Instrument the decision, not the page

I instrumented this site properly last week, and the interesting part wasn’t the code. It was how long it took me to admit that eighteen months of pageview data had never once changed a decision.

That’s the normal state of analytics on a personal site, and on a lot of products. You install something, you get a dashboard, you look at it sometimes, and it produces a feeling rather than an action. The feeling is usually mild anxiety.

The fix is not more tracking. It’s narrower tracking, chosen backwards.

Start from the decision

The question that fixes analytics is: what would I do differently depending on the answer?

Pageviews fail this immediately. Suppose traffic is up thirty percent. What changes? Nothing — I’d write the next essay either way. Suppose it’s down thirty percent. Same. It’s a number that describes weather, and I was checking it the way you check weather in a country you don’t live in.

Now try: which essay causes someone to book a call? If the answer is “the ones about how I build with agents”, I write more of those and put the offer link higher in them. If it’s “the ones about the Romanian market”, that’s a different site and a different plan. Either answer changes next month’s work. That’s a metric worth collecting.

Same test kills most of what people track. Time on page — what would you change? Bounce rate — on an essay, what does leaving even mean? Someone read the thing and left satisfied; that’s the product working.

The three questions worth instrumenting

For anything that exists to produce leads, sales, or signups, almost everything reduces to three:

Which entry point produces the outcome? Not which page is most visited — which page is most often the last thing someone reads before they act. Those are rarely the same page, and the difference is where all the leverage is.

Where does the funnel leak? The step where people stop. You only find it if you instrument steps rather than pages.

Which channel brings people who act? Not which brings the most people. A channel with a tenth of the traffic and ten times the intent is the better channel, and raw traffic numbers will tell you the opposite every time.

Three questions. On this site that’s about six events. Everything else I deliberately didn’t collect.

Collecting less is a feature, not a compromise

The version I shipped is roughly a kilobyte of inline JavaScript with no cookies, no local storage, and no persistent identifier. Events go out with the URL that produced them and nothing else that could identify anyone.

I could have installed a full product-analytics SDK, gotten session replay and cross-visit funnels, and paid for it with fifty kilobytes of JavaScript and a consent banner. For a site whose entire pitch is performance and not making readers click through a privacy dialogue, that’s a bad trade.

But the more interesting reason to collect less: you don’t need identity to answer the three questions. The click event already carries the page it happened on. That single property answers “which essay causes bookings” without knowing anything about the person who clicked.

I’d assumed for years that funnels required stitching sessions together. For most questions I actually have, they don’t. The identity was there to support analysis I was never going to do.

There’s a general pattern here, well beyond analytics. Ask what the minimum data is that answers your question, and it’s routinely less than what the standard tool collects by default. The default is built for the largest possible customer, not for you.

The dashboard is not the point

One more habit worth breaking: the daily check.

A metric you look at every day but act on every quarter is not instrumentation, it’s a nervous tic. Worse, daily numbers on a small site are almost entirely noise, and staring at noise trains you to see stories in it.

What I do instead: the events accumulate, and I look once a month, with a specific question written down before I open anything. Which three posts produced the most booking clicks? Did the Romanian section produce any at all? Then I decide something and close the tab.

The number’s job is to settle an argument I’m already having with myself. If I’m not having the argument, I don’t need the number.

The uncomfortable part

Doing this honestly means accepting that some things you’ve been doing have no measurable effect, and you’ll keep doing some of them anyway.

I write essays that will never produce a lead. I’ll keep writing them, because the site is a notebook as much as a funnel, and a portfolio that only contains commercially optimal work is a worse portfolio.

The point of instrumenting the decisions isn’t to make everything justify itself commercially. It’s to know which things are doing which job, so that when you cut, you cut the thing that isn’t working rather than the thing you were least attached to.

That’s the whole argument. Measure what changes a decision. Don’t collect the rest. Look monthly, with a question in hand.