Skip to content

Cash, in real time

Currency

Most finance teams can tell you their cash position. Fewer can tell you how old it is. Balances sit in four bank accounts and three currencies; receipts come from the billing platform and the pipeline, commitments from payroll and payables. When half of that arrives as a monthly export, the view is days or weeks old before anyone forecasts from it.

A stale view is paid for in idle cash: the less you can see, the bigger the buffer you hold against it. Wiring the stack together — nightly syncs first, then APIs and webhooks, with AI doing the matching in between — narrows the forecast and gives that buffer back. Connect it system by system and watch what it releases.

Three stacks

The stack · seven systems

Tap a system to step it up: export, sync, live
  • Monthly export
  • Daily sync, in batches
  • Live feed
  • Ring: share of the view arriving live

Cash view

19 days

old, value-weighted

Forecast band

±€1.3m

90% range at week 13

By hand

24 h

a week · ≈ €75k a year

Cash released

€0

against the spreadsheet buffer

Verdict

Your cash view is 19 days old, so you hold a €1.9m buffer against what you cannot see.

The biggest single step from here is moving the banks to a daily sync: it releases €164k and saves 3.9 hours a week.

13-week cash forecast

Same flows in every stack. Only the certainty changes.
  • Expected balance
  • 90% range
  • Minimum buffer
  • Cash released
Week by week, as a table
WeekEventsInOutExpected90% range
NowOpening position——€4.6m€3.9m–€5.3m
W1Customer receipts€1.1m€137k€5.6m€4.8m–€6.4m
W2Supplier run€380k€442k€5.5m€4.7m–€6.4m
W3—€357k€142k€5.7m€4.8m–€6.6m
W4Payroll, Supplier run€365k€1.5m€4.6m€3.6m–€5.5m
W5Customer receipts€1.1m€151k€5.5m€4.5m–€6.5m
W6Supplier run€347k€447k€5.4m€4.3m–€6.4m
W7VAT€339k€604k€5.1m€4.0m–€6.2m
W8Payroll, Supplier run€335k€1.5m€3.9m€2.8m–€5.0m
W9Customer receipts€1.1m€144k€4.8m€3.6m–€6.0m
W10Supplier run€386k€474k€4.7m€3.5m–€5.9m
W11—€365k€154k€4.9m€3.7m–€6.2m
W12Payroll, Supplier run€378k€1.6m€3.7m€2.4m–€5.0m
W13Customer receipts€1.1m€136k€4.6m€3.3m–€5.9m

System by system

Share of variance: each system’s part of the week-13 forecast uncertainty

  • Banks

    Four accounts · EUR, GBP, USD

    19 days · 6.0 h/wk

    26%

  • Billing & payments

    e.g. Stripe

    19 days · 4.0 h/wk

    20%

  • CRM pipeline

    Deals and expected close dates

    19 days · 2.0 h/wk

    24%

  • Ledger (ERP)

    e.g. Xero or QuickBooks

    19 days · 3.0 h/wk

    5%

  • Bookkeeping partner

    Outsourced, e.g. Pilot

    19 days · 3.5 h/wk

    3%

  • Payroll

    Three countries, monthly runs

    19 days · 1.5 h/wk

    2%

  • Spend & payables

    Cards, bills, payment runs

    19 days · 4.0 h/wk

    12%

  • Timing no feed removes

    Customers pay when they pay. Live data cannot narrow this part.

    6%

Notes on the model

What it shows

  • One monthly export is enough to age the whole cash view: leave only the CRM on a spreadsheet and the view is most of a day old, however live everything else is.
  • A narrower forecast is worth cash, not just comfort: the buffer you hold scales with what you cannot see, so visibility frees cash to invest or to pay down debt.
  • The gains flatten. Once every feed is live, most of the remaining doubt is customers paying when they pay — a collections problem, not a systems one.

How it works

  • Each system carries an error on the opening balance and on every forecast week. Its connection mode scales both: a daily sync keeps about half the weekly error, a live feed about a fifth.
  • Weekly errors are independent, so the variance of the balance grows linearly with the horizon. The band is ±1.645σ, a 90% range.
  • Minimum buffer is an operating floor plus 1.645σ at week 13: cover a 1-in-20 miss and still pay everyone. Cash released is the fall in that buffer against running everything on spreadsheets.
  • Cash-view age is the average age of each feed, weighted by how much of the cash view it carries.

What it leaves out

  • The flows, error sizes and hours are illustrative, calibrated for a mid-sized group with a quarter of steady trading. Your own forecast-versus-actual history is the real input.
  • Errors are treated as independent. In a bad quarter they are not: late customers and a slipping pipeline tend to arrive together.
  • Integration cost, data quality and the controls needed to trust an automated feed are left out. So is FX: balances are shown in one reporting currency.

Next experiment · 08

Automate or hire?

The AI business case, costed honestly: implementation, ramp, rework and oversight included.

Test the business case

Using this on a real decision?

Talk it through

The defaults are illustrative. The inputs that decide the answer are the ones you only know from inside the business.

Get in touch