What is actually driving the number?

What is actually driving the number?

A performance number is an outcome, not an explanation. When revenue, margin, churn or productivity moves, the first discipline is to separate arithmetic contribution from causal influence. Without that distinction, sophisticated analytics can produce a persuasive story that management cannot safely act on.

Begin with an identity the business recognises. Revenue may decompose into volume, price, mix and currency; margin into revenue quality, input cost, labour, yield and allocation. Quantify each contribution at a consistent grain, then segment by customer, product, channel, geography and cohort. This shows where the movement resides before asking why it occurred.

Next test the mechanism. Timing, correlation and feature importance are clues, not proof that changing a variable will change the outcome. Machine learning is powerful for prediction; causal inference addresses interventions. Use controlled experiments where possible, and credible natural experiments, matched comparisons or sensitivity analysis where randomisation is not feasible.

Triangulate model results with process evidence. A price effect should be visible in transaction histories, customer behaviour and commercial decisions; an operational driver should connect to capacity, queues or defects. Search deliberately for rival explanations, data-definition changes and external shocks. The objective is not one perfect model but a chain of evidence strong enough to support action.

Finally, express the conclusion as a decision: which lever, for which population, over what horizon, with what expected range and guardrail. Track whether the intervention produces the predicted response. Analytics becomes managerial when it moves from describing variance to learning which controllable mechanism actually changes performance.