Benchmarking

A number about yourself, on its own, says nothing about whether you are doing well. Knowing that 3 percent of the people who land on your signup screen finish signing up produces no decision. Is 3 percent embarrassing for this category, or already strong? Without an answer, nobody can say whether the screen is worth fixing.

So meetings stalled in the same spot every time. Someone would report that conversion rose 0.2 points from last month, someone else would ask whether that made it good enough, and from there the loudest opinion in the room decided. Comparing yourself only against your own past makes standing still look like steady progress. The fact that a competitor doubled over the same months never enters that comparison.

Benchmarking plants a ruler outside the building. You find out what the strong performers score first, then measure the distance between that score and yours, and suddenly the same number reads clearly as ahead or behind. It helps to notice that it runs in the opposite direction from its parent activity, competitive analysis. Competitive analysis compares in order to find what rivals are not doing. Benchmarking takes what rivals do well and measures how far short you fall. If a competitor's signup takes two screens and yours takes six, that is not a gap in the market to exploit — it is a place where you are behind.

Run it one item at a time, in five steps. Measuring several items at once means the conditions drift between them and none of the resulting numbers holds up.

1. Pick a single item to measure. "Product competitiveness" is not an item. Choose things any two people would measure the same way: screens to finish signup, hours to a first support reply, documents allowed on the free plan.
2. Choose whose level becomes the reference. Not automatically the category leader. Find whoever is best at that one item. For signup, a service in a different industry that gets you in within three minutes beats another document tool that is also slow.
3. Measure both sides the same way. Reading your own number out of internal data while eyeballing the competitor's breaks the comparison. Sign up for both yourself and count screens, or send a support request to both and time the reply.
4. Write the gap down as a number. Not "much worse," but "six screens against two, a gap of four." Only a number lets you check six months later whether it narrowed.
5. Find out what causes the gap. This is the step where you actually learn something. Whether a rival finishes in two screens because of social login or because it collects company details later decides what you change.

Finish by attaching a target and a date — cut the four-screen gap to two this quarter, for instance — so the next measurement judges itself. Benchmarking that stops before this step is sightseeing.

It fits wherever good answers already exist in public. Signup flows, the layout of a pricing page, how fast search results appear, how quickly support replies — the shape of "done well" is roughly the same at every company. There is no reward for originality in those places, and catching up to a known reference beats inventing your own version.

It does not fit when you are deciding what the product should be. Following a reference line takes you exactly where someone else already stands. If everyone measures everyone, five products converge into something customers cannot tell apart. What you should do differently is a question for user research, which gets its evidence from customers instead of competitors.

Numbers also fail to travel when the conditions behind them differ. A ten-minute support reply at a 2,000-person company exists because there is a night shift. Pagedeck, at 45 people, cannot adopt that as a target without either burning people out or hitting the number while gutting the quality of the answers. When you import a reference level, import the conditions that produce it too.

Copying the surface is the other common failure. Cut your signup from six screens to two without checking why the competitor needs only two, and the details you stopped collecting come back as questions in week one, along with more churn than you started with.

"I don't know whether a 3 percent signup rate is bad, so I benchmarked five comparable products first."

There was no reference line to judge against, so the outside numbers came first.

"This isn't something to benchmark. It's something we have to decide."

A warning that following a reference here would erase the thing that makes the product distinct.

"Their ten-minute reply time comes from having a night shift. It's hard to use as our reference."

The conditions differ enough that the number cannot be lifted across as a target.

At Pagedeck, Mia runs one of these per quarter on a single item. Last quarter it was signup. She and Noa signed up for four competing products and counted screens and input fields: Pagedeck came in at six screens and eleven fields, the shortest rival at two and four. Once that table hit the meeting, a conversation that used to end at "yeah, ours is a bit long" turned into an argument about why four extra screens exist.

The cause turned out to be company name and team size, collected at signup for the benefit of a sales report and irrelevant to whether the product worked. Theo moved those two screens to after the first document was created, and signup landed at three screens and five fields. The benchmark did not supply the answer. It pointed a number at the place where an answer was missing.

The most common reading is that benchmarking means copying whoever is in first place. Since the reference comes from a stronger company, the assumption is understandable. But the work is measuring distance, not imitating. That is exactly why step five asks for the cause behind the rival's number. Skip the cause and transplant the result, and you gain one more screen that does not work in your context. A competitor that gives away unlimited documents on its free plan may be able to afford it; lifting that policy into a different revenue model just produces losses.

Treating the measuring itself as the deliverable is the other trap. A 40-page study of five competitors takes two weeks, so it feels like real work. If that document contains no line saying what will change, by when, and by how much, nobody opens the file again after a month. The output is not a report — it is one gap and a date for closing it.