Coordination used to be cheap enough to do by hand.
For thirty years the operating assumption of supply chain management held: signals arrived slowly enough that humans could coordinate the response. A weekly planning meeting was a reasonable cadence because the world moved at roughly that speed. A cold chain problem discovered on Tuesday could be settled by Thursday and nothing was lost. Judgement lived in experienced people, coordination lived in meetings and email, and the ERP recorded the result afterwards.
That system was not primitive. It was correctly matched to its conditions. Enterprise software was built for it, which is why every major system in a distribution business today is a system of record rather than a system of action. SAP records what was decided. A warehouse system records what moved. A BI dashboard records what happened. None of them decide, because deciding was never their job.
- Planning cadence
- Weekly
- Signal arrival
- Daily
- Coordination cost
- Low
- Where judgement lived
- People
The assumption that has quietly expired
Every one of those systems still assumes a human will notice the signal, gather the right people, agree a response, and type it back in. That assumption is no longer true, and almost no software has been rebuilt around its failure.
The human became the execution layer, and therefore the bottleneck.
Weather, regulation, geopolitics, supplier stability, freight rates and customer expectation now change hourly rather than quarterly. The volume of signals rose by an order of magnitude. The number of experienced people available to coordinate a response did not.
So the human stopped being the judge and became the transport layer. A person notices a breach, a person opens a spreadsheet, a person books a meeting, a person carries the decision to whoever can approve it, a person types the result into SAP. Judgement is perhaps four minutes of that chain. The rest is coordination, and coordination is now the expensive part.
- Decision latency today
- 48 to 72 h
- Of which is judgement
- Minutes
- Roles unfilled by 2033
- 1.9M
- Plan AI for decisions
- 94%
Sources: Deloitte 2025 on unfilled supply chain roles. ABI Research 2025 on operators planning AI for decision support. Decision latency measured across Groflex pilot environments using timestamps already present in the customer ERP.
The forecast was usually right. The decision was late.
At a distributor of roughly 150 million in revenue, the measured gap between a signal appearing and a documented decision executing runs to four to six million a year in spoilage, expedited freight, stockouts and write-offs. Not one of those losses is a forecasting failure.
Two things have happened at once, and this is the part most software has missed. Forecasting became a solved commodity, and coordination became the binding constraint. The industry responded by buying better forecasts.
AI should not replace the decision maker. It should remove the coordination.
There are two available responses to a coordination bottleneck. Hand the decision to the machine, or take the coordination away from the human and leave the decision where it belongs. The first is what most of the market is currently selling, and in a regulated industry it is unbuyable: no Qualified Person, Head of Quality or CFO will accept an autonomous agent moving stock or money, and no regulator will accept an unattributed decision.
The second is Groflex. The signal is detected the moment it crosses a threshold. The options are priced. The policy is checked by article number. The decision is routed to the named human with the authority to make it. The approved action is written back into the system of record. Every step is logged as one artefact.
Humans keep authority. Machines keep continuity.
Everything else follows from that sentence, including the rule that the platform has no approve button of its own.
Why it has to be a new layer
| What exists | What it does | Why it cannot close the loop |
|---|---|---|
| ERP and WMS | Record what happened | No reasoning, no routing, no priced options |
| BI and dashboards | Show what happened | Ends at the chart. Cannot act, cannot write back |
| Forecasting and planning | Predict what will happen | Solved, and the forecast was rarely the failure |
| RPA | Repeat fixed steps | No judgement, no policy, breaks on exceptions |
| Autonomous agents | Decide and act | Unbuyable in a regulated industry. No accountable owner |
| Consulting | Fix it once | Leaves no system behind |
Supply chain is the first market, not the company. Every regulated industry has the same structural problem: signals arriving faster than authority can be exercised, and an auditor arriving later to ask who decided. The layer generalises. The wedge does not, and the wedge is what gets built first.
See the gap on your own data before you take our word for it.
- Backtest01Your data · 1 to 3 days · risk-freeYour last 12 to 24 months replayed. Deleted on delivery, confirmed in writing.
- Pilot02$10,000 · 90 daysOne facility. A written outcome report against your own baseline.
20 minutes. No deck. No sales script.