0% completed
Questions for the First Ten Minutes
On This Page
Why a Fixed Order Helps
Minutes One and Two: The Decision
Minutes Three and Four: The Current Process
Minutes Five and Six: What Was Tried
Minutes Seven and Eight: The Data and Access
Minutes Nine and Ten: Limits and Success
What the Ten Minutes Produced
How to Ask
Using This in an Interview
Key Takeaways
Practice Questions
You have ten minutes with a customer's manager before the manager leaves for another meeting. You know the customer bought an AI product, and you know the project has stalled. What do you ask first?
Most engineers start with the technology: which model, which database, which cloud. Those questions can wait, because the answers rarely change what gets built in the first week. Early questions should be about the work.
Why a Fixed Order Helps
A first conversation should follow a fixed order, because the early answers decide which later questions are worth asking. If the decision is unclear, questions about data come too early.
The FDE used five groups of questions, about two minutes each:
- The decision: what choice the system should make or support, and how often.
- The current process: who does the work today, how long it takes, and what it costs.
- What was tried: any earlier attempt, and why it stopped.
- The data and access: where the inputs are stored and who can grant access.
- Limits and success: rules the system must obey, and the result that would count as success.
Minutes One and Two: The Decision
The claims manager had asked for "AI to take over our claims", so the FDE narrowed it:
"Which single step in handling a claim takes the most staff time before the claim is assessed?"
Routing did: about 1,200 claims a day into six queues, all by hand.
Plan change: the project became claim routing. Payment and damage assessment were left out of the first version.
Minutes Three and Four: The Current Process
"How long does routing take per claim, and how long does a claim wait before it reaches the right queue?"
The answer was about 5 minutes a claim and about 2 days of waiting. The FDE did the arithmetic aloud: 1,200 claims at 5 minutes is 6,000 minutes, or 100 adjuster-hours a day. The claims manager confirmed that it matched the team's own estimate.
Plan change: the project now had a baseline, a measure of today's process that later results are compared against.
Minutes Five and Six: What Was Tried
A stalled project's history often contains the most useful fact.
"You mentioned an earlier test of the model. What did it do, and why did it stop?"
The insurer's engineers had sent each claim form to the model as text. The demo misrouted 1 claim in 4, and nobody could say why. The FDE asked: "Did the test include the photos or the police reports?" Only the form text had been sent.
Plan change: the FDE stopped assuming that the model was the problem and planned the first week for reading real claims with the adjusters.
Minutes Seven and Eight: The Data and Access
Data questions in a first meeting are about location and permission, not format. Formats can be studied later, but access can take weeks, so it must be requested early.
"Where are the form, the photos, and the police reports stored, and who approves access to each?"
All three were in the claims system, and access to its API needed approval from the insurer's security team.
Plan change: the FDE asked the claims manager to start the security request that same day. Had it waited until the end of the week, the prototype would have started late.
Minutes Nine and Ten: Limits and Success
"Are there rules about how claims must be handled that any system would have to follow?"
The regulator requires the insurer to show how each claim was handled, including automated decisions. The claims manager did not know what record it expected and offered a meeting with the compliance officer. Success, in three months, meant "adjusters spend their time on claims, not on sorting them."
Plan change: a decision log was added to the design before any code existed, and a compliance meeting was booked for the first week.
What the Ten Minutes Produced
| Question group | Main answer | Change to the plan |
|---|---|---|
| The decision | Routing takes the most staff time; 1,200 claims a day, six queues | Scope narrowed to routing |
| The current process | About 5 minutes a claim; 2 days of waiting | Baseline of 100 adjuster-hours a day |
| What was tried | Demo sent only form text; misrouted 1 claim in 4 | Week 1 spent reading claims with adjusters |
| The data and access | Claims API needs security approval | Access request started on day 1 |
| Limits and success | Regulator requires a record of every decision | Decision log added; compliance meeting booked |
Five answers changed the plan five times, and none of them involved choosing a model or a database.
How to Ask
How a question is asked matters almost as much as which question is asked.
- Ask for numbers. "How long does it take?" gets "a while". "How many minutes per claim?" gets "about 5".
- Ask about the last real case. "Tell me about the last claim that went to the wrong queue" gets a specific story. "Do claims go to the wrong queue?" gets "sometimes".
- Repeat the answer back. "So routing is 1,200 claims at 5 minutes, about 100 hours a day?" lets the customer correct a mistake before it enters the plan.
End with a written summary sent the same day: five lines, one per question group, with next steps.
Using This in an Interview
Customer simulation rounds often begin with a vague request and a time limit. Ask one question from each group before proposing technology, and say aloud how each answer changes your plan.
"Before I suggest anything, I would like to ask about the decision, how it is done today, what you tried, where the data is stored, and what rules apply. Routing at 5 minutes for 1,200 claims is about 100 hours a day, so I would start there. What happened with the earlier test?"
Expect two follow-up questions:
- "The customer only wants to talk about which model to use. What do you do?" Answer briefly, say the model choice depends on the data and the decision, and return to those.
- "What if the customer does not know the numbers?" Ask for an estimate, note who could measure it, and plan to measure it yourself in the first week.
Key Takeaways
- Cover the decision, the current process, what was tried, the data and access, and the limits and success, in that order.
- A stalled project's history often contains the most useful fact, like the demo that sent only form text.
- Ask about data location and permission early, because access approval can delay the whole project.
- Ask for numbers and recent real cases, repeat answers back, and send a written summary the same day.
The next lesson, Mapping the Customer's Data, continues the data questions and maps every source the routing system needs.
Practice Questions
1. Why does the FDE ask about the earlier demo before asking about the data?
<details><summary>Show answer</summary>Because the reason the demo failed decides which data questions matter. The demo sent only form text, which told the FDE to study the photos and the police reports.
</details>2. Why did the FDE ask the claims manager to start the security request on the first day?
<details><summary>Show answer</summary>Because access approval can take longer than any engineering task in the first week. A later request would have delayed the prototype.
</details>3. The claims manager did not know what record the regulator expected. What was the right response?
<details><summary>Show answer</summary>Book a meeting with the person who does know, and add the requirement to the design now. Leaving it until the end risks a system that cannot go to production.
</details>Reading Progress
0%
On This Page
Why a Fixed Order Helps
Minutes One and Two: The Decision
Minutes Three and Four: The Current Process
Minutes Five and Six: What Was Tried
Minutes Seven and Eight: The Data and Access
Minutes Nine and Ten: Limits and Success
What the Ten Minutes Produced
How to Ask
Using This in an Interview
Key Takeaways
Practice Questions