We build the systems that tell your business what to do next, and why.
The ratio went up. No order follows from it, no change to staffing, no price adjustment. Nothing follows from it.
Sales, stock, staffing, accounting: each system knows a part. None knows why the week went badly.
The good decisions depend on whoever happens to be in charge. At the second site that knowledge is no longer in the room.
“While reading, I kept asking myself which decision I would actually make differently because of these reports.”
A report ends at the observation. Here come the recommendation, the simulation, the release and the execution, and after that the measurement of whether it worked.
Signals from sales, stock, staffing, capacity and external factors arrive continuously and are checked against the model.
The system proposes a named measure: which item, which supplier, which assignment, with the expected contribution in euro.
The measure is computed against the model in advance, including knock-on effects on stock, capacity and staffing.
You decide per measure type: automatic, released by a role, or manual only. A release workflow, not a switch.
Released measures go into the connected systems as a procurement proposal, a staffing change or a price adjustment.
The effect is held against the expectation. What does not deliver becomes visible instead of disappearing into the next recommendation.
No step runs without an audit record, and none without a way back. What may run automatically is set by you, per measure type and per role.
A model of your business. On it: forecasting, simulation, and recommendations ranked by what they are worth in euro.
See itAutonomous systems for sales and operations: containerised, in your cloud, with documentation and handover.
See itCreative, content and video. One-off or continuous, alongside the systems.
The interface is the smallest part. Underneath it sits what makes a recommendation trustworthy at all.
Sits above ERP, inventory, time management and whatever systems of record the sector runs on. Integrated by API or export, across sites. Released measures are written back, with legally sealed systems excepted.
Objects, attributes and relationships as a knowledge graph. The substrate that makes simulation possible at all.
Demand per object and day, as an interval rather than a point estimate, backtested against your own history.
Scenarios are computed against the twin. You see the expected effect before a measure runs.
Run id, period, inputs, computation path. No value without provenance. That is the precondition for audit.
OpenTelemetry traces across the pipeline, structured logs, alerting on drift and source-feed failure.
The AI layer calls verified functions and phrases their output. It invents no number and alters none.
A role and permission model per tenant, enforced in the database. Full change log, EU endpoints with no retention, subprocessors declared per service.
Which is why any number can be traced back to its source: run, period, origin.
The fastest way to lose trust is a system producing attractive numbers nobody can verify.
Before we build, we check whether your data can answer the question. If it cannot, we say so.
Every model runs against your own past before you have to believe anything it says.
The new runs next to your way, not instead of it. You switch when it is measurably better.
Documentation, tests, handover. You can carry on without us at any time.