EY-Parthenon

A forecast is only as good as the assumption about what shape its history has. Fit a straight line to something compounding, or a compounding curve to something that has already saturated, and the model will still produce a number — a precise one, with decimal places, carrying no warning that its premise was wrong.

That failure is invisible in the output. It only shows up later, in a decision that was made on it.

We built the models in Excel and Python on historical financial and growth data, and started each one by diagnosing the growth pattern rather than by choosing a technique. What the data was doing decided what was fitted to it, in that order.

Then we tested rather than trusted. Quantitative modelling backed by statistical hypothesis testing — t-tests and regression — so an apparent relationship had to survive being challenged before it was allowed into a forecast a client would act on.

Forecast reliability improved 25%.

The models supported client initiatives driving 20–40% revenue uplift. That range is wide because it spans several engagements, and the models were one input into each of them rather than the cause of the result.

The problem was never a shortage of measurement. It was that nothing said which metrics answered which question, so every meeting spent its first half re-deciding what to look at and its second half deciding on the basis of whichever number was raised most forcefully.

My co-lead and I structured measurement frameworks across more than fifteen business KPIs — ROI, conversion, churn, MRR — with each one tied to the decision it was there to inform.

Then we monitored them. A framework delivered and left alone decays quietly: definitions drift, one team starts counting something slightly differently, and within a quarter two dashboards disagree and nobody can say which is right. Monitoring is what keeps a framework worth consulting.

Stakeholder decision-making accelerated 30%.

The speed came from removing an argument, not from adding a report. Once each KPI had a definition and a threshold attached, the meeting could start at the decision instead of arriving at it.