Brand Tracking: brand health monitoring system

What Brand Health Tracking Actually Is

Brand Health Tracking (BHT) is not a one-time study but a repeated measurement of the same questions on comparable samples at regular intervals. The key word is 'comparable': if the question wording, recruitment channel, or sample size changes between waves, comparing the numbers is meaningless because the difference will reflect methodological error, not market movement.

BHT is usually approached in two ways. The first is an intuitive feeling that the brand is weakening: sales to new customers are slow, and a competitor with a clearly weaker product is capturing market share. The second is that money has already been invested in brand media, and it is necessary to determine whether it delivers an effect that is simply not visible in standard ad attribution with a window of 30-90 days. In both cases, a one-time survey does not solve the problem — a system with historical data across several waves is needed.

Which Metrics Actually Predict Sales

There is no universal set of KPIs: spontaneous brand awareness is critical for FMCG with a frequent purchase cycle, but almost meaningless for B2B with a deal cycle of 6-9 months. A working approach is to distribute metrics across three levels, each requiring its own actions when deviation occurs.

Inside this model, Brand Funnel operates — not absolute figures for each level, but conversion between them. An example from practice: spontaneous brand awareness is 70%, but only 38% make it to consideration. This is not a reach problem — the brand is known well enough. This is a positioning failure: the audience knows the brand but does not consider it a relevant solution to their task. If the awareness itself had dropped (say, from 70% to 45%), the diagnosis would be the opposite — do not touch the message, increase media weight.

Share of Search is a separate metric — the share of brand search queries relative to the category in Google Trends. Its difference from classic survey metrics is that it does not depend on the sample and precedes changes in market share by 1-2 quarters, so it is convenient to use as a quick indicator between waves of full BHT. A metric that fluctuates from wave to wave without connection to revenue or market share over the last 2-3 measurements — is ballast in the report, not a KPI, and it should be abandoned rather than continue tracking it 'just in case'.

How to prove to the CFO that the brand affects money

The main reason why brand metrics fail to hold up in a conversation with the CFO is that they are compared to the performance advertising attribution window. Classic attribution calculates the effect over 30-90 days and almost always shows zero return from brand campaigns. Marketing Mix Modeling (MMM) on historical data over 2-3 years gives a different picture: the influence of brand investments on sales stretches over 6-18 months, and it is at this horizon that the link between the growth of consideration and revenue growth becomes visible through 2-3 quarters.

The working combination — three tools together, not separately:

• BHT — long-term trend in perceptual metrics, wave to wave.
• Brand Lift (built-in tools of YouTube and Meta) — a point measurement of the increase in awareness from a specific campaign, before/after.
• MMM — proof that the growth of Tier 1-2 metrics truly translates into revenue over 6-18 months, not just a coincidence with seasonality.

Without MMM tracking, it remains a presentation for the marketing department. With MMM, it becomes an argument in a budget debate with the performance team, because it shows money, not NPS points.

Frequency and sample size

The frequency of measurements depends on the dynamics of the category, not the budget of the research department: fintech and e-commerce with high purchase frequency require monthly or quarterly tracking, stable B2B and industrial categories — 1-2 times per year. Separately, outside the regular schedule, a measurement before and after each major campaign is needed — otherwise, the effect of the campaign cannot be distinguished from seasonal market noise.

The sample size is specific: 500 respondents per wave provide a confidence interval of approximately ±4,4% at p=0,5 and confidence 95% — this is sufficient to see only major shifts. 1000 respondents reduce the margin of error to ±3,1%, and this is the minimum if comparing waves to each other is planned. If subgroup analysis is needed — by region, age, users and non-users — at least 100 people per subgroup are required separately, otherwise the difference of 2-3 points between waves will be statistical noise mistakenly taken for a trend and used to change strategy.

Tools: from quarterly waves to continuous monitoring

The classic model of ‘wave once per quarter’ is outperformed by continuous data collection. Kantar BrandDynamics builds its model through the Meaningfully Different, System1 Test Your Brand framework, which evaluates emotional response on a Spike scale from 1 to 5 stars and links it to long-term profit growth. Qualtrics and Nepa provide daily collection instead of discrete waves and automatic anomaly detection via algorithms — without waiting for the next quarter when the problem has already damaged sales.

The practical value of these platforms is not in continuous collection itself, but in integration with CRM and MA systems: perceptual metrics (what the audience thinks of the brand) can be directly compared with transactional data (what they actually buy). This intersection is usually missing in companies that conduct tracking 'just for the sake of it' — marketing looks at its dashboard, sales at their CRM, and figures never meet in any report.

Errors that devalue tracking entirely

Most BHT failures are not in the research methodology, but in the discipline of conducting it from wave to wave.

1. Changing question formulations between waves. Even editing a single word in a question breaks comparability — the trend you will see after this reflects the questionnaire edition, not the market.
2. Mixing recruitment channels. Panel, social networks, and email databases provide different respondent profiles by age, income, and brand loyalty — comparing waves collected by different methods is inherently incorrect.
3. Overloaded questionnaire. More than 60 questions — respondents get tired by the end and answer formally, data quality drops precisely on the subtle metrics needed for decision-making. The optimum is 10-20 minutes to complete, 25-35 questions.
4. Ignoring the confidence interval. A difference of 2-3 points between quarters in a sample of 500 people may fall within the margin of error ±4,4% — and yet it is taken as a trend and the strategy is adjusted to statistical noise.

Each error individually looks like a minor issue. Together, they produce exactly the presentation that fails the first question from the CFO: the numbers look good, there is growth, but there are no connections to revenue — because those were never checked.