The trial that keeps on giving.

Fictional scenarios and synthetic data. Reviewed 6 November 2026.

Hollow Oak Supplements runs a 14-day trial for EUR 1, converting to EUR 49 a month. In October they ran a campaign and trials tripled. In November the trial-to-paid conversion looked fine. In December Stilling Bank called Northbank about Hollow Oak's chargeback ratio, and Tomas cannot see why, because December's fraud numbers are ordinary. The answer is in October, and it is in two places.

What is actually being decided.

A one-off purchase is one risk decision. A subscription is a decision followed by a series of smaller ones that the merchant makes alone, without the customer present, every month until something stops it. The question for a risk team is not only "is this signup genuine?" but "will this signup still be a willing customer on the fifth rebill, and if not, how will they leave?"

There are two ways to leave. One is cancellation, which costs the merchant a customer. The other is a dispute, which costs the merchant the customer, the amount, a fee, and a tick in a ratio that the acquirer is watched on. the provider account of the dispute lifecycle lays out the fee and the process; the ratio side is the subject of the companion piece on VAMP and ECP. A subscription business that makes disputing easier than cancelling has designed its own ratio problem.

Three parties carry something. The merchant carries the cancellations, the disputes and the cost of acquiring trials that never meant to pay. The customer carries the friction of signup and of leaving. Northbank carries the portfolio exposure, and that exposure is where the December phone call came from.

Two populations wearing one trial.

A trial cohort contains two kinds of people who will never pay, and they need different treatment.

The first population never intended to. Some are individuals collecting a free fortnight; some are the mass-created accounts OWASP catalogues as OAT-019, set up to harvest whatever the trial includes, which for a product shipping physical goods means the goods. these signups show the shape of promo abuse described in the Volume 01 case: new accounts, clustered creation times, shared device references or delivery addresses, and payment instruments that fail or are cancelled before the first rebill. They do not usually dispute, because they do not usually get charged. Their cost is acquisition and stock, and it shows up in the conversion rate, not the dispute rate.

The second population did intend to, or thought they did, and then the rebill arrived. Some cancel. Some dispute, often under the dispute families that cover a cancelled recurring transaction or a transaction not recognised, and the piece on reason-code families explains why those families are answered with different evidence from a fraud dispute. A share of them are the subject of the article on first-party misuse: the cardholder is the customer, used the product, and is now saying otherwise. But the larger share simply did not remember, did not understand, or could not find the cancel button, and the dispute is the cancel button they found.

The two populations look the same at signup and completely different sixty days later. Trial abuse depresses conversion. Rebill regret inflates disputes. Treating them as one "trial risk" problem means applying signup friction to fix a cancellation problem, or loosening cancellation to fix an abuse problem, and neither works.

The cohort is the unit of analysis.

Why could Tomas not see the December problem in December's numbers? Because he was looking at December's transactions. Disputes on a rebill arrive weeks after the rebill, and the rebill arrives weeks after the trial; the Volume 01 lesson on time and outcomes and the Volume 02 article on dispute lag and cohorts are both about this. The October campaign cohort hit its first rebill in late October and early November, its disputes arrived through November and December, and December's ratio is October's decision.

A synthetic illustration. Suppose a normal month brings 2,000 trials, of which 1,000 convert, 950 of those stay past the second rebill, and 20 dispute at some point. The campaign month brings 6,000 trials. Conversion holds at 50 per cent, so 3,000 convert, and the headline looks healthy. But the campaign brought a different audience: suppose 12 per cent of converters dispute instead of 2 per cent. That is 360 disputes rather than 60, arriving across the following two months.

Now inspect the denominator and its period. If a monitoring metric uses a previous-month transaction count, a campaign ending can change that denominator while older purchases continue to generate disputes. Compare purchase cohorts as well as the agreed reporting ratio. Confirm current network criteria separately; a fictional example does not establish programme entry.

Measured by signup cohort, none of this is a surprise. The October cohort's conversion, retention and dispute rate are three lines on one chart, and by mid-November the third line was already steeper than any cohort before it.

The first transaction carries the rest.

The recurring-transaction exemption in Delegated Regulation 2018/389 allows a series of transactions to the same payee for the same amount to proceed without strong customer authentication after the first one is authenticated. The structure is the point: the first transaction is where the cardholder is present and can be authenticated, and every rebill afterwards rests on it. A EUR 1 trial authenticated properly is a cheap foundation. A EUR 1 trial waved through on an exemption, because it is low-value and friction at signup hurts conversion, leaves every subsequent EUR 49 resting on nothing, and the article on who carries the risk when 3DS is skipped explains what "nothing" costs when the dispute comes.

The alternative explanation to keep alive, for a merchant with rising rebill disputes: the problem may not be the customers. A rebill descriptor that does not match the brand the customer remembers, a trial length that does not match the marketing, or a cancellation path that requires a phone call during office hours will produce rebill disputes from perfectly ordinary customers. Before concluding that the campaign bought bad traffic, read ten dispute narratives. If they say "I did not recognise this", the fix is on the statement. If they say "I could not cancel", the fix is on the website. Neither is a fraud rule.

A failure mode to watch.

A consequential mistake is to fight rebill disputes at signup. Trials get harder to start, conversion falls, and the dispute rate on the trials that do convert barely moves, because the people who dispute a fifth rebill were not the people the signup friction would have stopped. The merchant has paid for a worse funnel and kept the ratio problem. Fight trial abuse at signup, with the link-analysis methods that find clustered accounts. Fight rebill disputes at the rebill: a reminder before it, a descriptor that matches, a cancel button that works, and a refund-on-request path for the first rebill after a missed cancellation, which the piece on refund abuse would call a goodwill refund and would keep open.

A way to work it.

1. Build the cohort table: signup month against conversion, second-rebill retention, cancellation and dispute rate. Lag the dispute column by at least 60 days before reading it.
2. Split each cohort's non-payers into "never charged" and "charged, then disputed". Neither category establishes intent; investigate legitimate cancellations, payment failures, unclear terms and misuse separately.
3. For trial abuse, run the promo-abuse checks at signup: account clustering, shared device and address, instrument status before first rebill.
4. For rebill regret, read ten narratives. Fix the descriptor, the reminder and the cancel path before touching a rule.
5. Confirm the first transaction is authenticated and that the recurring exemption on rebills rests on it. Ask the acquirer what their exemption data shows.
6. Watch the ECP denominator across a campaign and its ending. Tell Stilling Bank about the cohort before the ratio tells them.

Take it to the desk.

For last quarter's trial cohort, what share converted, what share disputed, and what share did both? If we do not know the third number, what would it take to know it? the answer is one join between the subscription system and the dispute feed, and nobody has owned it.

Try it: Refund ring at Hollow Oak, and the Volume 01 cases Promo abuse and Dispute spike. Read next: A chargeback arrives in its own time. Look up: cohort, first-party misuse.

Edited 6 November 2026. Primary references are listed above; fictional scenarios are not claims about Antifrog customers or operating experience.

Fictional learning material. Any numerical examples are synthetic. Use only authorised, appropriately minimised records for workplace exercises; do not upload customer information to this learning site.