What this worksheet is for
Use this structure before selecting an MMM library or writing model code. It turns “we have the data” into a field-by-field record of what exists, what changed, and what can be trusted.
Sheet 1: Decision definition
| Field | Question to answer |
|---|---|
| Business decision | What allocation, planning, or forecasting decision should change because of this model? |
| Outcome | What single primary result will the model explain? |
| Time grain | Daily, weekly, or monthly—and why does that grain fit the buying cycle and data quality? |
| Geography | National, regional, market, or store—and where is there real variation? |
| Decision cadence | How often can the business actually change the budget or plan? |
| Constraints | Which channel floors, ceilings, commitments, or capacity limits must scenarios respect? |
Sheet 2: Variable inventory
Create one row for every outcome, media, control, and context variable.
| Field | Record |
|---|---|
| Canonical name | Stable model-facing variable name |
| Business definition | What the variable means and does not mean |
| Category | Outcome, media, promotion, price, seasonality, distribution, macro, or event |
| Source and owner | System of record and accountable person |
| Unit | Dollars, impressions, orders, rate, index, or another explicit unit |
| Available grain | Native time and geography detail |
| Trustworthy history | Earliest date after known tracking or business-definition breaks |
| Missing-value rule | Whether missing means zero, unknown, delayed, or not applicable |
| Known breaks | Platform migrations, tracking changes, acquisitions, outages, and definition changes |
Sheet 3: Media inputs
For each channel, record spend, exposure measures, buying method, major format shifts, and whether the channel has enough variation to identify a relationship. Flag channels that always move together; the model may not be able to separate their effects reliably.
Also record periods when a channel was intentionally off, geographically limited, supply constrained, or tested. Those periods can be more informative than another year of uniform spending.
Sheet 4: Outcome and control quality
Check whether the outcome is measured consistently across the full window. Document returns, cancellations, offline conversions, delayed revenue, currency treatment, and accounting restatements.
List non-media forces that can move the outcome: price, promotions, availability, sales coverage, holidays, weather, economic conditions, competitor events, and product changes. Include a control because it represents a credible causal alternative—not merely because a column exists.
Sheet 5: Experiments and known events
| Event | Date and market | Expected direction | Measured result | Data source | Useful for calibration? |
|---|---|---|---|---|---|
| Geo holdout | |||||
| Lift study | |||||
| Major promotion | |||||
| Tracking outage | |||||
| Product launch |
Readiness decision
Do not reduce readiness to one score. End with three lists:
- ready: stable definitions, sufficient history, and credible variation;
- usable with limits: understood gaps that can be represented in assumptions or sensitivity tests;
- blocking: missing outcome integrity, unresolved definition changes, no meaningful variation, or confounded channels that the proposed model cannot separate.
For every blocking item, name an owner and a next action. The first useful output of an MMM project may be a better measurement plan rather than a model.