Dunning

Where the word comes from, and its two meanings

The verb to dun, meaning to press someone persistently for payment of a debt, has been in English since the seventeenth century. The popular story that it derives from a bailiff named Joe Dun is folk etymology and the real origin is uncertain. What survived intact is the tone: dunning has always meant chasing money.

Two distinct practices now carry the name, and the people searching for the word are looking for both. In accounts receivable, dunning is the escalating sequence of reminders sent against an unpaid invoice — a reminder, a firmer notice, a final demand, then referral to collections. The money is genuinely owed, the customer knows it is owed, and the sequence is a negotiation.

In subscription billing, dunning is the automated recovery that runs after a stored payment method is declined at renewal. Nothing is owed in the accounts-receivable sense. The customer intended to pay and a credential or a balance got in the way, so the sequence is mostly machinery, and in the majority of cases it resolves before anybody reads an email.

The distinction is load-bearing, because the accounts-receivable playbook imported wholesale into subscriptions produces exactly the wrong behaviour: escalating tone aimed at a customer whose card was simply reissued, and a collections mindset applied to somebody who is not a debtor.

The retry ladder, and what each rung is for

A dunning ladder is a sequence of re-authorisation attempts spread across a recovery window, usually paired with outreach after the first failure or two. The useful way to design one is to ask what each attempt is testing, because an attempt that tests nothing new is a guaranteed decline with a cost attached.

The first retry tests whether the failure was transient. The middle rungs test whether the account has been replenished. The last rung tests whether the customer acted on the outreach. If two attempts land in the same conditions with nothing changed between them, one of them is waste.

Hard declines never enter the ladder at all. They exit to a credential request on the first failure, because there is no condition that can change on a closed account. The rungs below are placed against the customer's cash cycle rather than at even intervals.

Why timing matters more than count

Adding a fifth rung to a ladder costs almost nothing and recovers almost nothing, because the fifth attempt lands in the same conditions as the fourth. Moving the second rung from day two to day four costs nothing at all and changes what it is testing. Attempt count is a dial with a very low ceiling; timing is not.

Three constraints set the shape. Insufficient funds is a timing problem, so an attempt is only worth making when money has plausibly arrived, and income arrives weekly, fortnightly or monthly — evenly spaced retries can miss every replenishment point in a cycle. Time of day matters for the same reason. And card network rules bound how long a merchant-initiated transaction stays valid against the original authorisation, so a ladder that runs past that window is generating failures that were invalid before they were attempted.

A worked example, with inputs stated as an example rather than a benchmark. Take a store billing 10,000 renewals a month with a 6% failure rate: 600 failed renewals, each carrying a $45 charge, so $27,000 of at-risk billing in one month. Recovering ten percentage points more of those failures — 60 subscriptions — returns $2,700 of that month's billing. The figure that matters is larger, because each recovered subscription keeps its remaining renewals too: at $18 of contribution per renewal and eight further renewals expected, those 60 subscriptions carry another $8,640 of contribution.

Soft and hard declines are your classification, not the issuer's

The card networks do not return a field marked soft or hard. What comes back is a response code from the issuer, and the soft/hard split is a mapping that a merchant or a gateway lays over those codes. Two consequences follow.

The first is that the mapping is a policy decision you own and can get wrong. A generic do-not-honour is the ambiguous case: issuers return it for short balances, for risk holds and for reasons they will not disclose. Treating every one as hard abandons recoverable subscriptions; treating every one as soft burns attempts on subscriptions that were already gone.

The second is that some failures never reached an issuer at all. Gateway validation errors, invalidated tokens, missing authentication and network timeouts appear in the dashboard alongside real declines and carry no issuer opinion whatsoever. Retrying those is not a recovery strategy; the failure is on your side of the wire.

• Hard under any reasonable mapping: closed account, stolen or lost card, revoked mandate, invalid account number. Attempts here are not merely wasted, they accumulate declines against your merchant profile.
• Soft under any reasonable mapping: insufficient funds, over limit, issuer or network unavailable, transaction not permitted at this time.
• Ambiguous, and worth a bounded number of spaced attempts: do not honour, generic decline, restricted card.
• Not a decline at all: gateway errors, tokenisation failures, authentication required. Fix or escalate these; retrying them only reproduces the bug.

The email leg, and where dunning ends

Outreach carries the residual: the failures where the stored credential is genuinely dead and no retry will ever succeed. Its entire job is to get a working payment method onto the account with as little friction as possible, which makes the update link the whole feature. A link that lands on a login wall loses a share of people who had every intention of paying, and a password reset placed between a customer and their own subscription is the most expensive interface decision in the sequence.

Tone follows from the accounts-receivable distinction. The customer has not refused to pay; something broke. Two or three messages across the recovery window, stating plainly what happened and what to do, do more than an escalating sequence, and a discount belongs in win-back after recovery has genuinely failed rather than in the second email — offering one earlier concedes a price objection nobody raised.

Dunning ends when the ladder is exhausted and a terminal decision is taken: pause, downgrade or cancel. That moment is also where involuntary churn is booked, which makes the boundary partly a setting rather than a fact. A business that stops retrying at day 7 and one that stops at day 21 report different involuntary churn on identical customer behaviour, so the recovery window belongs beside the number whenever the number is quoted.