What metrics should I track on my landing page besides conversion rate?

Track visitors, signups, activated users and paid accounts, all split by page version and traffic source. The ratio that judges the page is activated users per 1,000 visitors, because it counts the people who reached real value, not just the people who filled in a form. Conversion rate alone can rise while revenue falls.

My signups doubled after a landing page rewrite but revenue did not move. Why?

The rewrite probably changed who signs up, not just how many. A broader headline persuades curious visitors who never had the problem the product solves, so they sign up and leave. Compare activated users per visitor before and after the rewrite, one traffic source at a time, to see whether the page brought in better users or only more of them.

How do I connect landing page analytics to product activation?

Store the page version and traffic source in a first-party cookie when someone lands. Copy both onto the account when they sign up, and fire your activation event against that same account inside the product. One query can then count visitors, signups and activated users by page version and source.

Is a lower signup rate ever a good sign for a landing page?

Yes, when the drop comes from visitors who were never going to use the product. A headline that names exactly what the product works on turns away poor-fit visitors in seconds. If activated users per visitor rises while signups fall, the page got better at its real job.

Should I A/B test my landing page on signups or on activation?

Test on activated users per visitor whenever the change could alter who signs up, which includes most headline and offer changes. Signup-only tests reward pages that attract curious people. If traffic is too low for a clean test, compare two full weeks of each version inside the same traffic source and count activated users directly.

Landing Page Metrics to Track Beyond Signups

Q: Why do free signups cost more for an AI product than for normal SaaS?

Every free user who runs the product triggers model calls, and that cost lands in cost of goods whether or not they ever pay. AI product builders average around 52 percent gross margin against 70 to 80 percent for traditional software. So a landing page that attracts many signups who never activate costs an AI company more than it would cost a typical SaaS company.

The landing page metric to track is activated users per 1,000 visitors, split by traffic source. Signup rate tells you how many people the page persuaded. Activated users per visitor tells you how many of them the page should have persuaded, and that is the number revenue follows.

This matters most to a founder who has just rewritten his page and watched signups jump. A page can double its signups and still send fewer real customers into the product, because it started persuading curious people instead of the ones with the problem. If you only track the top of the chain, a worse page looks like a win.

The month signups doubled and published translations fell

Picture an AI tool that translates a software company's help center into other languages. A support lead connects the help center, picks Spanish and German, reviews the drafts side by side, and clicks Publish to help center. That click is the moment the product has done its job. Everything before it is setup.

The old hero read Translate your help center into 12 languages in an afternoon. The button said Connect your help center. It was narrow, a little dry, and it only made sense to someone who runs support docs. The numbers in this example are illustrative: 10,000 visitors a month, 210 signups, and 65 of those signups publishing a translated article.

Then the founder rewrote it. The new hero read AI translation for anything. The button became Translate a file free, with a drop zone that took any PDF. The next month the same 10,000 visitors produced 410 signups. Visitor to signup went from 2.1 in every 100 to 4.1. The GA4 tile turned green and the screenshot went into the team's wins channel that afternoon.

Inside the product, a different month was happening. Most of the new signups dropped one PDF, downloaded the result and never came back. They had no help center to connect. Only 53 people clicked Publish to help center, down from 65. The page had doubled its signups and lost a fifth of its real users. Nobody saw it for six weeks, because the product dashboard lived in another tool that nobody on the marketing side opened.

Most landing page advice treats signup rate as the score. Test the headline, test the button color, keep whichever version gets more signups. That advice is fine when every signup is the same kind of person. It breaks the moment a change alters who signs up, and headline changes almost always do.

Here is the part people miss. The rewrite would have been a real win if activation had only slipped a little. At 19 activated signups in every 100, the new page produces 78 activated users against the old 65. The rewrite loses only when activation falls below about 16 in every 100. So neither number can judge the page alone. Only the product of the two can, and that product is activated users per visitor.

We would go further. A landing page that loses signups can be the better page. If a sharper headline turns away the student with one PDF and keeps the support lead with 300 articles, signups fall and revenue rises. A team that scores the page on signups will revert that change within a week.

The four numbers in the chain, and the one that judges the page

Track the chain as four counts, not three percentages. Percentages hide the size of the base, and a founder reading a green tile does not stop to check it.

Visitors, by page version and source. This is the only count most sites already have.
Signups, with the page version and source written onto each new account at the moment it is created.
Activated users, meaning the signups who reached the moment the product does its job. For the translation tool that is a published article, not a translated file. If you have not pinned yours down, how to define activation for an AI product walks through choosing it.
Paid accounts, from the same signups, counted later. This one lags by weeks, which is exactly why you cannot wait for it before judging a page.

The ratio that judges the page is activated users per 1,000 visitors. It folds the page's persuasion and the visitor's fit into one count. Benchmarks for the first ratio alone are in our piece on SaaS website conversion rates. This post is about the number to put next to it.

Wiring the page and the product to the same person

The reason most founders cannot see activated users per visitor is plumbing, not discipline. The website analytics know the visitor. The product analytics know the user. Nothing joins them. The website is measured to the last click, and the moment a user gets lost inside the product produces no event at all.

The join takes an afternoon of an engineer's time. When a visitor lands, store the page version and the source in a first-party cookie. The source is the utm tag if there is one, the referrer if not. When that visitor creates an account, copy both values onto the account as properties, for example signup_page_version and signup_source. When the product fires its activation event, it fires it for that same account. Now one query can count visitors, signups and activated users by page version and source.

Dualite is a useful picture of why the two halves belong together. Its case study covers the website and also the first-run moments where users stalled: the first run, the empty project, build failures. The product spoke to engineers while most signups were non-technical builders. A page and a first run that disagree about who the user is will show up first in exactly this chain.

Split it by source before you trust any change

A page version is not one audience. In the translation example, imagine the new hero got picked up by a directory of free AI tools. That one listing sent 3,000 of the month's visitors. They signed up at a high rate, because a free file translator is exactly what they came for, and almost none of them had a help center to connect.

Blended together, the month looks like a page problem. Split by source, it looks like two stories. Search visitors who typed help center translation behaved the way they always had. The directory visitors filled the signup count and left the activation count alone. The fix for that is not the same as the fix for a weak headline.

So read the chain one source at a time, and only compare page versions inside the same source. A launch week, a newsletter mention or a paid campaign can change your mix more than any headline, and the blended number will credit or blame the page for it.

Every signup who never activates was paid for once already, in ad spend, content or the founder's time. In an AI product they get paid for twice. The student who translated one PDF ran real model calls on your account. That cost sits in your cost of goods whether or not anyone pays you.

The margin makes this sharper than it is in ordinary software. AI product builders average around 52 percent gross margin, against 70 to 80 percent for traditional software. A page that doubles free usage from people who will never pay is spending against a thinner margin than the SaaS playbook it came from assumed.

The upside runs the same way. On product-led SaaS benchmarks, activation sits around 20 to 40 percent for most products, and a ten point improvement typically drives a 15 to 25 percent increase in free-to-paid conversion. Those are SaaS numbers, not AI ones, but the direction holds. A page that brings in fewer, better-fitting signups is often the cheapest way to move activation, because it changes who arrives instead of what they see.

When signups rise and activation falls: what to do this week

You do not need a new tool or a redesign to run this. You need one table and one rule.

Write your activation event in one line, as an action a user takes. Published a translated article, not created a project.
Add the page version and source to every new account. If that will take more than a few days, tag accounts by signup date and match them to the page version that was live then.
Build the table for your last two page versions: visitors, signups, activated users and paid accounts, one row per source.
Compute activated users per 1,000 visitors for each row. Circle any source where signups rose and that number fell.
For those rows, read the first screen after signup next to the hero that sent them. Checking the promise against the first screen is usually where the gap shows. A hero that says anything leads into a product that only does one thing.
Put the qualifying words back into the headline. Name the thing the product works on, help center articles, so the wrong visitor can tell in a few seconds that this is not for them.
If you want to keep the broad traffic, add one question at signup with two answers, My help center and A single file, and send each group to a different first screen.
Judge the next change on one rule. It wins only if activated users per 1,000 visitors rises inside the same sources, over at least two full weeks, whatever happens to signups.

If your traffic is small, count the activated users themselves rather than the ratio, and expect to wait. A slow honest answer beats a fast green tile.

The page your team cannot read as a stranger

The founder in this example did not write a bad headline out of carelessness. He knows exactly what the product is for, so AI translation for anything read to him as a bigger version of the same promise. To a stranger it was a different product. The landing page is where his knowledge of the product meets someone with none of it, and the chain above is how you see what that stranger actually did next.

Studio Maydit designs websites and products for AI founders in the US, UK and Europe, in Framer, Webflow and custom code, and we keep going into the product when the leak is past the signup. Our fixed-scope projects end with a diagnosis of what is leaking in the product, so a page that wins signups and loses users gets caught before the next rewrite. If you are comparing studios first, our ranking of landing page design agencies for AI startups sets out who does what. If your signups went up and your activated users did not, book a 30-minute call with Studio Maydit.