Most launch teardowns are written by the agency that ran the launch, which is why they read like victory laps. This one keeps the published numbers and spends most of its length on what we got wrong. A three-week beta campaign for a Series A AI platform produced 8.5M impressions, 3,500 signups, and a 12% conversion to active users. It also produced one structural mistake that we now design around on every engagement.

If you are planning a beta, the useful question is not “how did you get 8.5M impressions?” but “which of those impressions did anything?” Here is the honest accounting.

What we actually ran

The client was a Series A company building a collaborative workspace for ML engineering teams. Their beta worked. Their distribution did not: signups arrived in a thin trickle from founder outreach and the occasional organic mention, and a trickle never accumulates enough early users to learn from.

A funding announcement had already bought them attention. Attention is the easy half. Turning it into people who log in twice is the hard half, which is why the campaign was built as funding announcement PR with a second act, not a single news push.

The shape: a 14-day pre-launch, then 20 creators in three weekly cohorts, each with a different job. Week one was the reveal. Week two was demonstration, with creators given extended hands-on access so they could post screen recordings and workflow walkthroughs instead of paraphrasing the landing page. Week three was social proof: early user quotes, community screenshots, and an X Space with two beta users.

The audience segments and week-by-week breakdown live in the Series A beta release case study. What follows is the reasoning behind the structure, and where it bent.

Why three cohorts beat one spike

A single-window launch works when the goal is a moment: a Product Hunt rank, a funding headline, a first-hour velocity spike. A beta has a different job. It needs a population of users large enough to learn from, and it needs them to still be there in week three.

Concentrating all twenty creators into one afternoon would have produced a taller spike and a worse funnel. Beta adoption moves at the user’s pace: someone sees a post on Monday, reads a technical thread on Thursday, and signs up the following week. A one-day campaign is invisible at two of those three moments.

The cohorts also reached different slices of the audience, which mattered more than reach itself. Cohort one carried the reveal to ML practitioners and adjacent AI-industry accounts. Cohort two spoke to developers who needed to see the product operate before trusting it. Cohort three addressed people who had seen the first two waves and needed a reason to move now. Same product, different argument, different moment.

That is the case for sustained work over a spike, and why we run Sustained Amplification separately from Launch Surge. They solve different problems. A surge for a beta buys you a crowd that leaves.

The first 24 hours still carried the campaign

Here is the uncomfortable part. The first cohort posted on day one, and that day produced 2M of the campaign’s 8.5M impressions. Roughly a quarter of total reach came from one day out of twenty-one, and the following weeks bought repetition and depth rather than raw distribution.

Published results: 8.5M impressions across three weeks, 2M in the first 24 hours, 3,500 beta signups, 12% conversion to active users. Numbers as reported in the case study.

Read that plainly: even a sustained campaign is decided by a spike. The following weeks buy repetition and depth, not raw distribution. Anyone selling three weeks of steady reach as a substitute for a strong first day is selling the wrong shape.

A beta launch still needs a first-hour plan. Velocity on day one decides whether later cohorts inherit an audience that has heard of you or start cold each week. Week two’s demonstrations worked because week one had already put the name in front of the right people.

Where it broke: the late third cohort

We added the third cohort late, after the first two were mapped. On paper it was a sensible extension: more creators, more days of visibility, a stronger finish.

In practice, we recruited them from the same pool we had already mined. Their audiences overlapped heavily with cohort one’s, so a large share of the “new” impressions landed on people who had seen the reveal two weeks earlier. Reach kept climbing. The argument was being repeated to the same room.

The signups were real, and they count toward the 3,500 total. But they came from an audience already primed twice, and the impressions were the most expensive of the campaign because they were largely repetition. We paid for reach we already had.

That is the honest version of “more creators is better.” It is not. Beyond a certain point you buy frequency, not new audience, and frequency does not convert people who already decided.

Lesson one: define the activation metric before launch

The 12% activation rate was discovered in reporting, not designed into the campaign. We tracked signups from day one. Activation — the percentage of signups who became regular active users within 30 days — was measured after the fact.

That sequencing cost us. If we had agreed on an activation target before the first brief went out, the third cohort would have looked different: not more reveal posts for an audience that had already seen the reveal, but the behavior we wanted demonstrated: which workflow, which integration, which first-session moment predicts a return visit.

Signups flatter a campaign. Activation tells you whether it did anything. Decide which one you are optimizing before you brief a single creator, and write the definition down, including the measurement window. Our measurement methodology now requires that agreement before kickoff.

Lesson two: cap creators at the overlap threshold

The overlap threshold is the point where a new creator’s audience is mostly people who have already seen your argument. Past it, additional creators add frequency rather than reach, and frequency is the cheapest thing you can buy.

You can estimate it before you sign anyone. Ask each candidate who follows them, which communities they sit in, which accounts they overlap with, then compare against the creators already on your list. When two candidates share most of their engaged audience, you are looking at one slot, not two.

We now cap cohort size at that threshold, even with budget for more. A smaller network with genuinely distinct audiences beats a larger one that keeps talking to the same room, and it is cheaper. It also protects against the failure mode we hit: a final week that looked busy in the dashboard and thin in the signup data.

Lesson three: plan week two before week one ends

Week two did the quiet work. Seven creators posted across five days, one or two at a time, with screen recordings and workflow demonstrations. Those posts converted the people who had seen the reveal and needed proof. It was also the least improvised week, because those creators had extended product access during the 14-day pre-launch and had already built their content.

Week three was assembled while week one was running, and it shows. Late recruitment meant less preparation time, thinner product familiarity, and content that leaned on social proof because that was quick to produce.

The middle week is where a sustained campaign earns its money. If it is still being planned when the launch goes live, you have already given up the part that converts.

Plan week two before week one ends. Brief those creators first, give them the longest product access, and let them build demonstration content while everyone else is still writing a launch post. The reveal will take care of itself. The middle is what needs defending.

None of this changes the headline result, and we would run the campaign again. Twenty creators, three weeks, 3,500 signups from an audience that had never heard of the product a month earlier is a good outcome. The difference between a good sustained campaign and a great one is entirely in the boring decisions: what you measure, how many voices you need, and when you write the middle week.