Seasonal businesses face a measurement catch-22. You have only a few months of revenue data each year, but MMM needs multi-year history to work reliably. The solution is not to give up on measurement: it is to build it correctly from the start.
A garden centre, a ski resort, a fireworks retailer, or a sunscreen brand each concentrates the vast majority of its revenue into a short window. For some businesses, 70 percent of annual revenue arrives in eight weeks. That creates a measurement challenge. Most marketing happens before and during the peak, and connecting that spending to peak-period sales is genuinely hard. The standard approaches fall over badly in this context, and teams often end up either not measuring at all or drawing the wrong conclusions from incomplete data.
Why standard measurement approaches fail seasonal businesses
Digital attribution tools are built around the assumption of a steady conversion flow. Last-click attribution needs enough conversions throughout the year to estimate channel performance. In a seasonal business, there are no conversions to attribute for six months of the year. ROAS calculations during peak look artificially high because volume is high regardless. ROAS during off-peak looks artificially low because media investment is building for the upcoming season, not converting immediately.
Year-on-year comparisons are also unreliable for seasonal businesses. If Easter falls in March one year and April the next, or if a heat wave arrives three weeks early, those external factors can shift peak timing by weeks. Comparing week 14 of this year to week 14 of last year is misleading if the buying peak has shifted by two weeks.
What measurement approaches work
- MMM with multi-year data: three or more years of weekly data allows the model to learn seasonal patterns and separate them from media effects reliably
- Controlled geo experiments: test a higher media investment level in one region versus a control region during the peak, then compare outcomes
- Pre-peak brand tracking: survey your target customers before the season to measure awareness, consideration, and intent as a leading indicator
- Cohort analysis by acquisition timing: track customers acquired in pre-peak periods separately from peak-period customers to understand which media drives early buyers
- Booking and reservation data: for businesses where customers reserve in advance, use booking data as an early signal of marketing effectiveness weeks before peak revenue arrives
Building the right MMM for a seasonal business
A seasonal MMM requires careful treatment of the off-peak periods. When sales are near zero for several months, the model needs to understand why. Including explicit off-season indicator variables prevents the model from trying to explain zero sales through media variables and generating nonsense coefficients. You should also include weather variables if weather is a demand driver, because weather is often the primary reason sales start earlier or later than average in a given year.
Three years of data is the minimum for a seasonal MMM. With three years, the model sees three complete seasonal cycles and can learn what is seasonal pattern (consistent across all three years) versus what is media-driven variation (correlated with spend changes). With less than three years, the model cannot reliably separate these effects, and the results will be unreliable.
Keep your historical spend data clean and consistent. If your agency changed how it reported media spend three years ago, align the old and new data before you start modelling. Inconsistent historical data creates model errors that are hard to diagnose.
The pre-peak investment question
The most common planning question for seasonal businesses is: how much should we spend before the peak to prime demand, versus spending during the peak to convert it? MMM answers this directly by estimating the adstock decay of each channel. A channel with a long adstock decay (TV, for example) retains its effect for many weeks after the campaign runs. A channel with a short adstock decay (paid search) converts in the same week. This means TV investment in the six weeks before peak can be just as effective at driving peak sales as the same TV spend running during peak, at lower cost because peak CPMs (cost per thousand impressions) are typically much higher.
Using off-peak data and media efficiently
Seasonal businesses often over-concentrate media into the peak period when costs are highest. MMM can show the effect of spreading investment across a longer window. A garden centre that front-loads TV into February and March (before peak garden demand in April and May) can build awareness at lower media cost and then use paid search and social during the peak to capture the demand that TV created. This is a smarter allocation than running all channels simultaneously in April when CPMs are elevated and the audience is already in buying mode.
Planning for the next season
Seasonal businesses should build their MMM refreshes on an annual cadence, ideally immediately after the peak so that fresh data is incorporated before the next season's planning begins. Use the model outputs to optimise the channel mix and timing for the following year. Then test any major changes (a new channel, a different pre-peak timing, a higher or lower total budget) in a controlled way, ideally in a geo test, so you are adding experimental evidence alongside the model evidence.
We have only two years of data and our peak just ended. Can we run a model now?
You can, but two years gives you only two seasonal cycles for the model to learn from. The results will have wider uncertainty ranges than a three-year model. Run it with the two years of data and use the outputs as directional guidance. Flag the uncertainty clearly in any planning decisions, and prioritise getting to three years of consistent data before making any large budget changes based on the model.
Weather drove an unusually strong peak last year. How does that affect our model?
Include weather as a variable. For most seasonal categories, temperature and rainfall data are available at weekly granularity from national weather services. If last year was unusually warm or sunny, the model needs to know so it does not attribute the weather-driven sales uplift to whatever media was running. Without a weather variable, the model will overestimate media effectiveness in a good weather year and underestimate it in a bad one.
We want to extend our season. Can MMM help us understand how to do that?
Yes. MMM can show you whether media investment in the shoulder periods (just before and after your traditional peak) generates meaningful sales, and at what cost per sale. If the marginal return on shoulder-period media is positive at your target threshold, you have a data-driven case for investing there. Many seasonal businesses find that shoulder-period media is more efficient than peak-period media because competition for media inventory is lower and the audience who buys early tends to be highly engaged.