How to benchmark your brand against a competitor

1. Pick the competitor a buyer actually compares you to

Not the biggest company in the category, and not the one that annoys you most. The one that appears in the same shortlist when somebody is deciding.

One is enough to start. A benchmark against a single well-chosen rival produces a usable list of gaps in an afternoon; a benchmark against seven produces a spreadsheet that nobody reads and a set of averages that describe no actual competitor.

2. Choose the measures before you see who wins them

This is the step that decides whether the result is evidence or decoration. Written down first, a measure is a test. Chosen afterwards, it is a way of describing a conclusion you already reached, and with enough measures available something always favours you.

Keep the list to what would change a decision: entry price and the shape of the range, review volume and rating, publishing and ad cadence, catalogue depth, traffic scale and channel mix, presence in AI answers. Anything you would not act on in either direction is taking up room.

3. Measure both sides the same way, in the same week

Here is the failure that produces the most confident wrong answers in this whole field.

If you read your own numbers from your analytics and theirs from a modelled estimate, the comparison is between two different instruments and the difference you find is partly the instruments. If you count their products from a category page and yours from the admin, you are comparing a sample to a census. If their reading is from Tuesday and yours is from three weeks ago, the gap includes three weeks of drift.

The rule we hold ourselves to is that a difference between two sides is only reportable when both sides were collected the same way. Where that is not possible, the honest move is to say which side is estimated rather than to publish a difference that is really a difference in method. A benchmark that skips this measures your own collection and calls it their performance.

4. Label what is measured and what is modelled

Some of these numbers are facts about the world: their entry price, their review count, the number of pages they have indexed, what their site says today. Some are models: traffic estimates, revenue bands, ad spend.

Both are useful and they must never look alike on the page. Mark every modelled figure as an estimate wherever it appears, and keep it out of anything that gets summed or scored. A single unlabelled estimate inside a total is enough to make the whole total unarguable, and the reader cannot tell which number did it.

For anything you cannot know from outside, publish a band with its assumptions rather than a confident figure, for the reasons set out here.

5. Date every row, then repeat it

A benchmark without dates rots silently. Somebody quotes a figure from March in a meeting in September, it sounds plausible, and nobody can tell because it looks exactly like a fresh one.

Repeating it is what turns the exercise from a position into a line, and the line is where the value is. One reading tells you where you stand. Twenty readings tell you whether the gap is closing, which is the only version of this that can settle an argument about whether the last quarter worked.

What to do with the gaps

Sort them by width against cost and take the wide cheap ones first. Resist the urge to close every gap: some of them are decisions your competitor made that would be wrong for you, and a benchmark that turns into a to-do list of parity items is how a company becomes a worse version of a rival.

And when the gaps run out of interest, the subject is usually the problem rather than the method. That is the argument in the note on watching the wrong companies: a rival tells you where you stand, and somebody growing faster than both of you tells you what to do.

How to do it here

Name a domain and both sides get collected the same way, in the same run, with estimates labelled and every figure clickable through to the page it came from. The reference configuration carries 176 structural measures, so the gaps are found by counting rather than by whoever spoke last in the meeting. The first one is free, and the reasons to do it before you change anything are the subject of their own note.