# Case: fix the onboarding funnel

> Find where new Halves users drop off and what to fix first.

- Course: [Product Analytics: From Data to Decisions](https://productovi.com/courses/product-analytics.md)
- Module: Where do new users get stuck?
- Lesson 12/32, ~42 min
- Source: https://productovi.com/courses/product-analytics/onboarding-funnel

## One sprint, one fix

**Alex Tutor:** Planning time on the Halves product team: Maya needs numbers from you before the onboarding sprint.

**Maya:** The onboarding team has one sprint in September. I need to know where new users drop off and what they should fix first. With numbers, please.

**Alex Tutor:** Then we build the funnel properly. Steps: sign-up → group created → member invited → invite accepted → first expense. The last one is where activation starts: a second member in the group and something to split.

**Alex Tutor:** The rules: only self-signups enter, since invitees skip the first steps. Only complete cohorts, sign-ups from 6 July to 23 August. A step counts if it happened on days 0–6 and every earlier step was reached.

**Dataset (CSV):** [Halves new users, 6 July – 30 August 2026](https://productovi.com/datasets/product-analytics/onboarding-funnel/halves-new-users-jul-aug.csv)

192 sign-ups, data pulled on 31 August. For the funnel: `channel` other than `invite`, `signup_date` up to 2026-08-23.

- `user_id`: User id
- `signup_date`: Sign-up date (UTC); day 0
- `platform`: ios or android
- `channel`: invite (joined someone's group), app_store_search or paid_social
- `group_created_day`: Day after sign-up the user created their first group; invitees never do
- `member_invited_day`: Day the user sent their first invite
- `invite_accepted_day`: Day someone first accepted the user's invite
- `first_expense_day`: Day of the user's first expense
- `expenses_week1`: Expenses the user added on days 0–6; empty if week 1 isn't over
- `active_week4`: Opened the app on days 21–27 (true/false); filled only for sign-ups 6 July – 2 August

**Sofia:** I get 119 users entering, then 100 with a group, 71 who invited someone, 37 whose invite was accepted and 29 with a first expense.

**Alex Tutor:** Now turn the counts into step conversions: each step divided by the step before it. That's where a drop becomes visible.

![A chain of water tanks joined by pipes; one joint in the middle leaks much more than the rest.](https://productovi.com/courses/product-analytics/inline/onboarding-funnel-illustration-2.webp)

*Dividing each step by the one before shows where most people are lost.*

> **Exercise:** Halves's onboarding funnel for self-signups (sign-up → group created → member invited → invite accepted → first expense, each within 7 days) gives 119 → 100 → 71 → 37 → 29. From these counts, what is the step conversion from **member invited** to **invite accepted**? In %, one decimal.

**Alex Tutor:** And the number Maya will ask for first: how many of the 119 made it all the way?

> **Exercise:** Halves's onboarding funnel for self-signups (sign-up → group created → member invited → invite accepted → first expense) gives 119 → 100 → 71 → 37 → 29. From these counts, what is the **overall** conversion from sign-up to first expense? In %, one decimal.

## Split before you decide

**Alex Tutor:** Step conversions: group 84.0%, invite 71.0%, accepted 52.1%, first expense 78.4%. Acceptance is the weakest step. Before blaming it, split the funnel by platform.

**Sofia:** iOS: 61, 50, 34, 26, 19. Android: 58, 50, 37, 11, 10.

**Maya:** Everything looks similar until the invite. Then Android falls off a cliff?

> **Exercise:** Halves's onboarding funnel (sign-up → group created → member invited → invite accepted → first expense), **Android** self-signups only: 58 → 50 → 37 → 11 → 10. What is Android's invited → accepted step conversion? In %, one decimal.

**Alex Tutor:** 29.7% on Android against 76.5% on iOS. The friends who get the invite are the same kind of people; what differs is the inviter's phone. That points to how the invite link behaves on Android.

![iOS and Android self-signups move alike until the invite; then only 11 of 37 Android invites are accepted (29.7%) against 26 of 34 on iOS (76.5%).](https://productovi.com/courses/product-analytics/diagrams/m3-l4-funnel-by-platform.png)

![Two phones send envelopes to friends; from one they arrive and get opened, from the other most get stuck in a tangled pipe.](https://productovi.com/courses/product-analytics/inline/onboarding-funnel-illustration-3.webp)

*Same kind of friends, different invite path: the drop belongs to one platform.*

**Theo:** Or my paid social users are the problem. I've heard that before.

**Alex Tutor:** Check it the same way. By channel, acceptance is 48.7% for store search and 56.3% for paid social. Paid users aren't the drop here.

**Alex Tutor:** Maya is ready to write “the Android invite link is broken” in the memo. Can she?

> **Exercise:** **True or false:** because Android's invite acceptance is 29.7% against 76.5% on iOS, the funnel split proves the Android invite link is broken.

## Signal or noise?

**Sofia:** 11 of 37 against 26 of 34 are small numbers, though. Could it be chance?

**Alex Tutor:** Put an interval on each. With small counts use the Wilson interval, not the normal one. 95% Wilson: iOS 60.0–87.6%, Android 17.5–45.8%. How to compute and compare them is in our [conversion rate tests](/courses/statistics-for-pm-and-marketers/conversion-rate-tests) lesson.

**Maya:** So what do those two ranges tell me?

> **Exercise:** 95% Wilson intervals for invite acceptance: iOS 60.0–87.6%, Android 17.5–45.8%. What's the right reading?

## How big is the prize?

**Alex Tutor:** Last question before a recommendation: what would fixing it give? Suppose Android invites were accepted at iOS's rate. How many more Android users would have reached a second member in these seven weeks?

**Sofia:** 37 Android users sent invites. At iOS's rate, 26 of 34, far more of them would have been accepted than the 11 we got.

**Theo:** And compare that with making group creation a bit better. That's the one my old agency always wanted to redesign.

> **Exercise:** In Halves's onboarding funnel, 37 Android self-signups invited someone and 11 of those invites were accepted; on iOS, 26 of 34 were. If Android invites were accepted at iOS's rate, how many **more** Android users would have reached “invite accepted” in these seven weeks? One decimal.

**Alex Tutor:** Now Theo's alternative, sized the same way: improve group creation and follow the extra groups down the funnel.

> **Exercise:** If group creation rose from 84% to **90%** of the 119 self-signups, and every later step kept its current conversion (71.0%, 52.1%, 78.4%), how many **more** users would reach a first expense? One decimal.

**Alex Tutor:** About 17 more users past acceptance, most of whom add a first expense. Raising group creation from 84% to 90% would add about 7 groups at the top, and after three more steps only about 2 would reach a first expense.

**Alex Tutor:** Two sprint plans are on Maya's desk. Pick the one the numbers support.

> **Exercise:** Which plan should the onboarding team take into September's sprint?

![A plumber seals a large gushing leak in one pipe, while beside it another worker polishes a tap that only drips.](https://productovi.com/courses/product-analytics/inline/onboarding-funnel-illustration-4.webp)

*Fix first where the change adds the most users, not where it looks nicest.*

## The recommendation

**Theo:** Before Maya's memo, give me something for my team. They're convinced paid social brings the wrong people.

**Alex Tutor:** Answer with the channel split, step by step, and say what it doesn't cover.

**Sofia:** Store search: 67, 57, 39, 19, 16. Paid social: 52, 43, 32, 18, 13.

> **Exercise:** Theo's team believes paid social brings the wrong users. Reply in 3–4 sentences using the funnel split by channel (store search 67 → 57 → 39 → 19 → 16; paid social 52 → 43 → 32 → 18 → 13), and say what this split doesn't cover.

**Maya:** Now the memo for me: where we lose people, what to fix first, how sure we are and how we'll know the fix worked.

**Alex Tutor:** And keep the funnel's rules in it. Whoever reads it in October should be able to recount every number.

**Sofia:** Rules, the drop, the split, the intervals, the prize, and a way to check the fix.

> **Exercise:** Write Maya's memo: where new users drop off and what the onboarding team should fix first.

> **Key ideas**
>
> - Fix the funnel's rules first: who enters (self-signups), complete cohorts only, a 7-day window, each step after the earlier ones.
> - Read step conversions, not only the overall rate: 84.0% → 71.0% → 52.1% → 78.4%; 24.4% overall.
> - Split the weakest step by platform and channel. Android invite acceptance 29.7% (Wilson 17.5–45.8%) vs iOS 76.5% (60.0–87.6%); channels don't differ much.
> - Size the prize before choosing: +17.3 users past acceptance at iOS's rate vs about 2 more first expenses from better group creation.
> - The split points to a cause (the Android invite link) but doesn't prove it: confirm with the engineers, then check the fix on the next complete cohorts or with a test.

**Alex Tutor:** The biggest drop isn't always where the fix is. The split told us whose drop it was.

- Previous lesson: [Activation: finding the aha moment](https://productovi.com/courses/product-analytics/activation.md)
- Next lesson: [Retention: day N, rolling or bracket](https://productovi.com/courses/product-analytics/retention-types.md)

---
Productovi Learning. Free, no sign-up needed to read.