One decision layer across your supply chain
FLX4 connects to the systems you already run, turns their data into predictions and costed options, routes each decision to the person who owns it, carries the approved action back into those systems, and then checks what happened.
How the platform fits together
Seven parts, one loop. Each FLX4 product uses the same pattern for its own decisions, so they behave alike and can share signals.
Connect
Read-first connections to ERP, spreadsheets, machine data, trip data and partner portals.
For technology teams
File and CSV ingestion works from day one. ERP connectors are configured per deployment and confirmed in discovery.
Model
One operational data model per domain: items, locations, orders, suppliers, machines, trips, claims.
For technology teams
Data checks flag duplicates, unit mix-ups and stale feeds. Where data is too old or too thin, recommendations say so or abstain.
Predict
Forecasts with ranges, projected shortfalls, predicted arrival times and fair-price bands.
For technology teams
Statistical forecasting (for example AutoETS, AutoARIMA and Croston for sparse demand) with model selection per item; regression on price indices; trip-time prediction.
Decide
Costed options and a recommendation, with the reasons shown.
For technology teams
Optimization solvers (such as OR-Tools CP-SAT) handle allocation, MOQ, truck fill, changeovers and award splits. Language models explain and draft; they do not invent the numbers.
Govern
Approval rules, guardrails, evidence and an audit trail for every decision.
For technology teams
Value-based approval tiers, maker-checker steps, override reasons, and per-decision autonomy settings with a kill switch, where the product supports them.
Execute
Approved actions carried to the systems and teams that act on them.
For technology teams
Purchase orders and amendments, carrier indents and tenders, supplier and distributor portal tasks, and notifications, each linked to the decision that caused it.
Learn
Outcomes measured against the plan and fed into the next decision.
For technology teams
Decision logs and ledgers record what was recommended, what was chosen, who approved it and what happened, so parameters and models improve.
Your operation, running twice
ControlTower keeps a live software mirror of your operation: every order, machine, vehicle and pallet. Test the decision in the twin, then carry it into the real world, with a named person approving.
- 1:1physical ↔ digital mirror
- Livestreaming state, not reports
- What-ifsimulate before you commit
- Governedautonomy at your pace
- 01
Mirror
Orders, stock, machines, trips and claims kept in step with the real operation.
- 02
Simulate
Run the options in the twin: what it costs, what it saves, which promises it keeps.
- 03
Approve
A named person picks the option, with the evidence in front of them.
- 04
Carry out
The approved change goes to the systems and teams that act on it, with a rollback.
Your ERP is the system of record. FLX4 is a system of decision.
The two do different jobs, and FLX4 is built to work beside the ERP you have, not to replace it.
| Your ERP | FLX4 |
|---|---|
| Records transactions accurately | Decides what the next transaction should be |
| Holds master data and the ledger | Reads them, checks them and flags what looks wrong |
| Runs fixed rules and MRP settings | Weighs options against live constraints and cost |
| Reports what happened | Predicts what will happen and what it will cost |
| Posts the approved document | Prepares it, with evidence, for a person to approve |
Human oversight, built into the flow
Control is a feature of how decisions move, not a policy document.
Named approvers
Every decision that writes to a system has an owner and an approval rule. Value tiers decide who signs what.
Evidence attached
The data, the options and the trade-offs travel with the recommendation into the approval.
Overrides with reasons
People can change any recommendation. The reason is kept and used to improve the next one.
Abstain when unsure
Thin or stale data lowers confidence or stops the recommendation, and a person is told why.
Guardrails and limits
Spend ceilings, contract terms and policy limits that a recommendation cannot cross.
Kill switch
Automation for any decision type can be paused at once, returning it to manual approval.
Every decision leaves a receipt
The receipt is the unit of trust in FLX4: what was seen, what was weighed, what was chosen, who approved it, what was written where, and what happened next. An auditor, a CFO or a new planner can follow any decision months later.
Governed autonomy: earned one decision type at a time
FLX4 is engineered for autonomy, and it starts with people in charge. Pick a decision to see where it typically starts, what would let it move up a level, and what moves it back down.
Example policy Levels and thresholds shown are example policies. You set your own, per decision type.
Deployment and data
Where it runs
Agreed per engagement: our managed cloud or your cloud account. On-premises deployment is available for products designed for it; ask about the product you need.
Your data stays yours
Each customer’s data is kept separate, and is never used to serve another customer.
How AI models are used
Forecasts and optimization come from statistical models and solvers. Language models are used to explain, summarise and draft, through a controlled gateway.
Start small, scale on results
One decision, read-only, ninety days. Scale only when the agreed measure moves.
See the platform on one of your decisions
Read-only on day one. A baseline agreed in writing. Ninety days to see if the number moves.
Read-only on day one. Your people approve every action. If the number does not move, you do not scale.