--- title: "Cash forecasts: how much confidence should payment promises carry?" canonical: https://www.billabex.com/en/blog/cash-forecast-payment-promises-confidence/ lang: en alternate: https://www.billabex.com/fr/blog/prevision-tresorerie-promesses-paiement.md updated: 2026-10-03 index: https://www.billabex.com/llms.txt --- # Cash forecasts: how much confidence should payment promises carry? A payment promise contains information that the contractual due date does not: the customer has stated an intention after receiving your request. It deserves a place in cash forecasting. However, entering its full amount on the promised date treats a declaration as an assured receipt. Ignoring every promise creates the opposite problem, excluding useful operational information from the finance team's view of the coming days. The question is how much confidence to place in that information for a defined horizon. To prepare next week's outgoings, you need amounts expected to reach the bank before the relevant cut-off. To organise collections, you need commitments to verify and blockers to resolve. These uses share underlying facts, but they support different decisions and should not automatically produce identical figures in every report. ## Define what a fulfilled promise means Before calculating a rate, define the observed outcome. Is a promise fulfilled when the customer initiates a transfer, when your bank receives the money or when accounting reconciles it? For liquidity forecasting, the availability of funds is the relevant point. Reconciliation subsequently establishes which invoices the receipt covers and prevents the same payment from being mistaken for another expected cash inflow. Retain the initially promised date, amount, covered documents and observation date. If the customer postpones twice, the third date should not silently replace the first in the history used to assess reliability. You need the changes to organise the next action and the original version to understand what could reasonably have been anticipated when the first commitment was received. The approach to [making promises trackable](https://www.billabex.com/en/blog/verifiable-payment-promises/) provides this foundation. “Payment is being processed” should not automatically receive the same status as a confirmation of a specific amount and date. Label incomplete information honestly instead of assigning artificial precision simply because a forecast template requires every row to contain a date and a percentage. ## Read the historical rate alongside its denominator In a simulated review, you examine twenty promises whose observation dates have passed. Sixteen were received in full within the defined period: **16 ÷ 20 = 80%**. That figure describes those twenty observations under your chosen definition. It does not establish that each new promise has exactly an 80% probability of being fulfilled on the relevant date in the future. Even under the simplifying assumption of independent, comparable observations, the 95% Wilson interval for 16 successes out of 20 is approximately **58.4% to 91.9%**. NIST documents the formula; these figures are calculated for our example, rather than taken from a business survey. [NIST, confidence intervals for a proportion](https://www.itl.nist.gov/div898/handbook/prc/section2/prc241.htm). The width illustrates the weakness of a small history. If fifteen observations concern the same customer or payment run, independence itself becomes questionable. Do not present that interval as certification of your actual portfolio. Start by counting distinct customers, examining the circumstances and stating which comparisons the observations reasonably support. More rows do not necessarily mean more independent evidence about future behaviour. ## A high frequency does not guarantee the expected cash amount The sixteen fulfilled promises in the simulation each concern €1,000, producing €16,000. The other four each concern €10,000 and remain unpaid at the observation date. Total promised cash is €56,000, but receipts within the period amount to €16,000, or **28.6% of the promised value**, rounded. Both the large and small commitments remain in the review; none has been removed to improve the result. The 80% rate is still correct by number of commitments. Applying it mechanically to the whole value would be misleading: €56,000 × 80% gives €44,800, exceeding observed receipts by €28,800. This does not invalidate every weighted forecast. It shows why the relationship between commitment size and payment behaviour deserves examination before a count-based historical rate becomes a general cash conversion assumption. Excessively narrow segmentation creates a different problem. Four categories with three cases each do not necessarily produce more reliable probabilities. Look for explainable differences, such as incomplete customer approval, a partial receipt already received or several missed commitments. Describe the evidence available and retain an “insufficient history” category where the data cannot support a robust estimate of the outcome you need. ## Keep probability, promised date and collectible amount separate Microsoft Dynamics 365 Finance documentation illustrates the distinction: its payment prediction feature assigns probabilities of on-time, late and very late payment to invoices. These are estimates using historical information, separate from a commitment expressed by a customer. [Microsoft, payment predictions, updated 25 June 2026](https://learn.microsoft.com/en-us/dynamics365/finance/finance-insights/payment-insights-overview). In your own forecast, a probability must identify its target event. Does “80%” mean complete payment before Friday, receipt of an initial instalment or eventual payment during the month? Those outcomes produce different available cash balances. A score without a defined horizon and amount can appear rigorous while providing little help with a particular bank payment your business needs to make. Check that the underlying case is current as well. A credit note agreed yesterday may reduce the expected amount, and cash already received must leave the list of future inflows. The controls for [delayed accounting synchronisation](https://www.billabex.com/en/blog/delayed-accounting-sync-pause-reminders/) therefore matter before interpreting a promise that is still visible. An updated forecast must not count the same money twice under different statuses in separate reports. ## Use scenarios management can actually discuss Consider a second simulated example. Two commitments concern €6,000 and €4,000 before Friday. For a sensitivity exercise, management chooses hypothetical weights of 80% and 50%. The weighted amount is **€6,000 × 80% + €4,000 × 50% = €6,800**. These percentages are working assumptions, not measured performance figures for any software product or claims about the two customers' actual reliability. If each invoice is paid in full or remains entirely unpaid within the horizon, actual receipts can be zero, €4,000, €6,000 or €10,000. They are never exactly €6,800 in that simplified model. The weighted figure summarises an expectation; it is not an assured bank receipt that can be committed to an outgoing payment without examining an adverse outcome and its implications. Dependence between payments matters too. If both invoices are included in the same transfer instruction, their outcomes may be linked. In a variant where the full €10,000 arrives together with an assumed probability of 80%, the expected amount is €8,000, but the possible outcomes remain zero or €10,000. Adding weighted invoice values does not create diversification that the actual payment process does not provide. ## Adjust for new evidence, rather than the need for cash Hyndman and Athanasopoulos recommend supporting forecast adjustments with relevant additional information and documenting the changes. Their discussion highlights the limitations of small optimistic adjustments. This methodological principle is useful when assessing information contained in a customer's promise. [Forecasting: Principles and Practice, section 6.7](https://otexts.com/fpp3/judgmental-adjustments.html). A confirmed customer approval is new information to assess. The fact that your payroll falls on Friday does not make the promise more likely to be fulfilled. Keep the receipt scenario, financing requirement and decision about covering a shortfall distinct. The [largest-customer payroll stress test](https://www.billabex.com/en/blog/largest-customer-late-payment-payroll/) helps translate that uncertainty into an operational requirement with a date and a cash amount. The [payment promise workflow in Billabex](https://www.billabex.com/en/product/payment-promises/) brings together commitments that need verification in the customer relationship. To incorporate them into forecasting, preserve their content, history and connection to funds actually received. Confidence then becomes an explainable, revisable assumption accompanied by a cash scenario that your team can decide to accept or cover, instead of a reassuring percentage detached from the next financial decision. ## Sources - [NIST, confidence intervals for a proportion](https://www.itl.nist.gov/div898/handbook/prc/section2/prc241.htm), Wilson formula applied to simulated data. - [Microsoft, Customer payment predictions, 25 June 2026](https://learn.microsoft.com/en-us/dynamics365/finance/finance-insights/payment-insights-overview), payment timing probabilities. - [Hyndman and Athanasopoulos, judgmental adjustments, section 6.7](https://otexts.com/fpp3/judgmental-adjustments.html), method and limitations.