The five numbers worth watching

Analytics tools present several hundred numbers and a small business can act on about five of them. The problem is almost never insufficient data. It is a dashboard that gets looked at weekly and never changes a decision.

The test for whether a number is worth watching

One question, applied before anything is added to a report: what would I do differently if this number halved. If there is no answer, the number is entertainment.

Applied to the usual suspects, most of them fail. Sessions halving tells you to look at something else. Bounce rate halving tells you nothing, because you cannot tell whether people left satisfied or confused. Average time on page halving could mean the content got worse or the answer got clearer.

The five that pass

• The thing you actually want people to do, counted. Orders, enquiries, bookings, sign-ups. Not a proxy for it.
• How many people arrived, split by how they found you. The split is what makes it actionable: total traffic moving tells you nothing about what to change.
• The rate between the two: of the people who arrived, how many did the thing. This is the number that tells you whether to work on traffic or on the page.
• What it cost to acquire a customer, where you are paying for any of them.
• Whether they come back. The single number that decides whether the business compounds or starts from zero every month.

Five numbers, each of which changes what you do next. Everything else in the tool is there to explain a movement in one of these, which is a different job from being watched.

The numbers that actively mislead

• Pageviews, which rise when navigation is confusing.
• Bounce rate, which cannot distinguish a satisfied reader from a lost one. A page that answers a question completely should have a high bounce rate.
• Time on page, measured by most tools as the gap between loading two pages, so somebody who reads one page and leaves registers as zero.
• Social followers, where the platform decides who sees you regardless of the count.
• Email open rate, now inflated by privacy features that pre-load images.
• Any ranking position reported as a single number, since results differ by person, place and device.

Start from the decision

1. 01Write down the three decisions you expect to make in the next quarter. Whether to keep advertising, whether to change the price, whether to write more of a certain thing.
2. 02For each, write the number that would decide it.
3. 03Check you are collecting that number. Usually one of the three is not being collected at all, and that gap is the whole analytics project.
4. 04Stop reporting everything else. Not delete it: stop looking at it weekly.

That exercise usually produces a report of four or five lines, which looks embarrassingly thin next to a default dashboard and is the version that gets used.

The number that makes the others mean something

What a customer is worth. Without it, acquisition cost is a figure with nothing to compare against, and conversion rate is a percentage with no consequence attached.

The simple version is enough to start: average order value, multiplied by how many times an average customer orders, multiplied by your margin. Illustratively, an average order of 40 at a 40 per cent contribution margin, with customers ordering 1.6 times on average, gives about 26 of contribution per customer. That is the ceiling on what you can pay to acquire one, and everything else is judged against it.

Two warnings about the figure. Use contribution rather than revenue, or you will conclude that you can afford to spend your entire turnover on advertising. And measure the repeat part from actual data rather than assuming it, because the assumed version is reliably optimistic and it is the term the whole calculation is most sensitive to.

If you have been trading for less than a year you cannot know the repeat figure yet. Use one and be conservative, then update it when you can, rather than inventing a number that makes the advertising look affordable.