Running a discovery call
Find the business problem, the person who owns it, how success is measured and what stands in the way, before proposing anything.
Forward Deployed and Applied AI roles are scored on discovery as much as on code. OpenAI's FDE post asks you to "lead technical discovery" and to translate customer workflows and constraints into production systems; Anthropic's Applied AI posts ask you to "translate [requirements] into technical solutions" (official, Sources O6, A7).
A 45-minute structure
- Context (5 min): their goals this year, why now, who else is in the room and what each cares about.
- Current workflow (15 min): walk one real case end to end. Who does what, with which systems, how long it takes, where it fails.
- Success (10 min): what number moves if this works? Who reports it? What is good enough to launch?
- Constraints (10 min): data access and classification, security review, residency, integration points, budget, timeline.
- Next step (5 min): agree one concrete next action and owner, and play back what you heard.
Questions that earn their time
- "Can you show me the last time this went wrong?"
- "If this worked perfectly, what would you stop doing?"
- "Who has to say yes before this goes to production, and what will they ask?"
- "What have you already tried, and why did it stop?"
- "What data would we need, and who owns it?"
From your background: enterprise architecture reviews are discovery under another name. Your story is "I have sat on the customer side of the table and know what the security architect will ask". Use a specific example.
Before you leave the call, do you know...
Practice
Written for this site to drill the lesson. Not reported interview questions.
- Role-play: a bank's head of operations says "we want to use AI for KYC". Run the first 15 minutes of discovery with a partner.
- Write the one-page discovery summary you would send after the call.
Check yourself
What should you agree before ending a discovery call?
One concrete next step with an owner and a date, and a playback of what you heard.
Why walk one real case end to end?
It exposes the actual workflow, systems and failure points that a general description hides.
Saved in this browser.