sits above it, and approved decisions come back down into your systems of record.
What removing ADIP would leave behind
Your source systems, untouched and still transacting. Your data, in the open formats it was already in. The decisions your team made while ADIP was running, recorded in your own systems of record. The exit is real because there is nothing to unwind — no destination to migrate out of, and no process that was rebuilt around the platform.
- What deployment looks likeRapid assessment, first domain live, then expansion on your roadmap.Deep diveLearn more
- Systems we connect toERP, planning, WMS, commerce, and the data platforms underneath them.ReferenceLearn more
- Where A2go is not a fitThe situations where we say so early rather than late.Straight talkLearn more
- Security and controlsAccess, audit, and the limits your team sets before anything acts.FAQLearn more
The data travels. The systems don't.
Centralization asks your systems to move into one place. Coordination leaves them where they are and moves only what a decision needs. That distinction sounds architectural, but it is the whole difference between an eighteen-month programme and a first result in weeks.
What actually happens, in order
- PulledADOE reads from your existing systems where they already sit — multiple ERPs, planning, WMS, OMS, commerce, supplier feeds — through governed interfaces. No extract sits waiting for someone to refresh it.
- PreppedEntities are reconciled across systems that name the same thing differently, hierarchies are aligned, and time series are put on one clock. Only the fields a given agent needs, in the shape it needs them.
- UsedThe agents reason over that prepared data together rather than in isolation, so one agent's answer becomes context for the next, and what reaches a planner is one reconciled recommendation.
- ReturnedThe approved action — and the reasoning behind it — writes back into the systems it came from. That becomes part of the operating record those systems hold, and part of the input for the next decision.
Why the loop matters more than the pull
A one-way integration is a pipe. What makes this a loop is the return leg: the decision goes back to the source, so the next time the same question arises, the systems already reflect what was decided. Nothing has to be reconciled twice, and no separate history builds up inside A2go that your systems don't know about.
This is also what keeps your systems the systems of record. ADIP holds the reasoning; your ERP still holds the order.
The practical test
Ask any vendor what happens to a decision after someone approves it. If the answer is that it lives in their platform and your systems find out later, that is centralization with extra steps.
- The Coordination GapWhy the cost sits between your systems rather than inside any one of them.Point of viewLearn more
- Systems we connect toERP, planning, WMS, commerce, and the data platforms underneath them.ReferenceLearn more
- How governance holdsLineage, permissions, approval, and write-back controls end to end.FAQLearn more
- Anatomy of a DecisionOne supply chain decision traced end to end, hand-off by hand-off.Deep diveLearn more
At the centre of ADOE is a model of how your company decides.
Data alone cannot tell an agent what a Tier 1 account is worth protecting, which supplier substitution is permitted, or who is allowed to approve a change that costs money. That lives in the ontology — the governed model at the centre of ADOE that carries your business rules, your guardrails, and your decision responsibilities.
What the ontology holds
- Business rules. Allocation policy, substitution limits, minimum changeover, safety stock floors, service commitments — the standing logic that constrains every recommendation before a person sees it.
- Guardrails. The limits an agent cannot act outside of, regardless of what the data suggests. These are set by you, they are explicit, and they are enforced rather than advisory.
- Decision responsibilities. Who owns which decision, what each role may approve, and where a call must escalate. The ontology knows that a Tier 1 re-promise is not the same authority as a line reschedule.
- Accountabilities. Which approvals are recorded against which role, so an audit answers not only what was decided but who was entitled to decide it.
Why it sits inside ADOE rather than inside each agent
If every agent carried its own copy of your rules, changing a policy would mean changing it in thirty places, and two agents could reason from different versions of the same rule. Holding it once, at the data layer, means every agent above draws on the same governed model — and a policy change takes effect everywhere at once.
It is also what makes a recommendation explainable. When an agent reports that an option was ruled out, it can name the rule that ruled it out and where that rule came from. An auditor can follow the same trail.
Where the rules come from
Some are already structured in your systems and are read directly. Others live in an SOP or in a planner's head, and get captured and validated with you during discovery. ADIP does not read legal contracts and turn them into executable rules — where a term is contract-specific, the authoritative source stays your own contract or the system that already extracts it.
- Governance and auditLineage, permissions, approvals, and write-back controls end to end.FAQLearn more
- ADOE, in fullThe six continuous functions and the architecture underneath them.Deep diveLearn more
- The Decision LibraryEvery decision the agents cover, by pillar, with the metric each one moves.ReferenceLearn more
- Ask ArtiePut a question about your own rules to our agent and watch it reason.Try itLearn more
What a planner actually receives.
Not an alert and not a dashboard. A defined artifact with required fields, assembled once, carrying its own reasoning — which is what makes it both actionable and auditable.
- Recommended actionThe specific call, stated so it can be executed without interpretation — not a direction, an instruction.
- TriggerWhat changed, when it arrived, and from which system. Every package traces back to a signal with a timestamp and a source.
- Alternatives consideredThe other options the agents evaluated and what each one would have cost, so the planner sees the shape of the tradeoff rather than taking the answer on faith.
- Expected impactThe projected outcome in the metric that matters — service, working capital, avoidable cost — stated before the decision, so it can be measured after.
- Rules appliedThe business rules and guardrails from the ontology that constrained the reasoning, named individually rather than summarised.
- Coordinated withWhich other agents and domains contributed, so it is clear the recommendation reconciles across the chain rather than optimising one corner of it.
- Audit recordWho approved or rejected it, when, under which rule set version, with the full stage trace retained.
Seven required fields · every package carries all of them
Query-ready, through Ask Arti
A package is not a static document. Every line in it can be interrogated in plain language through Ask Arti — where a number came from, which rule permitted an action, what would change the answer, why an alternative was ruled out.
Ask Arti is the user interface to ADIP. It is not the AI engine and it does not make the decisions; the coordinated agents do that, on governed data, under the ontology. Ask Arti is how a person asks the platform to explain itself.
Why the artifact is defined rather than free-form
A recommendation with required fields can be audited, compared against the outcome it projected, and replayed months later against the rule set that was in force at the time. A paragraph of generated text cannot.
It also means the planner's review is the same shape every time. The trigger is always in the same place, the alternatives are always listed, and nothing arrives without the reasoning attached — which is what makes a fast decision a defensible one.
- Anatomy of a decision packageA worked example, field by field, as a planner would read it.Deep diveLearn more
- Ask ArtieAsk what a recommendation would look like for a decision you make every week.Try itLearn more
- From review to automaticHow teams decide which calls stop needing approval, and how to move that line back.FAQLearn more
- The Decision LibraryBrowse the decisions agents produce across all four pillars.ReferenceLearn more
What lands on a planner's screen.
All of it — the data, the agents, the judgment — arrives as one artifact a person can act on. Not an alert, and not a dashboard to go interpret. A decision package, with the reasoning attached.
- Judgment stays human. Full tradeoff visibility and override authority at every step. We remove the friction around the decision, not the decision-maker.
- Approved actions write back. Straight into ERP, WMS, planning, and execution systems, with a full audit trail attached.
- Promote what you trust. As projected outcomes prove out cycle after cycle, you choose which calls run without review. You set that line, and you can move it back.
- The required fieldsAll seven, and how Ask Arti makes each line query-ready.On this pageLearn more
- The Decision LibraryBrowse the decisions agents produce across forecasting, planning, inventory, and order promising.ReferenceLearn more
- From review to automaticHow teams decide which calls stop needing approval — and how to move that line back.FAQLearn more
- Ask ArtieAsk what a recommendation would look like for a decision you make every week.Try itLearn more
Pull the Thursday run forward to Wednesday second shift, and hold the Tier 1 allocation intact.
A supplier moved a commit at 09:14 against sell-through running 18% above forecast. Four orders lost cover.
- Triggered by ▸
- Supplier commit moved on a bound component at 09:14, against sell-through 18% above forecast. Either alone was absorbable.stage 01–02
- Permitted by ▸
- Changeover window open Wednesday second shift · 8-hour changeover minimum respected · safety stock floor unchanged.stage 03
- Alternatives ▸
- Expedite after the outage — misses two promise dates, adds freight. · Partial allocation — leaves three accounts short.stage 04
- Promise on ORD-4471held at 12 Sep
- Orders taken off risk4 of 4
- Days recovered2
- Expedite freightavoided
- Tier 1 accounts shortnone
- Commitments brokennone
Nothing is written back until you say why. The reason is what the Judgment Layer learns from.
Illustrative configuration · figures anonymised
Why it gets better without another project.
Most software is as good on day one as it will ever be for you. This runs the other way: the layer is fed by the work your team already does, so each turn of the cycle leaves it sharper than the last.
One turn of the cycle
Nothing in this loop asks your team to do anything they are not already doing. That is the point — the input is the ordinary work of reviewing and deciding.
- 01More usageYour planners work the way they already work — reviewing, approving, overriding.
- 02More tracesEvery one of those calls is captured with its reasoning, including the rejections.
- 03Sharper agentsThe next recommendation reflects the decisions your company actually made.
- 04Better decisionsWhich earns more use — and the cycle starts again, one turn further along.
Specific to you, retained by you. 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. Without it, AI is just automation; with it, expertise compounds instead of retiring. This is the piece no chatbot or generic assistant delivers.
- How we capture tribal knowledgeTraces, decision memory, and the loop that compounds with every approval.Deep diveLearn more
- Your logic stays yoursHow decision memory is isolated per customer and never shared across accounts.FAQLearn more
- Supply chain AI agentsThe agents this loop sharpens, by pillar, with the metric each one moves.ReferenceLearn more
- The Coordination GapOur newsletter on decision-making in complex supply chains.SubscribeLearn more