See the whole platform

Business rules: what the agents know about your operation

Before an agent touches a document, the operation it works in is written down: its objects, the rules that say what correct looks like, and the exceptions that go to a person. Agents work from that model, never from a generic template, and every correction your team makes adds to it.

What the model holds

The forward deployed engineer writes it with your team, in your operation's own terms, before the workflow goes live.

  • Your objects

    Projects, shipments, orders, assets, sites and suppliers, and how they relate: which order a delivery note belongs to, which supplier a shipment comes from.

  • Your rules

    What a correct order, quote or quantity looks like for you: matching rules, unit conversions, routing by volume, shipping calendars. A rule your team keeps in a spreadsheet becomes one the agent applies to every case.

  • Your exceptions

    What does not fit the standard pattern, and where it goes. A case the rules cannot settle goes to a person with its sources, instead of being guessed at.

  • Your reference data

    The catalogue, the customer file and the rates the rules are checked against, read from the systems that hold them, not copied into a new one.

Rules in production at Biodéal

Biodéal distributes organic dairy products, and its customers order by email in every format: PDFs from their own ERP, Word or ODT tables, scanned forms, plain text. These are some of the rules its order agent applies before an order is created in SAP Business One.

The volume routing rule was one Biodéal specified in a spreadsheet. It is now applied to every order with no human step, and locked in by automated tests.

See the Biodéal case study
Biodéal order rulesIn production since 22 July 2025
Not an order
Delivery confirmations, stock-out notices and Biodéal's own orders to its suppliers are rejected.
Customer
Matched exactly, then fuzzy, then by address.
Units
Quantities converted between kilos, cases and units. A quantity with no unit is inferred by comparing the document's unit price with SAP's rate for each possible unit, and the inferred line is flagged for the team to confirm.
Volume routing
A full pallet of milk leaving one site triggers the matching supplier order; below that threshold, the order ships as cases from another warehouse.
Shipping calendar
Fixed departure days, EDI cut-offs, automatic roll-forward to the next business day.
Reference data
The product catalogue and the customer file, kept in sync with SAP automatically.
Unreadable line
Never fails the whole order and never disappears: it goes out as an alert with a diagnosis and a suggested fix.

Inputs, outputs and who decides

The model is part of every workflow, so it is written, checked and changed with the people who own the work.

Inputs
Your documents as they arrive, the reference data already in your systems, and the rules as your team states them: in a spreadsheet, in a procedure, or as the forward deployed engineer sees the work done on site in week one.
Outputs
The checks the agent runs on every item, and for each item one of two outcomes: it goes through, or it is held for a person with the rule it failed and its sources attached.
What a person decides
Your team sets the rules and the thresholds with the forward deployed engineer when the workflow is scoped. A held item is decided by the person named for that workflow.See approvals and audit
Set per workflow
Each workflow has its own rules, and each entity can have its own. OFRET and SWILL, the two entities of OFRET Groupe, run on one TMS with different rules for what counts as a standard request, so each has its own environment.See the OFRET Groupe case study

Corrections become rules

When your team settles an exception, the decision becomes a rule the system applies from then on, so the same exception is not settled twice. The rules and integrations built for the first workflow carry into the next.

At OFRET Groupe, an improvement made for one entity, such as a new verification rule or a better extraction pattern, can be adapted for the other without starting from zero.

At Biodéal, the team completes a held line in SAP or corrects the customer's rule, then reprocesses the original email through a dedicated tool that runs the same guardrails as production.

What the model does not do

Said plainly, so nobody buys it for something it is not.

  • It is not a rules library you configure alone. The rules are written for your operation, with Mirage's engineers, during the deployment.
  • It brings no reference data of its own. Mirage supplies no material catalogue or labour-unit database: the agents check against your catalogue, your rates and your estimating system.
  • A rule does not make an ambiguous case certain. What the rules cannot settle goes to a person.
  • On a stable, high-volume workflow that is fully rule-based and has no exceptions, simpler RPA or template OCR is likely cheaper and faster to set up. See AI agents vs RPA
  • Mirage publishes no accuracy percentage for its checks.

Where it runs in production

Each line restates the case study it links to.

Biodéal
Its own business rules applied to every order before it is created in SAP Business One: 2,943 orders and 12,993 order lines entered automatically since launch, about 7% of orders flagged for a person.See the Biodéal case study
OFRET Groupe
Two environments, one per entity, each with its own rules for a standard request, and verification rules that check a quote or an extraction against what normal looks like for that entity. More than 90% of standard quote requests are handled automatically.See the OFRET Groupe case study
OCBM
Agents adapted to OCBM's own drawings and conventions. Every quantity carries a reference to the drawing it was measured on.See the OCBM case study
Transwin
Built one workflow at a time: each process was put into production before the next, which is how freight, fuel, mining and fleet ended up on one platform.See the Transwin case study

Frequently asked questions

  • No. The forward deployed engineer writes them with your team when the workflow is scoped, from your documents, your procedures and the way the work is actually done. A rule your team already keeps in a spreadsheet can be taken as it is: at Biodéal, a routing rule specified in a spreadsheet is applied to every order.

  • It goes to a person, with the rule it failed and its sources attached, instead of being guessed at. At OFRET Groupe, a request goes to an operator before anything is sent when data is missing, the extraction confidence is low, the price falls outside the entity's standard rate patterns, or the request matches neither entity's usual profile.

  • Yes. Corrections and decisions your team makes become rules the system applies from then on, so the same exception is not settled twice.

  • Yes. OFRET Groupe's two entities share one TMS but quote differently, so each runs its own environment with its own rules, on the same platform.

  • No. The agents check against your own reference data, read from the systems that hold it. At Biodéal, the product catalogue and the customer file are kept in sync with SAP Business One automatically.

  • At Biodéal, the rules are locked in by automated tests, and an email is reprocessed through a dedicated tool that runs the same guardrails as production.

Bring us the operation you need to improve.

Mirage will map the workflow, identify what can be automated and define the fastest path to production.

Start with one workflow

Leave your work email, then pick a time for a 30-minute call on the workflow where your team loses the most time. You leave with a written plan to automate it. No commitment.

We use your email to reply to you and, now and then, to tell you about Mirage. You can opt out at any time. Privacy policy