The post-launch 30 day measurement plan: what to read, what to ignore, and the one decision at the end

Most founders finish launch week with a chart that goes up and no idea what it means. That is the normal outcome, and it is not a measurement failure so much as a structural one: the launch spike is made of attention, and attention is not demand. Nearly everybody who arrives in the first seventy two hours came because somebody they follow mentioned you, not because they woke up with the problem you solve. Reading that traffic as market feedback is the single most common analytical mistake made after a launch, and it leads teams to scale a channel that was never going to repeat, or to abandon a product that was never actually evaluated.

The month after launch has one job. Not to prove the launch worked, which is unanswerable in thirty days, but to produce a decision you can act on: is the weakness in the message, in the product, or in distribution. Those three have completely different fixes, they compete for the same scarce weeks, and picking the wrong one costs a quarter.

What should you measure, and what should you deliberately ignore?

The distinction that matters is not vanity against real. It is whether a number changes a decision. Plenty of numbers are real and still useless, because whatever they say, you would do the same thing next.

Watched but not acted on	Acted on	Why the second one decides something
Total visits	New users per week in weeks three and four	The tail tells you whether anything compounds. The spike tells you who has a big audience.
Signups	Share of signups reaching first value, split by source	Signups measure the persuasiveness of a page. Activation measures whether the product fits the person the page attracted.
Upvotes, likes, shares	Unprompted mentions in places you did not post	Applause is not distribution. Somebody explaining your product to a third party without being asked is.
Average session duration	Day 7 and day 30 return rate for the launch cohort	Coming back is the smallest honest unit of value. Time on site is mostly confusion.
Aggregate conversion rate	Conversion rate of the worst performing source	The average hides your two best and two worst channels. The spread is the strategy.

The rule underneath the table: measure cohorts, not periods. A period metric mixes people who arrived on launch day with people who arrived three weeks later from a search result, and those two groups have almost nothing in common. Once you split by arrival week and by source, numbers that looked flat start saying something specific.

Week one: do you actually trust your instruments?

Spend the first week checking the plumbing rather than reading the output. This feels like a waste of the most exciting week of the year, and it is the difference between a month of findings and a month of confidently wrong conclusions.

Three things break in almost every launch. First, the activation event does not fire, or fires on the wrong action, so activation looks like zero or looks like one hundred per cent. Second, attribution collapses into direct, usually because a link was shared without parameters or because a redirect stripped them, and now half your traffic has no known origin. Third, something double counts: a page view fires twice, or a signup is recorded both client side and server side, and every downstream ratio is quietly wrong by a factor you do not know.

Test each one by hand. Sign up as a new user through each major source link and confirm the whole chain appears correctly. It takes an afternoon. Every subsequent week of the plan depends on it, and instrumentation errors found in week four invalidate everything you concluded in weeks two and three.

The only reading worth doing in week one is triage: is anything actively broken for real users. Errors, failed payments, a step nobody gets past. Fix those immediately and do not treat them as findings, because they are not learning, they are bugs.

Week two: activation by source, and nothing else

Now read one thing properly. Take every account created since launch, group it by where it came from, and calculate the share that reached first value. First value is the moment the product has done for them the thing it exists to do, defined before launch and instrumented, not invented now to make the number look better.

What you are looking for is the spread. It is common to find that one source activates at three or four times the rate of another. That single fact reframes the launch: you do not have an activation problem, you have a traffic quality problem, and the answer is to find more of the good source rather than to redesign onboarding. The reverse pattern, where every source activates at similarly low rates, points at the product or the onboarding, because the weakness follows the user regardless of where they came from.

Two cautions. Sample sizes in month one are small, so treat a gap of a few percentage points as noise and only act on gaps that are large and obvious. And exclude the people you personally know, because friends activate out of politeness and will make a mediocre onboarding look adequate.

Week three: the numbers have stopped talking, so go and ask people

By week three the dashboard has said everything it can. It will tell you that a certain percentage of people stopped at a certain step. It cannot tell you why, and the why is the whole decision.

Contact fifteen people who signed up and did not activate. Not the happy ones, whose feedback is pleasant and useless. Ask what they thought the product did when they signed up, what they were trying to accomplish that day, and what stopped them. Two patterns emerge, and they are easy to tell apart once you hear them.

If people describe a product different from the one you built, that is a message problem: your positioning attracted the wrong expectation, and no amount of onboarding work fixes it. If people describe your product accurately and then say it was too much effort, too confusing, or that they meant to come back, that is a product problem, specifically a time to first value problem. The first is fixed with words and takes days. The second is fixed with engineering and takes weeks. Confusing them is how a quarter disappears.

Fifteen conversations is enough. The patterns repeat by the eighth, and after that you are gathering confirmation rather than information.

Week four: make the decision and write it down

The month exists to produce one choice. Take the evidence from weeks two and three and pick exactly one of these to be the focus of the next thirty days.

1. Fix the message. Chosen when people describe you inaccurately, when activation is uniformly weak across sources, or when the people who love the product found it by a route you did not plan for. The work is positioning and page copy, and it is fast enough that you can rerun this whole plan a month later.
2. Fix time to first value. Chosen when the description people give matches reality and the drop-off clusters at one identifiable step. The work is engineering and design on the path from signing up to the first useful outcome.
3. Scale the one source that worked. Chosen when a channel produced people who activated and returned at a visibly better rate. This is the happiest outcome and also the rarest, and it is worth being sceptical: check the cohort is large enough to mean anything before committing a quarter to it.

Write the decision down in a paragraph with the numbers that led to it and the date. This costs ten minutes and is the most valuable artefact the month produces, because in six weeks you will be arguing about a course correction with no memory of why the original choice was made, and a written record turns that argument into a five minute check.

How do you tell a flat launch from a slow one?

This distinction decides whether to persevere, and both look identical on a chart of total traffic. Look instead at weeks three and four in isolation, after the announcement has finished circulating.

A slow launch has a small, steady flow of new users continuing, arriving from search, from a mention somewhere you did not post, or from word of mouth, and they activate at roughly the same rate as the launch cohort. That is a functioning product with insufficient distribution, and the answer is patience plus deliberate distribution work. It is a genuinely good position, and it feels awful, which is why so many teams abandon it.

A flat launch returns to near zero new users, and the few who arrive do not activate. Nothing was created that generates ongoing discovery, and there is no compounding to be patient about. The answer is not more distribution of the same message; it is going back to positioning and demand evidence, which is upstream of everything a launch can fix.

The visibility number to record before you forget

Add one non-obvious measurement to the end of the month, because it takes forty minutes and becomes far harder to establish later: run the twelve prompt test across the engines your buyers use and record whether you are named at all. Month one is the honest baseline, taken while your launch coverage is at its freshest and most likely to have been picked up. Repeating it quarterly turns it into the only trend line you will have for whether the answer layer has noticed you, and having the first data point already in hand is worth the time now.

Expect to be absent. Almost every product is at day thirty, and it means nothing negative about the launch. Its value is entirely in the comparison you can make in three months, once the third party coverage the launch generated has had time to be indexed and to accumulate.

What this plan deliberately leaves out

No revenue targets, because thirty days of revenue after a launch is a measure of your pricing page and your existing audience rather than of the business. No competitor comparison, because you cannot see their cohorts and the comparison will be made of guesses. No engagement scoring, no health scores, no composite indices: they average away the very spread that makes the data useful.

The month has one output. A decision, written down, with the evidence attached. Everything else in the dashboard can wait for a quarter when it might actually change what you do.