Several customer matches
A support agent looks a customer up and finds more than one account. The tempting answers pick one by a sensible-sounding rule: the most recent account, the closest name. Each is a guess, and a wrong guess changes a stranger's account. The exam's answer is to ask the customer for an identifier that singles out one record.
D5 Context & ReliabilityExam guide 5.2About 12 minutes
Start here: the pharmacist
A pharmacist looks up a prescription for “Priya Sharma” and finds three patients with that name.
A support agent is in the same spot. Its search tool (say get_customer) can return several accounts for one name, or for an email that a household shares. An identifier is a detail that only one record has: an order number, a postcode, a date of birth. Asking for one is cheaper than undoing a refund sent to the wrong card. A person is for the real triggers: the customer asks for one, the case falls outside policy, or the agent can't make progress.
- One match, or the customer already gave an order number: go ahead.
- Several matches: ask for an identifier that only one record has, then continue with the verified match.
- The customer can't or won't give one: now the agent can't make progress, which is a trigger for a person.
- Never pick by a rule of thumb (newest, closest name), and never show the other matches' details to choose from.
Words in the question that tell you which side you're on
Ask for an identifier
- two accounts match, same name
- several customers, more than one result
- ambiguous, a shared email
- “picks the one with the latest order”
Go ahead
- one match
- verified ID
- order number given
Try it: play eight chats
Pick what the search returns, then the policy for several matches. The same eight chats play out each time, and each card shows whose account was actually changed. The diagram above shows where your policy acts.
The chats, names and the 15-minute wait are illustrative, chosen to show the shape. The exam guide's point is that several matches call for identifiers, not a heuristic pick.
Things to try: three Priyas + most recent (five wrong people), then three Priyas + best name match (still three wrong). Then three Priyas + ask for an identifier, and watch chat 7. Last, shared email + escalate to a human: all right, all waiting.
The four traps
| Trap | What the wrong option looks like | Why it loses |
|---|---|---|
| Heuristic guess | “Pick the account with the most recent order” | Works most of the time, but when it's wrong the agent acts on someone else's account. |
| Heuristic guess, dressed up | “Pick the match whose details are closest to the conversation” | Closeness is still a guess. A town mentioned in passing or an old address points at the wrong person, and a wrong account is costly. |
| Overbuilt | “Escalate to a human whenever there are several matches” | It works, but one clarifying question settles it without a handoff. Several matches is not one of the escalation triggers. |
| Leaks data | “Show the customer the details of all matching accounts so they can choose” | It settles the choice, but it shows other customers' information to someone who hasn't been verified. |
Practice: three exam-style questions
Answer first, then open “Think it through”. The third is a close cousin: when a person is the right answer.
Rule to remember
Exam guide 5.2: “Multiple customer matches → ask for identifiers (not heuristic selection).” And, for when a person is the answer: “Triggers: customer requests a human; policy exceptions/gaps; inability to progress.”