Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
  1. Home
  2. Blog
  3. Prioritising Collections With Machine Learning
AI & Machine Learning

Prioritising Collections With Machine Learning

Chasing by age and value is the default and it is not very good. Predicting who will pay without chasing, and who needs a call today, changes the cash outcome.

Updated 2 min readBy SpiderHunts Technologies

Free estimateNo obligation

Get a free estimate

Tell us what you need. A senior engineer reads every enquiry.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →

Quick answer — TL;DR

Most collections processes chase in order of age and value, which wastes effort on customers who were going to pay anyway. Predicting payment timing lets a small team focus where contact changes the outcome - and the relationship cost of over-chasing is real.

The default policy and its flaw

Standard practice is a ladder: reminder at seven days, call at fourteen, escalation at thirty, ordered by value. It is simple and defensible, and it spends a lot of a small team's time on invoices that would have been paid this week regardless.

Meanwhile an account that has quietly changed behaviour - paying later each month, disputing more lines - waits its turn in the age queue.

What actually predicts payment

Payment behaviour is one of the more predictable business patterns, because customers are consistent. The strongest signals are usually internal and already in your ledger.

  • The customer's own payment history - average days beyond terms, and its recent trend
  • Whether the invoice has ever been queried or partially paid
  • Invoice characteristics - value relative to their norm, whether a PO was referenced
  • Whether previous contact was needed, and what worked
  • Seasonal patterns in that customer's own cash cycle

External credit data adds something, particularly for newer customers, but for an established ledger your own history is usually the stronger signal.

Prioritise by what changes, not by what is likely

The same distinction that matters in marketing applies here. An account certain to pay on Thursday does not need a call on Wednesday. An account that will not pay whatever you do needs escalation, not another reminder.

GroupPays without contactAction
Reliable, minor delayYesAutomated reminder only
Responsive to contactOnly if chasedPriority call - the real value
Genuine disputeNot until resolvedRoute to the dispute owner, not collections
Serious riskNoEscalate, consider credit hold

Separating the dispute group matters more than it sounds. Chasing a customer for an invoice they have queried damages the relationship and does not produce cash.

Relationship cost is a real constraint

Collections differs from most prioritisation problems because the action has a downside. Over-chasing a good customer is not neutral - it irritates people who pay you, and in some sectors it is remembered at renewal.

Build that into the design explicitly: a contact frequency cap per account, and a rule that strategic accounts are routed to their account manager rather than the collections queue.

Chasing someone who was going to pay anyway costs you twice - the time, and the relationship.

Measuring it honestly

Days sales outstanding moves for many reasons, including sales mix and seasonality. Attributing an improvement to the model needs more care than a before-and-after chart.

Where possible, run the new prioritisation on part of the ledger and keep the existing policy on a comparable part for a period. It is the only way to separate the model's effect from a good quarter. Our note on incrementality testing covers the general approach.

FAQ

Frequently asked questions

The questions readers ask us after this guide.

Still have a question?

Ask us directly — a senior engineer will get back to you.

Ask about your project

How much history do we need?

Enough to see each customer's pattern repeat - typically a couple of years of ledger data, though customers with many invoices build a pattern faster.

Does this work for B2C collections?

The principle transfers but the signals differ, with less per-customer history and more reliance on broader behavioural and external data.

Will this replace the credit control team?

No. It reorders their day. The work is still human, particularly on disputes and relationships.

What about customers with only one or two invoices?

They fall back to policy rules and external data. A model should be allowed to abstain where it has no basis.

Keep reading

More on AI & Machine Learning

Start here

Want machine learning project details from us?

Tell us what you are trying to predict and roughly what data you hold. We will come back with an honest view on whether machine learning is the right tool, what the work would involve and a realistic cost range. If a spreadsheet would do the job, we will say so.

  1. You tell us what you needTwo minutes on the form, or a message on WhatsApp.
  2. A senior engineer reviews itAnd comes back with questions, a realistic range and an honest view on fit.
  3. Free 30-minute scoping callWe talk through scope, options and a realistic estimate — with no obligation.
Free estimateNo obligation

Talk to someone who builds this

Send a short brief and we will come back with an honest view and a realistic range.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →