Three teams optimise the cycle. None of them can see it
Module · Working Capital and Transaction Finance
Operations runs the cycle: how customers, suppliers and inventory actually behave. Three teams optimise three legs of it and none of them sees the whole.
Prophesee Operations predicts when cash really moves, ranks the accounts worth working today, and tests a change before it is offered. Treasury turns the result into funding.
Every leg of the cycle is optimised. Nobody is optimising the cycle.
Collections chases the largest invoice, payables holds the run to the last legal day, and planning protects service with cover. Each is doing its job well. Every collector is working hard and almost none on the right account, because no one of them can see the cash the other two are moving.
Sources: UK Government, Large Businesses' Payment Practices and Performance Statistics 2025, published 14 Jul 2026 (n=11,178 statutory reports) · Ardent Partners, AP Metrics that Matter in 2025, published Feb 2025 (n=212 accounts payable and finance professionals, 53 percent from organisations above $1bn revenue) · The Hackett Group, US Working Capital Survey, Aug 2025 (top 1,000 US listed non-financial companies, drawn from financial statements).
From chasing the ledger to predicting the cash
ForesightThe worklist is ordered by value and days overdue, because that is what the report can compute. Nothing on it says which account is about to go late.
Time to event models predict when each receipt lands, when each payable falls due and when stock actually turns, each with a confidence interval. The worklist reorders around expected cash at risk rather than around age, so the same team releases more by working a different order.
PulseDays sales outstanding is reviewed monthly, by which point a deteriorating account has had four weeks to deteriorate quietly.
Write the deterioration in plain English: this customer slipping two weeks against its own pattern, this category of dispute rising, this entity stretching payables past policy. Rules run continuously against actuals and the forward projection, severity scored and routed with the action attached.
HorizonA terms change or a settlement discount is argued on one spreadsheet, with no model of what would have happened anyway and no way to prove afterwards that it paid.
Predict the no-action cycle, then model the intervention: a terms change, a discount, a dunning cadence, a payment run policy. Watch the projected curve move, then track actuals against that plan with counterfactual modelling separating the effect from the trend.
NexusThe same customer appears in three ledgers under three spellings and two legal names, so exposure cannot be totalled and the same supplier can be paid twice.
Fuzzy entity resolution over names, addresses, tax identifiers and bank details, with vector embeddings catching the cases that string matching misses. One identity per counterparty across ledgers, remittances and contracts, so exposure is a number rather than an estimate.
Turning working capital challenges into decisions
Ordered by value and days overdue. Nothing in it says which account is about to go late, so effort follows size rather than risk.
ForesightOne customer, three ledgers, three spellings. Exposure cannot be totalled and the same supplier can be paid twice.
NexusOne wire arrives covering forty seven invoices with no remittance advice, and somebody spends a day deciding what it paid.
Nexus PulseThe run goes out on schedule. Nothing in it flags the invoice paid twice, the price above contract or the credit never taken.
PulseLimits are set at onboarding and reviewed annually. Between those two dates the exposure changes and nothing tests it.
Pulse16 AI applications that could be relevant
A sample of what becomes possible on the decision layer, not a fixed list: each application draws on the same data foundation and audit trail, and new ones are configured on the engines, not built from scratch.
The date each open item is actually paid, predicted per customer with a confidence interval.
The worklist reordered by expected cash at risk rather than by age.
Which invoices will be disputed, predicted at issue rather than discovered at day forty.
Where stock cover is drifting from plan, by line and location, before it becomes cash.
Exposure and limits tested continuously against behaviour rather than against the onboarding score.
Payments, credits and deductions that do not fit the pattern, scored before release.
Cover, ageing and obsolescence moving against plan, by line and location, while it is still cash.
Customer deductions classified and ranked by recoverability, with the evidence attached.
Payment terms, discounts and policy exceptions watched across every entity.
Cash sitting unapplied, aged and owned, so it stops being discovered at quarter end.
Model a change to terms, stock cover or payment policy and watch the projected cycle move.
When to pay each supplier to hold days payable without breaching a term or a relationship.
Actuals tracked against the modelled plan, so the improvement is attributed to the action.
Every customer and supplier reconciled to a single identity across ledgers, remittances and contracts.
Receipts matched to open items by probabilistic matching, with the residue ranked by how close it came.
Candidate duplicates surfaced before the run releases, not found in the following quarter.
A day when the collector calls the right account
Today: Three teams, three targets, and a cycle none of them can see whole.
ForesightThe worklist reordered by predicted cash at risk, with the reason each account moved up it.
Today: A customer who quietly slipped two weeks, noticed at the monthly review.
PulseStock cover drifting on two lines and a payment run about to release early. Scored, routed, with the cash effect of each already priced.
Today: Offer the discount because the paper argued for it, then hope.
HorizonAn early settlement discount tested against a dunning change on the projected cycle. She takes the one that costs less margin.
Today: The cycle improved. Nobody can say which action did it.
HorizonLast quarter's terms change tracked against the modelled plan, with the counterfactual separating it from seasonality.
Tomorrow's cash becomes today's call list.
Release trapped cash
We agree the metric and the baseline in week one, and measure the result on your data.