What the framework is
Most marketing reports are strong on numbers and weak on words. A table shows a figure, a chart shows a line, and each reader supplies their own explanation of what it means and what should happen next. The framework is a discipline for the words beside the numbers: every section answers three questions in order, and the answers are written before the report goes out rather than improvised in the meeting.
It applies at every scale. A single line on a KPIA metric tied directly to an objective., a channel section, the summary at the top of a monthly pack, an experiment readout and a short weekly note about an anomaly all benefit from the same three questions. The scope is the commentary itself. When reports are produced, who reviews them and how the pack is laid out are set by the marketing reporting cadence playbook, which already asks every section to say what happened, why and what will be done. How the charts are built belongs to the dashboard design guide. This page is about writing each of the three parts well and testing them before the report is sent. The same three questions also shape a spoken opening, as the guide to presenting your work to senior leaders shows for a single test or campaign put before the leadership team.
The three questions
| Part | What it answers | Test | Common failure |
|---|---|---|---|
| What | What happened, measured against the target, the plan or a fair comparison | Could a reader check it against the data and agree? | Restating the chart, giving a change without its base, or reporting a figure with no comparison at all |
| So what | Why it matters for the objective, and what most likely caused it, with confidence stated | Does it connect to an outcome the business cares about, and does it say how sure the team is? | A confident story the data cannot support, or a reflexive explanation such as seasonality |
| Now what | What will change, who owns it and when it will be checked, or a decision to change nothing | Could someone be held to it at the next review? | Vague intentions such as "monitor closely", actions with no owner, or an action for every metric |
The three parts depend on each other. A now-what built on a wrong so-what fixes the wrong thing. A so-what built on a vague what explains a movement nobody can verify. Getting the first part precise is what makes the other two possible.
Writing commentary with it
Writing the what. Lead with the comparison the plan set: against target, against the previous period or against the same period last year, whichever the plan uses for that measure. Use the metric as the plan defines it, under the same rule and source, so the commentary and the table agree. Keep it to a sentence that states direction and size; the table carries the detail. If a metric has no comparison and no link to an objective, it is probably one of the Vanity MetricsMetrics that look impressive but do not inform a decision or track a business outcome. that fill reports without informing decisions, and it may not need commentary at all.
Writing the so what. This part has two halves. The first is significance: why the movement matters, usually by connecting it to the outcome above it, such as revenue, pipeline or new customers. The second is cause: the best available explanation of why it happened. Before writing a cause, rule out the dull alternatives: a tracking change, a shift in AttributionAssigning conversion credit to touchpoints. rules or platform reporting, a one-off event, a change in the mix of traffic or products, or ordinary variation within the metric's normal range. Then state the confidence: confirmed, likely or unexplained, with the evidence named. A movement inside the normal range should say so plainly, because "no meaningful change" is a finding.
Writing the now what. Each now-what is an action, an owner and a date when the effect will be checked, or an explicit decision to continue the plan unchanged. Keep it proportionate: a missed target needs a response, while a small wobble inside the normal range does not. Where the cause is uncertain, the most useful action is often a test of the explanation rather than a fix built on it. Then close the loop: the next report opens by stating what happened to last period's actions.
Testing before it goes out. Three quick checks catch most problems. Read only the now-what lines: do they add up to a coherent plan, with owners and dates? Check each so-what against the data: is the evidence actually there, and is the confidence honest? And ask a colleague to read only the what and the so-what and predict the now-what; if they cannot, the reasoning has a gap.
The other reporting guides show where the structure lands. The monthly marketing report template gives each results row a short note, and the three questions are a way to write it. The guide to performance reporting covers presenting channel results fairly and reconciling platform figures, which is what a trustworthy what depends on. And the guide to social media reporting ends every report with decisions, which is the now-what by another name. When the report is a research study rather than a performance pack, the guide to reporting research findings uses the same three questions to turn each finding into a recommendation with an owner.
Worked example (illustrative)
A furniture retailer rewrites its paid social commentary
Copperlark Furniture, an invented online furniture retailer, used to report paid social in one line: "Paid social revenue £38,200, down 22%." The rewritten section reads as follows. What: paid social revenue was £38,200 against a target of £48,000 and £49,000 the month before, a 22% fall, on flat spend of £12,500. So what: paid social brings in most new sofa customers, so the miss explains about half of the shortfall in new customers this month. The likely cause is creative fatigue: the same 2 sofa ads have run for 4 months, click-through has fallen from 1.4% to 0.8%, and average frequency has risen to 6 views per person a month, while site conversion and checkout are unchanged and no tracking changes were made. Confidence: likely, not confirmed. Now what: the paid social manager will launch 2 new sofa concepts by 10 June and cap the old ads, and the team will review click-through and revenue per pound spent on 1 July. If revenue has not recovered by then, the next step is to examine audience targeting.
Limits and criticisms
The so-what step invites confident causal stories the data cannot support. Reports reward explanation, so every movement attracts a cause, and the cause offered is usually the one that fits the team's existing view. A change that coincided with a campaign was not necessarily caused by it; a fall that followed a creative change may have been driven by something else that changed the same week. Stating confidence and naming evidence reduce the risk but do not remove it.
Not every metric needs a now-what. A structure that asks for an action invites one, and a report where every metric produces an action is a report where actions mean little. Many figures belong in a table or an appendix with no commentary, and "no change" is often the right decision for the rest.
It can turn a report into a list of actions nobody owns. Each section generates a now-what, the list grows, and without owners, dates and a review in the next report, the actions quietly disappear. Fewer actions, each owned and checked, are worth more than a long list.
It does not replace testing the explanation. Commentary is a hypothesis about what happened. Where a decision of any size rests on the so-what, the explanation should be tested, with an experiment, a holdout or an analysis designed to rule out the alternatives, before money moves on the strength of it.
Used mechanically, the three parts can also become a formula that readers learn to skim. Treat them as a drafting discipline rather than a template to fill.
Related frameworks
The KPI tree supports the so-what by tracing a movement in an outcome down to the drivers that produced it, which narrows the possible causes before anyone writes one. Control charts answer the question that should come before any explanation: is this movement outside the metric's normal range, or is it noise? When the so-what reveals a recurring problem, five whys root-cause analysis digs beneath the first plausible cause.
Common mistakes
- Restating the chart. A what that repeats the figure in words adds nothing; give the comparison and the size.
- No comparison. A number with no target or baseline cannot be judged good or bad.
- Seasonality by default. Blaming the season without checking last year's pattern is a guess.
- Stating a cause as fact. Without a confidence level, a plausible story reads as a finding.
- "Continue to monitor." A now-what with no action, owner or date is not a decision.
- An action for every metric. Reserve now-whats for movements that matter.
- Never closing the loop. Actions not reported on next time teach readers to ignore the section.
