Skip to content

Lesson 4 of 6

The Customer Case: Scoping a Vague Ask Out Loud

Run a customer case interview from a vague opening line to a first piece of work the customer agrees with, using five steps you can practise.

6 minForward Deployed Engineering

After this lesson you will be able to run a customer case interview: ask the questions that find the real problem, agree what success means, propose a first piece of work, and say what is out of scope.

There is no code in this lesson. The skill is a conversation, and you practise it by speaking.

What the round is

In a customer case the interviewer plays a customer, and you play yourself on your first call with them. It usually lasts thirty to forty-five minutes. The customer opens with one or two sentences that are vague on purpose. For Calder Freight, the opening might be:

"We keep losing money on invoicing. We want to use AI to fix it. What would you build?"

The last sentence is the trap. If you answer it, you design a system for a problem you have not understood, and that is the mistake the round exists to catch. The interviewer holds a page of facts about the customer and gives them out only when asked. Your score depends on which facts you draw out and what you do with them.

Five steps to hold on to

You need a structure simple enough to remember under pressure. Use these five steps:

  • Restate. Say back what you heard in one sentence and check you have it right.
  • Walk the process. Ask the customer to describe what happens today, step by step.
  • Size the cost and learn the limits. Ask how often it goes wrong, how anyone finds out, who owns the systems, and what is not allowed.
  • Agree success. Ask what would be different in three months, and turn the answer into something you can count.
  • Propose. Offer the smallest first piece of work, say what is out of scope, and name the next step.

The first four steps are questions. Only the fifth is an answer. In a forty-minute case, expect to spend about twenty-five minutes on the first four.

The conversation, step by step

Here is how the Calder Freight case can go. Read it aloud, both parts, to hear the pace.

You: "So that I have it right: money is being lost somewhere between a shipment going out and its invoice being paid, and you want that to stop. Is that fair?"

Customer: "Yes. Finance says we are under-billing."

You: "Can you walk me through what happens after a truck leaves the warehouse?"

Customer: "The warehouse team marks the shipment as sent in the warehouse system. Every night that system exports a file of the day's shipments. Each morning someone in billing opens the file and types the invoices into the billing system."

You: "They type them by hand? How many?"

Customer: "By hand. About two hundred a day."

You: "How did finance find the under-billing?"

Customer: "At the end of the quarter they compared the totals. They were out by about two percent. Nobody knows which shipments."

You: "So if one shipment is never invoiced, who finds out, and when?"

Customer: "Nobody, until the quarter ends."

You: "Who looks after the two systems? Could I be given the exports?"

Customer: "IT owns both. The exports are easy. The billing system is ten years old, and the only way to get data out of it is an export file."

You: "Is there anything about this data that must not leave your network?"

Customer: "Customer names and prices. Our contracts say they stay on our servers."

You: "Last question for now. If this were fixed, what would be different in three months?"

Customer: "Every shipment invoiced, for the right amount, within one working day."

Look at what eight questions bought. The problem is not invoicing in general. Two hundred invoices a day are typed by hand, nothing checks them, and a mistake stays hidden for up to three months. You know who owns the systems, what you can get, and what must stay inside the network. You have a success measure in the customer's own words. And nothing so far calls for AI.

Propose the first piece of work

Now answer, and keep it small. Something like this:

"Here is what I would do first. Each night a small tool compares the shipments export with the invoices export. Each morning it gives billing a list: shipments with no invoice, invoices for the wrong amount, shipments invoiced twice. It runs on your servers, so no names or prices leave. It does not stop the mistakes, but each one is found in a day and not in a quarter. I would judge it by one number: how many findings are still open after one working day."

Then say what is out of scope, and why:

"I would leave automating the typing out of this first piece of work, and I would leave AI out too. We do not know yet which mistakes are common. The list will tell us within a few weeks. If most of them come from typing, the next step is to create the invoices from the export. If they come from prices that live in contract documents, that is where a model that reads documents could earn its place."

Then name the next step, with a date if you can: "Could I have last week's exports from both systems, and an hour with the person who types the invoices?"

The customer asked for AI and you proposed a comparison of two files. That is not a refusal. You have told them what you would need to see before AI is the right tool, and you have offered something that pays back in the first week. Interviewers who have worked with customers recognise that as the job.

Mistakes that sink a case

  • Designing in the first minute. Any sentence that begins "I would build" before you have walked the process is too early.
  • Taking the requested solution as the problem. "We want AI", "we want a dashboard" and "we want a chatbot" are ideas about a solution. Ask what happens today.
  • Interrogating. Twenty questions read from a list, with no reaction to the answers, is not a conversation. Let each answer choose your next question.
  • No number. "Make invoicing better" cannot be checked. "Findings open after one working day" can.
  • No limits. If you never ask who owns the systems or what cannot leave the network, your proposal may be one the customer is not allowed to accept.
  • Nothing out of scope. A proposal that promises everything tells the interviewer you will promise a customer everything.
  • Jargon. The person you are talking to may run finance or operations. Say "a file the system writes every night", not "a batch export".

How to practise

Reading about cases does little. Speaking them does a lot. Ask a friend to play the customer. Give them an opening line and ask them to invent the facts behind it in advance and keep them to themselves: what happens today, how often it goes wrong, who owns what, one thing that is not allowed. Set a timer for twenty minutes.

Afterwards, write half a page: the problem in one sentence, the success measure, the first piece of work, what is out of scope, and the next step. If you cannot write one of those five, that is the step to work on in the next practice. Keep each half page in notes.md.

If you have nobody to practise with, record yourself. Say your questions aloud, pause, and answer as the customer would. It feels odd, and it still shows you where you jump to a solution.

Key takeaways

  • The opening line of a customer case is vague on purpose; the round tests what you ask before you propose anything.
  • Restate, walk the process, size the cost and learn the limits, agree success, then propose.
  • The solution the customer asks for is not the problem; ask what happens today to find the problem.
  • Propose the smallest first piece of work, a number to judge it by, and what is out of scope.
  • Practise out loud, with someone playing the customer, and write a half-page summary afterwards.

Try it

Practise a case with this opening line: "Our support team is drowning in tickets. Can you build us a chatbot?" Ask a friend to play the customer and to decide in advance, without telling you, how many tickets arrive a day, what most of them are about, who answers them, and one thing the company will not allow. Give yourself twenty minutes, then write the half-page summary: the problem in one sentence, the success measure, the first piece of work, what is out of scope, and the next step. If you have no partner, write the questions you would ask, in order, with the reason for each. A good result is a first piece of work that is smaller than a chatbot, or a clear reason why the chatbot is the right place to start.

Official documentation

Ready to use this in a job?

Rolebound applies to matching roles for you.

Start now

Written with AI and checked automatically. Nobody has run these steps by hand, and tools change, so a command may behave differently in the version you install. Spotted a mistake? Tell us.

    Cookies on Rolebound

    We use necessary cookies to sign you in and keep the site working — those cannot be turned off. We would also like to set analytics and advertising cookies so we can see which pages help people get interviews and which adverts brought them here. Nothing optional is set unless you say yes, and you can change your mind at any time from “Cookie settings” in the footer. Read our Privacy Policy.