Effective debt recovery: what repayment research tells us

Reminders help, but not equally in every setting

In an Australian BETA experiment, people previously making minimum credit card repayments repaid an average AUD 134 more in the next month after an SMS than the control group. No equivalent email effect was detected. This is a consumer credit result, not a promise of improved Polish invoice recovery.

BETA — Credit when it’s due

Recognisable contact matters

In a Philippine microloan study, SMS reminders did not improve repayments on average. A message naming a previously known loan officer did help returning customers. Our design inference is to identify the creditor clearly, maintain sender continuity and refer to a specific invoice. This does not prove that adding a name improves every case.

IPA — Text Message Loan Repayment Reminders

A response calls for a different approach

Three experiments involving over 32,000 debtors with old arrears found a small effect from repeated reminders. Extra notices after payments resumed could worsen outcomes. Saldora therefore lets the creditor pause a case while reviewing a complaint or agreeing a date. Arrangements are currently handled manually; we do not promise automated negotiation or instalment management.

Saulītis (2023) — Three field experiments

Measure payments, not just clicks

FCA credit card research found that changing a customer’s choice does not necessarily reduce their debt. For Saldora’s pilot, we therefore propose reviews at 30 and 60 days: the share of money recovered, fully settled cases and time to first payment. Contact cost, complaints and incorrect recipients are tracked separately.

FCA — Helping credit card users repay their debt

What does this mean for Polish B2B invoices?

We start with correct data and a simple reminder, keep a consistent sender, and make payment and replies easier. Escalation requires a current balance and a timely decision. The +1, +4, +7 schedule is a pilot starting point, not a scientifically established optimum. Consumer loan findings guide testing; they are not our own recovery statistics.

What the number of messages sent does not measure

The easiest metrics to collect — SMS counts, email open rates, clicks — describe system activity, not debtor behaviour. You can raise them by sending more, which has no bearing on whether the money came back. So the first metric for any case should be the amount recovered against the amount submitted, and the second the time from submission to payment.

Market effectiveness for amicable recovery sits between 34 % and 42 %. That means most cases do not conclude at the pre-litigation stage — and that a fixed-fee model requires the client to understand what they are paying for. They pay for work performed and for documentation that holds up in court, not for a guarantee of outcome that nobody can honestly give.

A third metric, often skipped, is the share of cases stopped because the debtor had already paid before any contact. In another system we measured that 7 of 15 candidates for a demand had already settled, and data snapshots were on average 192 days old. Sending a demand to someone who has paid is not a technical error but the loss of a commercial relationship.

What to do when the debtor replies

Automation is excellent at sending. The case begins when the debtor writes back "I have paid", "I disputed that invoice" or "that is not my company". Each reply calls for different action and none may be ignored — this is where full automation ends and where a complete contact log is meant to help.

A report of payment has to be checked against your own records, not ours: Saldora deliberately does not connect to your bank account and does not guess whether money arrived. Recording the payment in the panel stops further contact, and before every irreversible step the system asks about the case status again.

A complaint or a dispute over whether the invoice is due changes the nature of the case. A disputed claim is a poor candidate for the electronic writ-of-payment procedure — the debtor will object, the order will lose force, and you will have paid the court fee to move into ordinary proceedings. Better to resolve the dispute early than to learn of it from the court.

How we measure this ourselves

Every case carries a full record in the panel: which messages went out, which were refused and why, when the demand was produced and who approved it. Statistics show how many recovery SMS messages and emails actually went out and how many cases reached which stage — figures about work performed, without interpretation.

We separately watch the things whose absence used to silence a channel without a sound: an unavailable provider, an unapproved sender name, an empty balance with the SMS operator. The readiness panel states them plainly, because a channel that does not send while looking healthy is more dangerous than one that is switched off.

We do not publish our own effectiveness figure until we have enough cases for the number to mean more than an anecdote. When we do, we will show it together with the number of cases it was computed from — because a percentage without a denominator is not a result, it is a slogan.