AFAnalyst Field Notes
Business Metrics

How Data Analysts Break Down Business Problems

When a stakeholder asks why performance changed, your first job is not to pull every chart. It is to break the problem into a measurable structure.

By Analyst Field Notes6 min read

Turn the question into a metric

Vague questions such as "why did performance get worse?" need a clear metric. Define the numerator, denominator, time window, and business owner before analyzing.

Build a decomposition tree

Break the metric into drivers. Revenue can be decomposed into users, conversion rate, order frequency, and average order value. Conversion can be decomposed by funnel step, channel, device, and user segment.

Revenue
  = Active users
  x Conversion rate
  x Average order value
  x Purchase frequency

A decomposition tree is useful because it makes the analysis explainable. Instead of saying "revenue fell," you can say "revenue fell because active users from paid search decreased while conversion and order value stayed stable."

Separate movement from contribution

A segment with the biggest decline is not always the biggest contributor. Contribution depends on both the size of the segment and the size of the change.

For example, a small channel can drop by 40% and still contribute less to the total decline than a large channel that drops by 5%. Always combine change size with business weight.

Use a driver table

Create a driver table with previous value, current value, absolute change, relative change, segment weight, and contribution. This structure makes your work easier to review and reuse.

driver | previous | current | abs_change | rel_change | contribution
paid users | 12000 | 9800 | -2200 | -18.3% | -42%
organic users | 18000 | 17600 | -400 | -2.2% | -8%
conversion rate | 8.0% | 7.4% | -0.6pp | -7.5% | -35%

Write hypotheses before deep dives

After the first decomposition, list possible explanations. Then use data to reject weak hypotheses instead of jumping between dashboards.

A good hypothesis has a mechanism. "Revenue dropped because users changed" is too vague. "Revenue dropped because paid-search traffic shifted toward lower-intent keywords" is testable.

SQL example

This simplified query summarizes revenue drivers by channel. You can adapt the dimensions to category, country, campaign, or customer segment.

SELECT
  acquisition_channel,
  COUNT(DISTINCT user_id) AS buyers,
  COUNT(*) AS orders,
  SUM(order_value) AS revenue,
  1.0 * COUNT(*) / NULLIF(COUNT(DISTINCT user_id), 0) AS orders_per_buyer,
  AVG(order_value) AS average_order_value
FROM orders
WHERE order_date BETWEEN DATE '2026-06-01' AND DATE '2026-06-30'
  AND order_status = 'completed'
GROUP BY 1
ORDER BY revenue DESC;

Deliver an action-oriented answer

A strong analyst answer includes what changed, where it changed, why it likely changed, confidence level, and what the team should do next.

Analyst checklist

  • Define the metric and owner.
  • Break it into drivers.
  • Compare previous and current periods.
  • Calculate contribution, not just movement.
  • Write hypotheses with mechanisms.
  • Recommend one next action and one follow-up analysis.