Did the launch work? What I check in the first two weeks
Two weeks of data can’t tell you if a product will last, but it can tell you which channel, message and onboarding step to fix next, if you decide what to look at before launch day.
A launch-day chart almost always goes up. On its own, that spike tells me very little.
When a Web3 or AI product launches, the first wave is mostly curiosity. People click because something is new, because a partner posted about it, or because they hope there might be an airdrop. Most of them try it once and leave. a16z calls these users tourists in its writing on AI retention, and Andrew Chen has written about how novelty effects distort early metrics. On a dashboard it looks like this: visits peak on launch day, far fewer people complete a first transaction or save an output, and fewer still come back later that week.
When I worked on launches at HeimLabs, after every launch someone asked the same question: did it work? Two weeks is too early to say whether a product will last. It is enough to decide which channel to repeat, which message to rewrite and where onboarding is leaking, as long as the tracking was set up before launch.
Before launch: write down what you will count
The night before a launch, I write a short note with three things in it.
First, the one action that means someone got value. For a Web3 app built around transactions, that might be a wallet connected plus a first real transaction. A connected wallet alone doesn’t count. For an AI tool it is the first output someone saves, exports or comes back to. A prompt typed into a box doesn’t count either.
Second, where people will come from. Every link that goes out gets its own UTM tag, per channel and per partner, so a creator’s post and our founder’s post never end up in one big “social” bucket.
Third, a baseline. If there was a waitlist, a beta or an earlier version, I write down what a normal week looked like. Without a baseline it is hard to tell whether a post-launch number is unusual or just what happens on any busy day.
I write this down first because after launch, the number that grew fastest starts to feel like the number that matters. Having it on paper means I can’t quietly swap one for the other.
Days 1 to 3: read the funnel by source
On the first few mornings I ignore totals. I look at each source on its own: visits, sign-ups and the value action, side by side. In Google Analytics I filter by UTM, and for product events I run a simple SQL query that groups first actions by the source tag from each user’s first visit.
What I want to see is where each source loses people. Some lose them between the landing page and sign-up. Others lose them after sign-up, before anyone uses the product. A simple comparison shows why totals mislead. With made-up numbers: Partner A sends 2,000 visits and 12 people complete the value action. Partner B sends 300 visits and 28 complete it. Partner A looks better in the launch-day screenshot. Partner B is the one I brief again and ask for a second post.
Web3 gives you one extra check. If the product runs on a public blockchain, I look at the addresses that took part: when they were created, how they were funded, and whether they did anything after the first transaction. A batch of brand-new wallets, funded from the same place, that all stopped after the minimum action can point to reward hunting. It isn’t proof, so I don’t delete them. I put them in their own group and compare their repeat use with everyone else’s before I report anything.
Days 4 to 10: did anyone come back?
Repeat use isn’t a perfect signal, but for a product people are meant to use often, it says far more than a launch-day visit. So in the second week my main question is simple: of the people who did the value action in the first three days, how many did it again?
I build a small cohort table in Excel. Each row is a signup day. Each column shows the share of those users who repeated the value action within one, three and seven days. With launch-sized numbers this isn’t statistically clean, and I say that when I share it. It can still show an early pattern. Either some users from each day keep coming back, or nearly every group disappears after its first session.
I also read the questions. Replies, Discord threads and support messages count as data. “How do I connect a second wallet?” or “Can it export to CSV?” comes from someone trying to use the product for real. If most of the questions are about rewards or a token, that tells me which line in the launch message pulled in the wrong crowd.
Day 14: decisions instead of a verdict
At the end of week two I write a one-page note for product, engineering and whoever owns the budget. It has four parts:
- Value-action rate per source, with the drop at each step.
- Repeat rate for users from the first three days.
- The five questions people asked most, in their own words.
- One activity to stop and one to put more into, with the evidence for each.
The note never says “the launch was a success”. Two weeks can’t support that, and a16z’s retention writing is a good reminder that the real test comes later, after the tourists have gone. What two weeks can support is a short list of decisions: which partner to brief again, which line of the message to rewrite, which onboarding step is losing the most people.
Those decisions go straight into the next launch plan. If users from the walkthrough post were the ones who returned, I put more time into that format next time. If one onboarding screen loses half the people who reach it, engineering gets the query that shows it, so the conversation starts from a number.
Before your next launch, write down the one action that counts and tag every link by source. By day 14 you will be comparing the results with a definition of value you set before the launch-day spike could sway you.
