Activation, year over year: taking day-zero engagement from 25% to 60% on a paid base

CAC tripled. The funnel had to do more work, not less.

Through 2025 SpeakX's paid acquisition costs roughly tripled. The same ad rupee was buying a colder, less qualified user. The PM team's read was unambiguous: we couldn't keep paying more for users who never came back after Day 0. The conversation moved from "how do we acquire" to "how do we make the user we just paid for actually stick".

Activation became the focal metric. Specifically: M0 Active, the percentage of paid users who actually started a lesson on the day they paid. It was the cleanest leading indicator we had for retention, renewal, and unit economics, and in Feb 2025 it sat at 25.5%. Three out of four people we'd just paid to acquire were paying us and then leaving without ever using the product.

Users were paying, then drifting. The first lesson was where the funnel went quiet.

The pre-redesign onboarding was a generic "sign up → land on a catalogue" path. Users picked from a long list of lessons with no signal about why any of them were right for them. Even the ones who started a lesson dropped off mid-way: M0 minutes spent averaged 36 in Feb 2025, which on a 20-minute first lesson meant most users barely cleared one session.

The data made three things obvious. The catalogue was overwhelming for a Day-0 user. The system had no idea why anyone had paid (was it for a job interview, a school exam, just for fun?), so its recommendations were generic. And there was no return loop: nothing pulled the user back on Day 1, so M0 minutes had to do all the retention work alone.

Twelve months, two platforms, no big-bang releases.

• No single-shot redesign. The team was shipping weekly across activation, content, and pricing. The activation flow had to evolve in small increments that each held their own. A big-bang launch would have collided with everyone else's work.
• Android-first, iOS shadow. Most experiments ran on Android (larger volume, faster build cycle), then ported to iOS once the variant won. The design had to translate cleanly across both without diverging.
• Cohort metric, not A/B. Most of the win is observable as a cohort trend across the year, not as a single controlled experiment. We had per-step A/B data for individual changes, but the headline number is a year-on-year comparison, which means I have to be careful about what design alone can claim.
• Quality, not vanity. M0 minutes is gameable: a long, padded first lesson can inflate it without producing real engagement. We had to keep watching renewal at M1 to make sure the activated users were sticky, not just front-loaded.

Users didn't want lessons. They wanted progress on a reason they already had.

We started running an open-ended "why did you sign up?" prompt in onboarding and clustered the answers. The clusters were sharp: job_interview, higher_education, workplace_communication, just_for_fun, travel. Users weren't fuzzy about their reason. They had it on Day 0, before they paid. The product just wasn't asking.

I sat in on [X] onboarding sessions. The pattern across all of them: users would land on the catalogue, hesitate, scan a few titles, and quietly close the app. None of the lessons were wrong. They just weren't recognisably theirs. A user signed up because of an upcoming interview; the home screen showed her "Daily Conversation Practice, Episode 14".

The reframe. Stop treating onboarding as account creation. Treat it as the first conversation: capture the reason, then make the very first lesson recognisably about that reason, then end it with a milestone the user wants to come back and continue. Activation and retention are the same designed moment, just split across two days.

Capture the reason, recommend on the reason, end on a milestone.

What landed across the year, in the order it shipped:

• Intent capture in onboarding. A single screen, one question, five tappable reasons. No skip. The output became a first-class user attribute the rest of the product could read.
• Intent-aware first-lesson recommendation. Each reason mapped to a curated short first lesson, 8–12 minutes, recognisably on-topic ("Answering the 'Tell me about yourself' question" for job_interview, not "Daily Conversation Practice").
• Milestone reveal at lesson end. A streak start, a visible "1 of 12 in your interview prep path", and an explicit next-lesson preview. The user finishes Day 0 with a reason to open the app on Day 1.
• Recommendation refinements. Quarter by quarter, the data analyst and I tuned which lessons paired with which intent tag based on completion and Day-1 return.

The honest framing. CAC tripling, content depth, pricing changes, and a stronger team all helped. Design's specific contribution was the funnel-side response: building the activation flow that converted whatever traffic the new CAC was buying into actually-active users. I claim the funnel mechanics and the intent taxonomy. I don't claim the full 25.5 → 60.7 by myself.

Twelve months of small wins, not one big launch.

What shipped, when. Q1 2025: intent capture screen v1 (three options, no recommendations on the back-end yet, just collecting). Q2: intent-aware first-lesson mapping for job_interview and higher_education, the two largest clusters. Q3: milestone reveal at lesson end, streak system, Day-1 push tied to the user's intent. Q4: extended intent taxonomy to five categories, refined the Q2 mappings based on completion data, ported all of it to iOS.

What got cut. A more elaborate "learning plan" reveal at the end of Day 0: too much to read after a first lesson, dropped completion. A second intent-capture moment on Day 7: felt like re-onboarding, killed in usability testing.

The Green Cohort. Somewhere mid-year, the data analyst started slicing on "users who hit M0 active and one Day-1 return". That cohort sat at ~80% M0 active versus ~50–60% for the base, with a +3–6 point lift on M1 renewal. It became the segment the team optimised against, a measurable definition of "engaged paid user" that came directly out of this work.

What I'd do differently. What I'll carry forward.

Where I have to be honest. Twelve months is long enough that a lot of things changed at once: content depth, pricing experiments, ad creative, the team itself. A clean A/B counterfactual for the full 25.5 → 60.7 doesn't exist. What I can defend is the per-step lift on intent capture, the completion lift on intent-mapped first lessons, and the Day-1 return lift on milestone reveal. Stitched together, those are the funnel-side response to a tripling CAC. The rest is product maturity, and I want to name that rather than dress it up.

What I'd do differently. Ship intent capture and the recommendation back-end together, not three months apart. Collecting intent for a quarter without using it felt responsible ("let's get the data first") but it cost us a quarter of compounding learning on which mappings actually worked.

What I'll carry forward. Activation and retention aren't separate problems. The Day-0 lesson and the Day-1 return are one designed moment with a sleep in the middle. If you treat them as one surface, you can move both with the same intervention. If you treat them as two, you'll over-invest in one and watch the other quietly leak.