Fictional scenario. Synthetic data. Your notes stay on this browser; do not enter real customer information.
Week three turns to the things that arrive after the fact, disputes, and the patterns that produce them. The thread is the same as before: observe, infer, keep the alternative, price the response.
Day 15 — Four families.
Dispute reason codes look like a long list and are really four families: fraud, authorisation, processing error, and consumer dispute. Each family wants a different kind of evidence and tells you a different thing about the transaction. A desk that files disputes by family rather than by code sees the shape of its problem in a week.
Today: pull last month's disputes and sort them into the four families. Write down the share of each. Then write one sentence on what the largest family suggests, and one on the innocent reason it might be large.
Read: Four families of dispute, four kinds of evidence.
Day 16 — A chargeback keeps its own calendar.
A dispute is raised weeks after the sale, and a month's dispute count is a mixture of several months' sales. Counting disputes by the month they arrive hides the sale month that caused them. Counting by the month of the original transaction, as a cohort, shows it.
Today: take last month's disputes and re-sort them by the date of the original sale. Draw the result, even roughly. If one sale week stands out, write down what happened that week. If you cannot find out, that is the note.
Read: A chargeback arrives in its own time. and the Volume 01 note Disputes and time.
Day 17 — Two prior transactions.
Under Visa's Compelling Evidence 3.0 rules, as summarised by the acquirer and PSP documentation we cite in the library, a merchant answering certain card-absent fraud disputes can point to prior undisputed transactions on the same credential, with matching data elements. The point for today is not the rule. It is that the evidence has to exist before the dispute, which means it has to be collected at the sale.
Today: for your top merchant, check which of these are stored per transaction: login or e-mail, IP address, delivery address, device reference. Write down which are missing. Those are disputes you cannot answer.
Read: Two prior transactions and a matching footprint.
Day 18 — When the cardholder is the customer.
Some disputes are raised by the person who made the purchase. Forgotten subscriptions, a family member's purchase, regret, or deliberate misuse. The transaction looks like every other good transaction because it was one. The signal is in the account's history, not the sale.
Today: find five fraud-coded disputes from last month where the account had prior undisputed purchases. For each, write one line: what in the account history suggests the cardholder was the customer, and what suggests they were not. Do not decide. Just write both.
Read: When the cardholder is the customer. and look up first-party misuse.
Day 19 — The refund that never shipped back.
Refunds leave through a different door from disputes, and most monitoring is pointed at the disputes door. A refund ratio by month hides the distribution by agent, by amount and by whether the original transaction exists at all.
Today: export last month's refunds and credits for one merchant. Count how many have no original transaction referenced. Count how many fall just below an approval limit, if you know it. Count refunds per agent. Write down the one distribution that surprised you.
Read: The refund that was never a return., then try the case Refund ring at Hollow Oak.
Day 20 — Forty welcomes.
An offer given once per account is an invitation to make accounts. The ring that answers it shares infrastructure rather than identity: devices, referral codes, timing, the shape of play, withdrawal destinations. The identities are the one thing it makes sure are different.
Today: if you have a sign-up offer, count accounts per device reference and per referral code over the last two weeks. If you do not, do the same for any once-per-customer benefit. Write down the top three and whether each has an innocent explanation you can check.
Read: One welcome bonus, forty welcomes. and the Volume 01 case Promo abuse.
Day 21 — A lead is not a finding.
Two accounts sharing a device is a lead. Twelve accounts sharing two devices, a referral code and a bank account is a finding. The distance between them is the work, and skipping it is how good desks accuse households and miss rings.
Today: take one shared attribute from yesterday and write down what else those accounts would have to share before you would call them one actor. Name the number. Then check whether they share it. Write the result in the second notebook, because this is the week's lesson about yourself.
Read: A shared attribute is a lead, a ring is a finding. and look up linked entities.
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.