# MMM decision memo

_A template for a short memo presenting a marketing mix model result as a decision. Fill in every field; delete the grey hints. From the Vexpower MMM library (vexpower.com/mmm/communicating-mmm-results), generalised from the Northstar case in the Vexpower marketing mix modeling course._

Author: ______________   Review date: ______________

## 1. Decision, period and owner

Lead with the action. A reader who stops after this section should know what you want to do, for how long and who decides.

**I recommend:** ______________________________________________
_e.g. increase paid social by £30k next quarter as a bounded test, holding search at its agreed minimum_

**Period:** ______________________________________________
_e.g. 1 January to 31 March, plus a six-week follow-up for carryover_

**Decision owner and decision date:** ______________________________________________

| Channel | Current spend | Proposed spend | Change | Agreed min / max |
| --- | --- | --- | --- | --- |
|  |  |  |  |  |
|  |  |  |  |  |
|  |  |  |  |  |
|  |  |  |  |  |

**Constraint check (total budget, minimums, delivery capacity):** ______________________________________________
_All constraints are management decisions, not model outputs_

## 2. What supports this action?

Keep these four categories separate. Mixing them is the most common way a memo overstates its case.

**Observation:** ______________________________________________
_A fact visible in the data or a checked record, not a causal claim_

**Model-based calculation:** ______________________________________________
_A result conditional on named assumptions, with units_

**Assumption:** ______________________________________________
_Something the recommendation needs that the evidence does not establish_

**Independent evidence:** ______________________________________________
_An experiment, test or outside source not already used to fit or calibrate the model. If none exists, say so_

## 3. Economics and range

**Outcome definition:** ______________________________________________
_Numerator and denominator in full: e.g. incremental net revenue (after returns) per £1 of media, UK, weekly, 13-week window_

**Central result:** ______________________________________________
_e.g. contribution after advertising changes by +£4.9k versus the current plan_

**Range:** ______________________________________________
_Low / high, and what kind of range it is: credible interval, range across specifications, or stress scenarios (not probabilities)_

**Breakeven:** ______________________________________________
_The marginal return at which the change stops paying for itself (1 ÷ contribution margin for revenue ROI)_

**Margin recheck:** ______________________________________________
_Result at finance's alternative margin_

**Costs excluded:** ______________________________________________
_e.g. fixed overhead, agency fees, tax, financing. Contribution after advertising is not accounting profit_

## 4. Conditions

**One operational condition:** ______________________________________________
_The practical limit that must hold for the plan to work: delivery capacity, stock, creative supply, minimum presence_

**Causal assumptions that matter:** ______________________________________________
_e.g. channel effects separated cleanly; response curve holds at new spend level; no major price change_

## 5. What would change the recommendation

**The strongest reason not to take this recommendation:** ______________________________________________

**I would change course if:** ______________________________________________
_A specific, observable threshold, e.g. the test's lift estimate implies marginal ROI below 2.5_

| Changed assumption | Why it is defensible | Resulting recommendation | Evidence that would resolve it |
| --- | --- | --- | --- |
|  |  |  |  |
|  |  |  |  |
|  |  |  |  |

## 6. Next measurement action

**What we will run:** ______________________________________________
_e.g. geo holdout, conversion lift study, spend pulse_

**Comparison and outcome:** ______________________________________________
_Treatment vs counterfactual, population, outcome metric_

**Minimum decision-relevant effect:** ______________________________________________
_The smallest effect that would change the decision, and a power or feasibility check before choosing sample size_

**Owner, start date, read-out date:** ______________________________________________

**Review point:** ______________________________________________
_When we will revisit this recommendation_

## 7. Evidence trail

**Data issues found and repaired:** ______________________________________________
_Source of each correction, and what remains unknown_

**Model and files used:** ______________________________________________
_Model version, specification, date range, notebooks or workbook rows_

---

## Review rubric

Score each dimension 0 (missing or misleading), 1 (present but incomplete) or 2 (specific, correct and challengeable). A high total does not cancel a fundamental error.

| Dimension | A strong memo earns 2 when it… | Score |
| --- | --- | --- |
| Decision and feasibility | Names the period, owner, complete allocation and every operating constraint. | |
| Data stewardship | Records the input issues found, uses source-backed corrections, and distinguishes detected errors from unknown omissions. | |
| Evidence categories | Separates an observation, a conditional calculation, an assumption and actual or proposed independent evidence. | |
| Economics | Uses contribution margin correctly, subtracts advertising once, compares plans under the same assumptions, states units and excluded costs. | |
| Uncertainty | Shows the range and any sign change, rechecks margin, and does not label scenarios as probabilities or a credible interval. | |
| Actionable learning | Proposes a relevant comparison with an outcome, owner, review point and decision threshold; considers power and validity separately. | |

Revise any claim that calls prediction causal proof, treats scenarios as a calibrated interval, double-subtracts advertising or breaks a budget constraint before presenting.

## A review conversation

1. Ask your reviewer to explain the recommendation back to you without seeing the spreadsheet.
2. Ask which assumption matters most to the action, not just which number is largest.
3. Ask for the strongest defensible counterargument.
4. Revise one sentence to make the action or its reversal condition clearer.

_Invite a colleague to state what would make them change their mind, and build that condition into the next review._