Skip to main content

Architecture

A decision layer over the systems you run.

OODARIS reads from your ERP, POS and planning data. By default it writes back only what a named person approved. Your systems of record stay the source of truth.

Decision record

Entry 4417Approved

CL-03 Suburban: markdown ladder B

  1. ReadSell-through, stock, traffic, price and plan for every cluster14 sources
  2. Checked1,412 overnight checks raised 3 exceptions, CL-03 among themOvernight
  3. ProposedLadder B: a shallower first cut, held two weeks87% confidence
  4. Policy checkInside the margin limit set by merch financeCleared
  5. ApprovedMerchandise planning08:47
  6. Written back
    • Price conditions, SAP S/4HANA, 214 SKUsOK
    • Transfer orders, CL-04 to CL-03, 3 packsOK
    • Open-to-buy and merchandise plan re-projectedOK
  7. RollbackReversible for 14 daysOpen
Trace 4417 links each row to its source data.Illustrative product view

Decision lifecycle

Every decision moves through the same five stages.

An architecture review is about the contracts between them: what each stage receives, what it is allowed to do, where a person signs, and what goes on the record.

  1. 01

    Assemble context

    Bring the relevant product, location, calendar, inventory, plan, and policy context into scope.

    Canonical data contracts
  2. 02

    Scope and route work

    Assign bounded tasks to the appropriate agent, model, optimizer, or enterprise service.

    Coordinator and shared context
  3. 03

    Generate a recommendation

    Combine evidence and constraints into a proposed action with rationale and stated confidence.

    Models, optimization, and rules
  4. 04

    Review and approve

    By default, a person reviews each recommendation before orders or changes are written back.

    Policy gate and human approval
  5. 05

    Write back and learn

    Send approved actions through controlled interfaces, then compare outcomes with the expected result.

    APIs, events, and audit trail

One control plane across the flow

Controls travel with the decision instead of being reconstructed at each handoff.

  • Policy limits your teams set
  • Confidence thresholds
  • Human overrides and approvals
  • Decision records and lineage

Integrations

Where OODARIS fits in your stack

OODARIS reads from the systems you run today and writes changes back to the system that owns them. Nothing is migrated, and your systems of record stay authoritative.

Reads from

Merchandising and ERP
SAP ECC or S/4HANA, Microsoft D365, Oracle Retail, NetSuite, APTOS
Stores and commerce
POS, Shopify
Data platforms
Data warehouses such as Redshift, object storage such as S3
Everything else
Custom sources through APIs, event streams or file exports
OODARIS
  1. 01One retail data modelProducts, locations, calendars and measures from every source, with lineage back to the source record.
  2. 02Five decisionsMerchandise financial planning, open-to-buy, assortment, allocation, and pricing and markdown.
  3. 03Named approvalBy default, nothing is written back until the decision's owner approves. If the owner chooses Autonomous, OODARIS acts inside guardrails they set. Every action is logged with its reason and can be reversed.

Writes back

  • Price conditions and markdowns, for example in SAP S/4HANA
  • Transfer orders and pack changes
  • Plan updates: the re-projected open-to-buy (OTB) and merchandise financial plan (MFP)

Each writeback uses the target system's own APIs or event paths and is logged with the decision that caused it.

To start a pilot, we need

  • Sales transaction history
  • POS or ERP access, through an API or a file export
  • Inventory data, ideally at SKU level

Security and data

How data, access and approvals are handled

These are the practices OODARIS is built and run with. The architecture review goes through each one against your own requirements.

Data handling
OODARIS reads the data each decision needs and changes your systems only through writeback: approved by a person, or run inside limits the decision's owner set. Your systems of record stay the source of truth.
Tenancy and environments
Each customer's data stays inside its own tenant boundary, and environments are kept separate.
Access control
Identity and role-based access decide who can view, propose, approve and execute each decision. Data is encrypted.
Approval trail
Every recommendation, approval and override is recorded with who made it, when and why. Policy checks and confidence thresholds run before anything reaches an approver.
Rollback
Each writeback has a rollback path, and a rollback goes on the same record as the decision.
Traceability
One trace ID links the source inputs, the agent and model runs, the approval, the writeback and the outcome.

Technical evaluation

What technology leaders ask first

The questions technology, AI and data leaders bring to an Agentic OS evaluation, and what to review together.

Enterprise fit

How does this fit the enterprise we already run?

It sits above the systems you keep, and it can replace the planning tools you no longer need: at ALDO Group, OODARIS replaced a legacy planning system and more than 20 spreadsheets. It reads from your ERP, POS and planning data and, by default, writes back only approved actions, through those systems' own interfaces. A pilot reaches its first live cycle in 8–12 weeks.

What to review together

  1. Integration with existing enterprise systems
  2. Security, access, and operating ownership
  3. A phased rollout with clear decision gates

Composite roles drawn from live retail engagements

Bring us your architecture questions.

In the review we map OODARIS onto your environment: system boundaries, how data moves, access and approvals, and who operates what. Bring your enterprise architects and your security team.

What happens next

  1. A reply within 24 hours

    A person on the OODARIS team reads your note and writes back to set up a first call.

  2. The first call

    Your architects walk through how OODARIS reads your data, records approvals, writes changes back and rolls them back.

  3. A pilot, if it fits

    8–12 weeks to a first live cycle, run on your data with your approvals.