- →Which brand health metrics fit a marketing mix model, and where each one falls short
- →What to use when you have no brand tracking data
- →A checklist to test whether a brand metric works in your model
- →How to smooth brand data and connect it to sales
MASS Analytics treats brand as a core part of a modern marketing mix model. Brand health metrics, such as awareness and consideration scores from consumer surveys, are how you bring it in. A weak metric adds noise, not signal.
Most teams assume the brand metric leads sales. The evidence says otherwise. Professor Koen Pauwels has studied for years whether brand perceptions drive market outcomes or the other way round. In an article with Oliver Koll, he tested 153 FMCG (fast moving consumer goods) brands in five countries, with at least 25 months of data per brand. For 92 brands, penetration and awareness drove each other. Where the link ran one way, it mostly ran from market share to attitudes.
So a brand metric has to earn its place before it goes anywhere near the model. Two questions guide the screen. Does it move when your media moves? And does it move before sales, or after them? Your marketing mix model explains less when it leaves brand out, but it explains less still when the brand metric is the wrong one. Brand data is also messy. Scales vary, and each tracking provider uses its own method. This article shows how to measure brand health in a way the model can use.
Which brand health metrics belong in an MMM?
Brand health metrics come from consumer surveys. They capture attitudes that connect well with marketing activity. A survey can be one question or hundreds of items. For a marketing mix model (MMM), the survey must also run often enough to match your model refresh. The table below lists the brand awareness metrics and brand tracking metrics that teams use most.
| Metric | What it tells you | Weakness | When to use it |
|---|---|---|---|
| Awareness, aided and unaided | How many consumers recognize or recall your brand | Moves slowly and stalls once most buyers know you | New brands, launches and markets with low awareness |
| Consideration | Whether shoppers would put your brand on their shortlist | Scores shift when the survey wording or category definition changes | Categories with a clear shortlist step |
| Preference | Whether consumers choose your brand over rivals | Needs a stable set of competitors to read well | Crowded categories where shoppers compare brands |
| Purchase intent | How likely consumers say they are to buy | Stated intent can run ahead of or behind real buying | Short buying cycles and categories close to purchase |
| Net Promoter Score (NPS) | How willing customers are to recommend you | Often reflects existing customers, so it reacts more to product experience than to media | As a supporting metric, or when it is the only tracker you have |
| Brand equity index | A single score that blends several brand items | Can hide which item moved, and can react slowly to media | When your tracker reports only a composite score |
Table 1: Common brand health metrics for a marketing mix model.
Pick one primary metric
No single metric fits every brand, because your category and the way people buy decide which one works. Start with what you already track, and check it against your media history before you trust it. Pick one primary metric for the model and keep the others as backup. Two metrics that move together add little. They can also compete for the same effect inside the model.
What can you use if you have no brand tracking data?
Tracking surveys cost money, and they do not cover every market. Without a tracker, behavior data can stand in for it. We have found these three proxies useful for brand health.
- Share of search compares how often people search for your brand with how often they search for your rivals. Include your direct rivals in the count, or the share means little.
- Branded search volume counts searches for your brand name. It is mainly a demand signal, closer to intent than to awareness, so read it next to share of search.
- Website traffic shows how many people visit your site. Growth can signal rising interest. It is the weakest proxy, because paid clicks feed it, so remove paid traffic where you can.
Media impressions do not belong on this list. They measure what you put out, not what consumers took in. In a model they sit on the input side, so using them as a brand outcome measures your own spend twice.
Proxies have a catch. They record behavior, and behavior also follows sales and promotions. A promotion can lift branded search, and so can TV.
A telecom example
A major telecom operator shows how this plays out, in a MASS Analytics case study. The brand ran TV, radio, social, display and search at once, and leadership could not say how the channels worked together. TV campaigns created brand interest, and people acted on it by searching for the brand and visiting its website. A model that treated search as independent would have given search the credit for demand that TV had built.
A nested model fixed that. It first explains branded search from media, then uses branded search to explain sales. So it treats branded search as a result of TV and social activity. It also showed that paid branded search during heavy TV flights was capturing traffic that would have converted anyway, so the operator cut that spend during those flights. The nested model found that digital media as a whole generated a 12% sales uplift. A proxy needs the same responsiveness test as a survey metric.
How do you test if a brand metric works in an MMM?
Say you spend heavily on a brand campaign and your metric stays flat. Either the campaign flopped or the metric cannot react. If this repeats campaign after campaign, suspect the metric. Brand KPIs must respond to media investment, or you cannot influence or optimize them.
Hanssens, Pauwels and colleagues proposed four criteria for judging attitude metrics in a 2014 Marketing Science paper: potential, responsiveness, stickiness and sales conversion. We use them here as a plain checklist. The fifth check, on the data, is our own addition. First, plot the raw series next to your media spend.
- Does the metric have room to rise (potential)? A score near its ceiling cannot show a gain.
- Does the metric move when your own media moves, within the time window your model covers (responsiveness)? A metric that never reacts cannot be an optimization lever. It may still work as a control variable, for baseline or competitor effects.
- When the metric rises, does the gain last, or does it fade within weeks (stickiness)?
- When the metric changes, do sales follow, or do sales move first? Test the lead and lag, or run a Granger test, which asks whether one series helps predict the other. Then you do not credit a metric that only follows sales.
- Is the sample large and representative, and does the survey run as often as your model refreshes?
If a metric fails these checks, drop it and try the next candidate.
Here is the whole path. With tracking data, pick one primary metric, run the five checks, and connect it if it passes. Without a tracker, use a proxy such as share of search and run the same five checks.
If you are buying tracking data
If you are buying tracking data, ask the provider for a test sample and run the five checks on it before you sign. Ask how each score is built, how often the survey runs, how much history exists and whether you can have the raw time series rather than a dashboard. You need to know what a one point move means. If the provider cannot explain a score, do not build a model on it.
How do you smooth brand data without hiding the signal?
Survey data is noisy, because small samples and wave to wave changes add random jumps. Smoothing removes part of that noise, and a moving average or a median filter are the two common ways to do it. Our guide to data smoothing methods explains both.
Smoothing can erase the real effect of a campaign, and a window longer than the burst you want to see will flatten it. A trailing average can also delay the timing, so the metric seems to move later than it did.
Smoothing has a second risk. It makes each point depend on its neighbors. That can inflate the model’s fit statistics (t-stats and R squared) and make a weak metric look significant. Check the residuals for autocorrelation, for example with a Durbin Watson test, and use standard errors that are robust to it before you trust the fit. Stretching a quarterly survey to weekly data creates the same problem.
Plot the raw and smoothed lines together. If the campaign bumps vanish from the smoothed line, you have smoothed too much. Run the checklist on both versions and keep the one that responds best to your media.
How do you connect brand tracking to your MMM?
There are two ways to bring brand health metrics into your model.
- Build a standalone model with brand health as the KPI. This works on its own. But a separate sales model with no brand KPI does not ignore brand. It absorbs the brand effect into the base and into whichever media variables move with it, so media that builds brand gets under-credited and the base gets over-credited.
- Use a nested model. Brand activity moves the brand KPI, and the brand KPI then moves sales. It captures the effect of your entire marketing mix, but it takes more work. We recommend it. Our nested modeling guide shows the method.
In a MASS Analytics case study with a European airline, awareness and consideration were modeled as nested stages between media and ticket sales, across ten destinations. The earlier flat model had blended central brand campaigns and local promotions into one effect. The nested model showed that central brand campaigns carried a longer effect than local activity, which paid back faster but faded sooner. The airline protected its brand budget, moved spend out of a saturated channel, and lifted marketing ROI by 17%.
One technical note. Estimate the brand and sales equations together where you can. If you fit them in two stages, carry the first stage error into the second, by bootstrapping or correcting the standard errors. Otherwise the sales equation looks more certain than it is.
Once your brand KPI is in the model, our post on performance marketing vs brand marketing covers what to do with the budget.
How do brand effects show up in model results?
When a brand metric works, a channel’s contribution splits into two parts: a direct effect on sales, and an indirect effect that runs through brand. Read the brand equation first, then the sales equation. The contribution split comes last, because it shows how much of each channel’s effect runs through brand.
In a MASS Analytics case study with a D2C media company, a check against an earlier attribution study revealed an effect the first model had missed. Influencers were lifting branded search. A nested model captured it.
Models built on sales alone tend to under-credit brand building. Its effects build over years, while the model sees a short window. Brand effects show up indirectly, as a rising baseline, stronger seasonal peaks and more resilience when competitors run promotions. A brand KPI in the model gives that effect a place to appear, so it separates from short term effects.
Two things should make you doubt the metric: it moves against your media, or it does not move at all. Fix that before you read anything else in the results.
Is MMM the right tool to measure brand?
MMM suits the question of how brand moves sales, because it puts brand and sales in one chain across your whole mix. Other tools answer other questions. A brand lift study is faster for the short term effect of one campaign. Attribution goes deeper into user level paths. A dedicated brand valuation does more for long run brand value.
Frequently asked questions
Start with the metric before you touch the model. Collect every brand source you have, plot it against media and test it for responsiveness. Smooth the noise lightly, then connect the best metric to sales, ideally in a nested model, so media gets credit for its brand effect.
Yes. Brand tracking gives you a time series, and an MMM works with time series. You line up the survey results with your weekly or monthly media and sales data. Then the brand metric enters the model as a KPI, or as a nested step between media and sales.
A sales only model sees a short window, so it misses effects that build over years. Adstock helps by measuring how long an ad keeps working after it runs, but it covers weeks to months, not years of brand building. A nested model goes further and splits the long term path from short term effects. On the long term path, media moves a brand KPI and the brand KPI moves sales. The total effect of a brand channel is its direct effect on sales plus its indirect effect through the brand KPI. In a MASS Analytics case study with a European airline, central brand campaigns carried a longer effect than local promotions, which paid back faster but faded sooner.
Brand data and brand equity
Track a brand health metric that passes the tests above, such as awareness or consideration. See how it responds to your brand media. Then check whether sales follow through the model. Brand building often shows up in the brand metric first and in sales later. If you have no tracker, use proxies such as share of search.
You can still use it. Model at the cadence you measure, which is often monthly, not weekly. Use a higher frequency proxy such as share of search to fill the gaps between survey waves. Pool waves to steady a small sample. If you must interpolate, keep it light, and check that the filled series still follows real campaign timing.
It means measuring the value a brand adds beyond the product itself. Teams usually blend survey items such as awareness, consideration and preference into one index. Others estimate the financial value of the brand. In an MMM, test an index like any other metric. A blended score can hide which part is moving.
What should you check before your next model refresh?
Before the next refresh, plot your brand metric next to your media spend for your full history. If it does not move after your brand bursts, start with the checklist above. Our Comprehensive MMM Guide covers the full modeling process.
To see how your own brand metrics would work in a model, book a demo with MASS Analytics.
- ✓A brand metric earns a place in your MMM only if it moves when your own media moves.
- ✓Test every candidate on potential, responsiveness, stickiness and its link to sales, and check that it leads sales instead of following them.
- ✓Smooth survey noise lightly, because heavy smoothing hides the campaign effect you want to measure.
- ✓Use proxies such as branded search with care, because they follow sales as well as brand. Media impressions are an input, not a brand outcome.
- ✓Nest the brand KPI in the sales model, so media gets credit for the sales it drives through brand.