What are your first five discovery questions for any new customer problem, and why those?
Answer it out loud first, then compare. That's what makes it stick.
One: walk me through a real case today, step by step, so I see the actual process. Two: how many cases, in what formats, and how often do the rules change. Three: what does a wrong answer cost and who catches it today. Four: what number would tell you this worked. Five: what blocks us, meaning data access, security review and who signs off. These move from understanding to measuring to constraints.
- Current process, one real case
- Volume, formats, change rate
- Cost of errors and who catches them
- Success metric and baseline
- Constraints: data, security, owners, deadline
A good question list is not a script. It is a map from what you do not know to what would change your plan. Each question should have a reason you can say in a sentence.
1. 'Walk me through the last case you handled.' Stories beat summaries. You learn the real tools, the hand-offs and the unwritten steps that no process diagram contains.
2. 'How many, what formats, how often does it change?' Volume decides whether automation pays off (100 cases a year versus 1,000 a week). Formats and template churn decide how hard extraction will be.
3. 'What happens when it goes wrong, and who catches it?' This sets your accuracy bar and your review design. A typo in an internal note and a wrong billing code sent to a payer are different products. Interview write-ups list cost of a wrong answer and who catches errors today among the things interviewers want pinned down.
4. 'What would make you say this worked, and what is the number today?' No baseline, no proof. Get a metric in business terms (hours per case, days to close, rework rate) before technical terms.
5. 'What could block us?' Data access, security review, who owns the systems, who signs off, any hard date. Most real deployments stall here, not on the model.
Order matters. Go from the customer's world (process, pain) to measurement (success) to constraints. Start with constraints and you sound like a compliance form; start with technology and you sound like a vendor.
Follow-up habits:
- Ask 'why' once more on any vague answer ('it takes too long' becomes 'it takes 3 days because it waits for sign-off').
- Prefer open questions first, closed ones to confirm.
- Write down assumptions you had to make and read them back.
Numbers to quote: one prep guide suggests asking three to five scoping questions before you propose anything; five good ones beat fifteen scattered ones.
- Which of these five would you drop if you only had two minutes?
- The customer cannot answer question four. What do you do?
- How do you ask about error cost without sounding negative?
- How Distyl AI hires forward deployed engineers (prep, unknown) Lists what to pin down first: current process step by step, volume and formats, template churn, cost of a wrong answer, who catches errors.
- Common reasons candidates fail forward deployed engineer interviews (prep, unknown) Suggests three to five scoping questions before proposing anything, asking about the current workflow, and defining success in business terms first.
- What the forward deployed engineer interview really tests (prep, unknown) Example questions such as who touches the data today and what breaks if an answer is a day stale instead of a minute.
- Decomposition interview guide (prep, unknown) Clarifying question examples: which metric (average, worst case), where the bottleneck is, what users do with the data, and what shape the sources have.