Gathering Requirements -- Who Uses the App and When

Before you write a single test script, you need to become a detective. Your job is to figure out the real usage patterns of the application. This is not guesswork -- you need actual data. Think of it like a restaurant owner planning for Saturday night. They do not just say "it will be busy." They check reservation counts, historical covers, average party size, peak arrival time, and how long people stay. That is exactly what you need to do for your application.

Where to Find Usage Data

In most organizations, nobody will hand you a neat spreadsheet with all the numbers. You have to go hunting across multiple sources. Here is where to look:

The Key Numbers You Need

Not all data is equally important. Here are the numbers that directly shape your test plan. I call these the "Big Seven" -- if you have these, you can build a realistic workload model.

1. Concurrent users at peak -- Not total registered users, but how many are active at the same time during the busiest hour. A site with 1 million users might only have 2,000 concurrent at peak.
2. Transactions per second (TPS) at peak -- How many requests hit the server per second during the busiest time? This is often 10-50x the concurrent user count because each user generates multiple requests.
3. Average session duration -- How long does a user stay? An e-commerce user might browse for 15 minutes, while a banking user logs in, checks balance, and leaves in 2 minutes.
4. Think time between actions -- The pause between clicks. Real users read pages, fill forms, scroll. This is typically 5-30 seconds. Skipping think time makes your test unrealistically aggressive.
5. Transaction mix -- What percentage of users do what? On an e-commerce site, maybe 70% browse, 20% search, 8% add to cart, 2% purchase.
6. Data volume -- How many records in the database? Products, users, orders, transactions. Performance degrades with data volume, so testing with 100 rows when production has 10 million is meaningless.
7. Growth projection -- If the business expects 3x growth in the next year or a big marketing campaign next month, your test needs to account for that.

Requirements Gathering Template

# Performance Requirements Gathering Template

application:
  name: "Banking Portal"
  type: "Web Application"
  architecture: "Microservices on AWS EKS"
  current_state: "Production (live since Jan 2024)"

user_metrics:
  total_registered_users: 250000
  daily_active_users: 45000
  peak_concurrent_users: 3200
  peak_hour: "9:00 AM - 10:00 AM IST (Monday)"
  average_session_duration_seconds: 180
  average_think_time_seconds: 8

traffic_patterns:
  normal_tps: 120
  peak_tps: 450
  seasonal_peaks:
    - event: "Salary day (1st of month)"
      multiplier: 2.5
    - event: "Tax filing deadline"
      multiplier: 3.0
    - event: "Festival season"
      multiplier: 1.8

growth_projections:
  expected_growth_6_months: "40%"
  expected_growth_1_year: "80%"
  planned_campaigns: "Mobile app launch in Q3"

sla_requirements:
  response_time_p95: "< 2 seconds"
  response_time_p99: "< 5 seconds"
  error_rate: "< 0.1%"
  availability: "99.95%"
  throughput_minimum: "200 TPS sustained"

data_volumes:
  accounts: 250000
  transactions_per_day: 180000
  total_transaction_history: 45000000
  products: 12
  branches: 340

Common Mistakes in Requirements Gathering

• Confusing registered users with concurrent users -- A site with 5 million users might only have 10,000 concurrent. Testing for 5 million is absurd and wasteful.
• Ignoring think time -- Without think time, 100 virtual users generate the same load as 1,000 real users. Your test is 10x more aggressive than reality.
• Forgetting background jobs -- Cron jobs, batch processes, and scheduled reports consume resources too. A nightly batch job running during your test will skew results.
• Not accounting for third-party dependencies -- Payment gateways, email services, SMS providers have their own latency and rate limits.
• Using averages instead of percentiles -- "Average response time is 500ms" hides the fact that 5% of users wait 8 seconds. Always ask for p95 and p99 targets.