Cash can be in the bank while the customer ledger still shows an unpaid invoice. That gap creates work: collectors chase the wrong balance, customers resend remittances and finance investigates receipts that have already arrived.
Cash application connects a receipt to the right account and open items, with the appropriate treatment of differences. Straightforward references can make that task highly repeatable. Partial payments, deductions and consolidated remittances make it more demanding.
AI is worth considering where the information needed to resolve a receipt exists but is difficult to assemble or interpret. The first step is to understand what the existing matching rules already handle.
Start with the capabilities in the finance system
Business Central documents automatic payment application based on bank statement information and open entries. Its rules can specify review requirements, and suggested matches carry confidence categories. These are useful controls and should be evaluated before adding another solution. Microsoft Learn, automatic payment application
Dynamics 365 Finance also documents automated cash application through reconciliation rules, including matching statement lines to open customer invoices using references such as invoice numbers or payment IDs. Microsoft Learn, automated cash application
The opportunity for additional workflow automation starts with the receipts those mechanisms do not resolve satisfactorily. Inspect whether the obstacle is missing data, inconsistent references, unclear deductions or a genuine customer dispute. Each calls for a different response.
Distinguish cash collection from cash application
APQC's June 2026 analysis reported a median days sales outstanding figure of approximately 38 days, compared with 30 days or less for top performers and 46 days or more for bottom performers. The article also reported that 41% of organisations in its AI research were using AI in order-to-cash processes. Those are broad research findings; the public article does not provide the sample detail needed to treat them as a precise peer benchmark for every business. APQC, invoicing and accounts receivable automation
Neither figure establishes that improving cash application will shorten collection time by a particular amount. A receipt already in the bank is already cash received. Allocating it sooner improves the accuracy and timeliness of the receivables position.
That improvement may help collectors focus on genuinely unpaid invoices and avoid unnecessary customer contact. Any resulting acceleration in future collections needs to be measured separately. Counting the same receipt as both newly collected cash and a faster application would overstate the benefit.
Follow a short payment to a defensible outcome
Imagine a customer pays £9,700 against a £10,000 invoice. The remittance says “less agreed credit”, but the expected credit note is not present in the ledger. A salesperson's email refers to a £300 concession.
An agent could find the remittance, locate the correspondence and assemble a proposed explanation. The workflow still needs to establish whether the concession was authorised, whether a credit note is pending and how the remaining balance should be treated under policy.
The ability to find a plausible explanation should not itself confer write-off authority. A sound process can apply the amount received as permitted, preserve the unresolved difference and route the commercial decision to the authorised owner.
This fictional example shows why interpretation and accounting action need separate acceptance criteria. The system may correctly understand the customer's claim while still lacking sufficient authority to settle the account completely.
Build a resolution path for each receipt type
| Receipt pattern | Likely source of difficulty | Evidence to establish before completion |
|---|---|---|
| Exact reference and amount | Duplicate references or an already settled item | Correct entity, account, invoice and current open balance |
| Partial payment | Allocation across several open items | Customer instruction or an approved allocation rule |
| Consolidated payment | One receipt covering many invoices | A reconciled allocation schedule with totals that agree |
| Deduction | Discount, dispute, fee or unapproved short payment | Reason, supporting record and required approval |
| Unidentified payer | Incomplete or ambiguous bank information | Reliable customer identification; ambiguity retained if unresolved |
Use deterministic checks for totals, currencies, entities and outstanding amounts. Use AI where interpreting or finding information adds value. Keep the permitted posting actions narrow enough to enforce the company's policy.
Cross-entity payments deserve particular care. A familiar customer name does not establish that a receipt belongs to the entity or invoice suggested by the first search result.
Evaluate accuracy alongside coverage
Two measures are especially useful. Coverage is the share of receipts the workflow can resolve automatically within scope. Precision is the share of those automatic allocations that are correct. A high coverage figure is unattractive if it comes with unacceptable misallocation.
Measure both by receipt type and monetary value. A small number of high-value errors can disappear inside a volume-based average. Track reversals, customer complaints, unapplied cash ageing and the time needed to correct mistakes.
Also distinguish a system-generated confidence score from a demonstrated error rate. A score becomes operationally useful when validation shows how it relates to actual outcomes for the relevant population. It should not be treated as a universal probability without that evidence.
Put the retained work into the economics
Suppose a team receives 5,000 payments a month and 20% need investigation. At eight minutes each, the 1,000 difficult receipts consume approximately 133 hours. Reducing that effort to four minutes releases about 67 hours a month.
At an assumed loaded rate of £35 an hour, annual capacity released is £28,000 before implementation and operating costs. These are illustrative inputs, not industry averages or Curia results. The calculation should include checking and correction time, and it does not assume any increase in cash receipts.
For some businesses, that labour benefit alone may be too small to justify an additional service. Others may also value a more current ledger, fewer incorrect collection requests and improved handling of growth. State those outcomes and measure them rather than assigning an unsupported financial value.
Where Curia fits
Curia assesses the workflow across existing systems, identifies the unresolved work and builds a contained proof using the client's data. For cash application, that proof should show which receipts can be resolved, which need judgement and whether the total operating effort falls.
The managed service proposition matters when finance needs the surrounding evidence gathering, routing and maintenance operated consistently. It should build on useful existing rules and address the demonstrated gaps. For an explanation of the cost boundary, see what an AI finance workflow costs.
Bring Curia a sample of unapplied receipts, remittances and the current matching rules. We will assess where additional workflow automation could help. Discuss cash application