MEASUREMENT / MMM

Find the signal.
Fund the response.

Understand the contribution of marketing alongside seasonality, pricing, product and the other forces that shape performance.

THE QUESTION BEHIND THE MODEL

What would change
if the mix changed?

Marketing mix modelling uses variation in business outcomes, media and relevant controls to estimate how activity relates to performance over time. Carryover and saturation represent delayed effects and diminishing returns.

We start with the business outcome and the decision horizon. A revenue model, a contribution model and an acquisition model answer different questions. Their definitions should be explicit before the result reaches the budget meeting.

MEDIA INVESTMENT →INCREMENTAL RESPONSESEARCHVIDEOSOCIAL

FROM INPUTS TO A DECISION

Four steps.
No mystery in the middle.

01

Define & reconcile

Agree the KPI, scope, time grain and financial definitions. Check missing periods and tracking changes.

02

Model & challenge

Inspect contribution, fit, carryover and saturation. Compare plausible specifications.

03

Compare scenarios

Test the budget against constraints, response assumptions and contribution economics.

04

Validate & refresh

Use experiments and subsequent outcomes to challenge the model and improve the next decision.

THE ENGINE / 2 MINUTES 10 SECONDS

See the workbench
in motion.

The recorded retail demonstration shows variable setup, model diagnostics, response curves, optimisation inputs and saved-model comparison.

Read the visual walkthrough

00:00–00:15: introduction to the next marketing-dollar decision. 00:15–00:30: project setup and variable groups. 00:30–01:00: model diagnostics, contributions and channel settings. 01:00–01:20: response curves and quarter comparisons. 01:20–01:40: optimisation inputs, saved models and comparison. 01:40–02:10: decision use and future campaign-level extensions. This is a summary of the screen sequence, not an audio transcript.

Confidence comes
from challenge.

A strong model fit does not prove that every channel estimate is correct. We inspect out-of-time performance, residuals, channel correlation, parameter stability and business plausibility.

Recommendations should carry their assumptions. Where the historical data cannot separate effects, the next step may be a controlled experiment rather than a more elaborate model.

Read: average versus marginal ROAS

THE NEXT DECISION

Bring a question.
Leave with a plan.

Let’s define the brief