Brand tracking: the metrics, the sample size and the free proxies

A brand tracker is not hard to buy. It is hard to read.

Building awareness is a separate job — our brand awareness guide covers it. This page is only about measuring it: the five numbers, how often to take them, how big a sample each needs, and what to use while you have no sample at all.

Start with the thing almost no brand-tracking article mentions — the margin of error on your own numbers.

The five metrics, defined by their question wording

The definitions matter because the wording of the question decides the number. Change the wording and the number moves, with nothing changing in the brand.

Three wording rules keep the series comparable:

• Never change the question once the tracker is running. A reworded question starts a new series, and you lose the comparison that was the whole point.
• Randomise the brand list order on aided questions. A fixed list rewards whoever sits at the top.
• Put one fictitious brand in the aided list. Anyone who "recognises" it shows you how much of your aided number is polite guessing.

Satisfaction and recommendation are different questions, about existing customers — net promoter score is their usual home. A tracker surveys the market, not your customer base.

The same move, twice.

The sample-size problem, worked out

Working through the maths — hypothetical inputs, standard formula for a proportion. Say aided awareness comes back at 30% from a wave of 200 people. The 95% margin of error on that single number is:

margin = 1.96 x square root of ( 0.30 x 0.70 / 200 )
       = 1.96 x 0.0324
       = 0.0635,  so plus or minus 6.4 percentage points

So your awareness sits somewhere between 24% and 36%. That is the honest reading of a 200-person wave.

Now the part that catches people. You do not report one number — you report a change between two waves. A difference carries a wider margin than either number in it, because both ends can be off at once (Pew Research Center, 8 September 2016). For two separate waves of equal size, the margin on the gap is about 1.4 times the margin on one wave:

Hypothetical: awareness near 30%, 95% confidence, simple random sample.

Read the top row again. With 200 people a wave, awareness must move about nine points before you can say anything happened. A three-point move is invisible — it sits inside the noise.

Run the formula the other way and you get the real cost of a readable tracker. Detecting a three-point move reliably needs roughly 3,700 people per wave; a five-point move needs about 1,320. That is why so many trackers get commissioned, reported for three quarters and quietly dropped.

Two things make this worse in practice, and neither is in the formula:

• Sub-groups collapse fast. A 600-person wave split by four cities is 150 per city, with a wider margin on each. If you plan to read cuts, size for the cut — the same trap as in customer segmentation.
• Online panels are not simple random samples. The formula assumes they are, so treat every margin above as the best case. Agree your market research method once, then freeze it.

Cadence: why annual tracking reads nothing

An annual tracker gives you one comparison a year. One comparison, carrying a margin of several points, is not a trend — it is a coin toss you pay for.

Quarterly gives you four points a year, so you can see shape. Continuous — a small sample monthly, pooled on a rolling basis — gives you direction and precision from the same budget. The arithmetic on a fixed annual budget of 3,600 interviews:

Continuous wins because of that last cell. You report the tight rolling average and still get a monthly reading to spot a step change. Seasonality stops being an argument, because every month is in the series.

The cost consequence: at the same total sample, continuous buys no more interviews — it buys more fieldwork set-ups, twelve launches instead of one. Providers price that in, so ask for the per-wave fee, not just the per-interview cost.

The free proxies to start with instead

Most Indian mid-market brands should not buy a tracker yet. Watch four free signals for a year first, so that when you do buy one, you already know what you expect it to say.

Be straight about share of search: a real idea, a real author, a contested evidence base.

Les Binet presented it at the IPA's EffWorks Global conference on 14 October 2020, using free Google Trends data, and reported that share of search correlated with market share in automotive, energy and mobile handsets, leading market share by up to a year in cars (IPA, 14 October 2020). He also said it is not a perfect predictor, because conversion depends on things like price and not all search is positive.

It was argued with quickly. Ken Roberts of Forethought wrote on 4 November 2020 that search share is a leading indicator rather than a cause — people search because intent already exists — and cannot tell you what to change (Forethought, 4 November 2020). He also sells a competing survey-based model.

So: treat it as a cheap, fast directional signal in categories where people search before they buy. Not a law, and not a substitute for asking people questions.

Two caveats on the Trends data, from Google's help page: only a sample of searches is used, and numbers are scaled 0 to 100 against total searches rather than reported as volumes (Google Trends Help). Your brand must also clear the zero threshold, which rules the proxy out for genuinely small brands. The GA4 Direct rule above is Google's own (Google Analytics Help).

Put all four on one screen with your sales line beside them — our marketing dashboard note covers the layout. Social listening adds a fifth, noisier signal.

Brand tracking versus a platform brand lift study

These get confused constantly. Different tools, different questions.

A tracker tells you your altitude. A lift study tells you whether one engine fired. Neither replaces the other — the randomised design is covered in our brand lift study page.

For the revenue side, marketing mix modeling and incrementality testing connect spend to sales. Brand metrics belong on their own line of your marketing KPIs, not mixed into the performance ones.

What we tell a client with no survey budget

DH's own view, from client work: we do not start brand clients on a panel. We start them on a sequence.

The honest version we give clients: brand work takes about 3–6 months to show meaningful movement in organic and proxy signals, then compounds. Paid activity moves fast and stops when the spend stops. Anyone promising a measured awareness jump in a quarter is either buying a very large sample or hiding the margin of error.

If your question is about positioning rather than level, it changes shape — that is brand positioning work, and it needs qualitative input before any number.

A brand number your board can actually act on

The point of brand tracking is not a dashboard tile. It is being able to say, with a stated margin, whether the brand moved and which way.

Three decisions before fieldwork starts settle that: the exact question wording, a sample size matched to the move you care about, and a cadence you will still be running in two years. Get them right and a modest tracker beats an expensive one read badly.