Requirements Gathering for System Design

How architects turn a vague problem into scoped, testable requirements before drawing a single box, and how to run the first ten minutes of a design interview.

What it is

Requirements gathering is the structured conversation that defines what a system must do, how well it must do it, and within which constraints. In an interview it is the first five to ten minutes; in a real program it is discovery workshops, stakeholder interviews and document reviews. Skipping it is the single most common reason designs fail: architects optimize for the wrong quality attribute.

A useful mental model is to collect five kinds of information:

Kind Questions to ask Example answer
Goals What business outcome does this enable? How is success measured? Reduce checkout abandonment from 70% to 60%
Users & journeys Who uses it? What are the top 3 journeys? Shoppers browse, search, buy; sellers list items
Functional scope What must it do in v1? What is explicitly out? Card + wallet payments; no BNPL in v1
Quality attributes Availability, latency, consistency, security, compliance 99.95% for checkout, p95 < 300 ms, PCI DSS
Constraints Budget, deadline, team skills, existing systems, regulation, cloud Must run on existing AWS org, Java team, 6 months

Real-world example

A retailer asks for "a new order management system". Discovery reveals that the real pain is that orders from three channels (web, app, stores) land in different systems and customer service cannot see them. The actual requirement is a unified order view with a single order lifecycle, not a rewrite of every channel. The architecture becomes an order hub with channel adapters and CDC from legacy systems, which is far cheaper than the rewrite that was initially requested.

Architecture diagram

flowchart LR
  A[Business goals] --> R[Requirements backlog]
  B[User journeys] --> R
  C[Constraints<br/>budget, time, skills, regulation] --> R
  D[Existing landscape<br/>systems, data, contracts] --> R
  R --> F[Functional requirements]
  R --> N[Quality attributes / NFRs]
  F --> S[Scope v1 / out of scope]
  N --> Q[Measurable targets<br/>SLOs, RPO/RTO, budgets]
  S --> H[High-level design]
  Q --> H

A repeatable interview script

  1. Restate the problem in one sentence and confirm it.
  2. Users and scale: who, how many, where (regions), growth.
  3. Top three use cases: design for these; park the rest.
  4. Quality attributes: availability, latency, consistency, durability, security/compliance. Ask which matters most if they conflict.
  5. Constraints: cloud, existing systems, team, budget, timeline.
  6. Write assumptions down visibly and move on. Do not interrogate for 15 minutes.

[!TIP] Interview tip Say: "I'll assume X; if that's wrong the design changes in this way." It shows you know which assumptions drive the architecture.

Advantages and disadvantages of investing in discovery

Advantages Disadvantages / risks
Prevents building the wrong thing Can become analysis paralysis
Makes trade-offs explicit and agreed Stakeholders may not know answers yet
Produces testable acceptance criteria Requirements change; over-specifying early is waste
Surfaces compliance needs early (cheap to fix) Needs access to busy business owners

Technology and technique options

Technique When to use
Event storming Complex domains; finding bounded contexts and events
User story mapping Prioritizing scope into releases
Quality attribute workshop (QAW, from SEI) Eliciting and prioritizing NFRs with scenarios
Context diagram (C4 level 1) Clarifying system boundary and external actors
Domain glossary Avoiding ambiguous terms (e.g. "order" vs "basket")
Assumption log / RAID log Tracking risks, assumptions, issues, dependencies

Architect-level considerations

[!DEVELOPER] Understand acceptance criteria and edge cases for the stories you build; ask what "done" means for performance and security.

[!ARCHITECT] Decide which quality attributes dominate and get explicit sign-off on trade-offs (for example, "eventual consistency on stock display is acceptable, but not on payment"). Turn vague words like "fast" into numbers.

[!EA] Check alignment with capability maps, existing platforms and standards. Can an existing enterprise capability be reused instead of building a new one?

[!CTO] Confirm the business case: what revenue, cost or risk outcome justifies the investment, and what is the smallest release that proves value?

Interview questions and answers

The interviewer says "Design Instagram". What do you do first?

Clarify scope and scale before designing: core journeys (post photo, follow, view feed), daily active users, read/write ratio, media sizes, latency expectations for feed load, and whether features like stories, search or DMs are in scope. State assumptions, for example 500M DAU, 100:1 read/write, p95 feed load under 500 ms, and design for those.

How do you handle conflicting requirements from two stakeholders?

Make the conflict explicit, quantify the impact of each option (cost, time, risk), and take it to the person who owns the business outcome. Record the decision in an ADR. An architect facilitates and informs the decision; they do not silently pick a side.

How do you know you have gathered enough requirements to start designing?

When the top use cases, the dominant quality attributes with numbers, and the hard constraints are known, and the remaining unknowns would not change the high-level structure. Further detail is gathered iteratively.