Discovery & Requirements
Customer and technical discovery — the most undervalued FDE skill.
Discovery
Discovery is the process of understanding what the customer actually needs before building anything. Customer discovery: who, what pain, what workflow, what constraints. Technical discovery: what systems, what data, what APIs, what permissions, what scale. Most AI project failures are discovery failures — you built the wrong thing.
Why discovery matters
Skipping discovery = building the wrong thing fast. In FDE work, you deploy directly to customers — if you misunderstand the constraint (RBAC, audit, latency, cost), you ship something that gets rejected in security review or fails in production. Discovery is cheap; rebuilding is not.
Discovery process
Stakeholder interviews (business owners, end users, IT/security) → document the workflow (current state) → identify pain points → map constraints (compliance, scale, budget, latency) → map systems (data sources, APIs, identity) → write a requirements doc → validate with customer → sign-off before build.
Discovery question bank
discovery_questions = {
"business": [
"What is the workflow today, step by step?",
"Where does it break or slow down?",
"Who are the users (roles, frequency, peak concurrency)?",
"What does success look like in 3 / 6 / 12 months?",
"What is the cost of the current problem (time, money, risk)?",
"Who are the stakeholders and decision-makers?",
"What is the budget envelope (build + ongoing)?",
"What is the timeline and what drives it?",
],
"technical": [
"What systems hold the relevant data?",
"What APIs are available, and what are their limits?",
"How is identity / access control managed (AD, Okta, custom)?",
"What are the integration patterns (sync, async, batch)?",
"What is the data volume and growth rate?",
"What is the latency / throughput requirement?",
"What is the uptime SLA?",
"What is the deployment environment (cloud, on-prem, hybrid)?",
],
"constraints": [
"What compliance regimes apply (GDPR, HIPAA, SOC2, FedRAMP)?",
"What is the data retention / audit requirement?",
"Can data leave the environment? If so, under what conditions?",
"What is the model output restriction (advice, citations, confidence)?",
"What is the acceptable false-positive / false-negative rate?",
"What is the rollback plan if the AI is wrong?",
"Who is on-call when it breaks?",
],
"success_criteria": [
"How will you measure success (quantitative)?",
"What is the user acceptance test?",
"What is the pilot scope and duration?",
"What would cause the customer to NOT renew?",
],
}
def run_discovery(customer: str) -> dict:
"""Structure the discovery conversation."""
notes = {}
for category, questions in discovery_questions.items():
notes[category] = []
for q in questions:
answer = ask_customer(customer, q)
notes[category].append({"question": q, "answer": answer})
return notes
def requirements_doc(notes: dict) -> str:
"""Turn notes into a validated requirements doc."""
return f"""# Requirements
## Workflow (current state)
{notes['business'][0]['answer']}
## Users
{notes['business'][2]['answer']}
## Constraints
{[n['answer'] for n in notes['constraints']]}
## Success criteria
{[n['answer'] for n in notes['success_criteria']]}
## Sign-off
[Customer to confirm before build]
"""Experiment: discovery questions
See which discovery questions uncover hidden constraints.
What to observe
Surface discovery answers 'what' and 'who' but misses the constraints that make or break an AI deployment: RBAC, audit, cost, latency, compliance. Deep discovery uncovers these BEFORE you build. The cost of deep discovery is hours; the cost of missing constraints is weeks of rework or a failed security review.
Production discovery
Production discovery: structured question bank, written notes, a validated requirements doc signed off by the customer, a stakeholder map (who decides, who is affected, who can block), and a constraints register. Never start building before sign-off.
Challenge
A customer says 'we want an AI assistant over our docs'. Run a 30-minute discovery call that uncovers the constraints that will shape the architecture. (See the FDE Simulator.)
Production checklist
Production checklist
0 of 8 checked
Knowledge check
A customer wants an 'AI assistant over their docs'. What's the FIRST thing to do?
Complete
You can now run effective discovery. Next: solution architecture — designing under the constraints you uncovered.
Mark this chapter as complete
Track your progress and unlock the next chapter.