Alitycs
Start free

// customers

Three teams, three questions they could not answer

Product analytics earns its keep on specific questions: which onboarding step is really losing people, which early behaviour predicts retention, and what happens in the first four minutes of a mobile app. Here is what that work looks like end to end.

Start free Read the stories

These case studies are illustrative examples. Northwind, Peakline and Driftwork are invented companies, and every figure on this page is made up to show the shape of the analysis — not a real customer, a real result or a real quote. Quotes are written by us and attributed to a role, never to a person. When we publish a real customer story it will say so and it will be named.

// the companies in these examples

NorthwindLoomlyPeaklineCobaltHalcyonDriftwork

// worked examples

Situation, instrumentation, finding, change

Each story follows the same four beats, because that is the order the work actually happens in. The interesting part is rarely the dashboard; it is which event someone decided to send.

NorthwindB2B SaaS · 40 people

Onboarding looked fine in aggregate and was broken in the middle

Illustrative example

Situation

Northwind sells a reporting tool to operations teams. Signups were healthy and paid conversion was flat, and the weekly dashboard everyone looked at only showed the two ends of the journey: accounts created, and accounts on a plan. Nobody could say where the middle went wrong, so every fix was a guess.

What they instrumented

Six events across setup, named for what the user did rather than which screen they were on: account_created, workspace_named, data_source_connected, first_query_run, report_saved, report_shared. Each carries the source type and whether the person arrived by invite.

What they found

Connection was not the wall. Eighty-one per cent of accounts connected a source, and then a third of them never ran a query — they sat on an empty workspace waiting for something to happen. The drop concentrated in accounts where the connector took longer than about ninety seconds to backfill, because the interface said nothing while it worked.

What changed

They shipped a progress state with a row count and a sample of the incoming records, then pre-built one query per source type so the workspace was never empty. The funnel step they had assumed was fine turned out to hold the largest recoverable loss on the whole path.

Setup completion63%↗ 22 pts vs. last week
Time to first report9h↘ from 3d vs. last week
Setup tickets−38%↘ per 100 signups vs. last week

Setup funnel, before

Illustrative figures. The connect-to-query step lost 265 of 810 accounts — more than every other step combined.

We had been redesigning the signup form for a quarter. The step that was actually losing people was the one we thought had already worked.

Head of product, B2B SaaS · illustrative

PeaklineTwo-sided marketplace

The activation event they had chosen was a coincidence

Illustrative example

Situation

Peakline matches independent tradespeople with local jobs. Activation was defined as 'profile completed', because it was the first thing new sellers did and it was easy to measure. Growth spent two quarters optimising profile completion, and retention did not move.

What they instrumented

Every seller-side action in the first fourteen days, with a shared seller_id and a days_since_signup property: profile fields saved, portfolio images uploaded, quote sent, quote accepted, message replied, payout method added, first payout received.

What they found

Profile completion correlated with retention because retained sellers eventually completed their profile — not the other way round. The behaviour that actually separated the two groups was replying to a first enquiry within twenty-four hours. Sellers who did were 3.4× more likely to still be active at week eight, whichever way the cohort was cut.

What changed

Activation was redefined as 'first reply within 24 hours', and the whole first-week experience was rebuilt around it: enquiry notifications went to SMS, a canned first reply became one tap, and the onboarding checklist stopped asking for a portfolio before the first job existed.

Week-8 seller retention27%↗ 9 pts vs. last week
Replies within 24h58%↗ from 31% vs. last week
Sellers reaching payout2.1×↗ in first 30 days vs. last week

Week-8 retention by first-week behaviour

Illustrative figures, percent of sellers still active at week eight. Profile completion barely separates from the base rate.

We spent two quarters moving a number that was a symptom. The question that unstuck it was 'what do retained sellers do that churned sellers do not', asked in those words.

Growth lead, marketplace · illustrative

DriftworkConsumer mobile

Week one was decided in the first four minutes

Illustrative example

Situation

Driftwork is a habit-tracking app with strong install numbers from app-store features and a day-7 retention figure that had drifted down for three releases. The team could see the fall in the store console but not what caused it, because nothing was instrumented between install and the first completed session.

What they instrumented

A minimal in-app taxonomy — app_opened, permission_prompted, permission_granted, habit_created, reminder_set, session_completed — plus a single install_source property carried through from the attribution SDK, so cohorts could be compared release to release.

What they found

Users who created a habit in their first session retained at 3× the rate of those who did not, and the notification permission prompt fired before anyone had a reason to say yes. In the worst release, forty-one per cent of first sessions ended within the permission dialog. The regression was not in the store listing; it was four screens in.

What changed

The prompt moved behind the first reminder a user actually wanted, and the empty state was replaced by three one-tap starter habits. Both changes shipped in a week, and the day-7 curve stopped falling on the release after that.

Day-7 retention31%↗ 9 pts vs. last week
Habit in first session64%↗ from 38% vs. last week
Week-one uninstalls−24%↘ per cohort vs. last week

Retention curve, day 0 to day 14

Illustrative figures. The curve flattening around day eight is the signal worth chasing, not the height of day one.

The store console told us retention was falling. It could not tell us that a permission dialog was the fourth screen of the product.

Founding engineer, consumer app · illustrative

// in practice

What teams actually use it for

Six questions cover most of the work. None of them need a modelling layer, and all of them need the definition written down next to the number.

Defining activation

Compare early behaviours against week-four survival and pick the one that actually separates the cohorts, then hold the whole team to it.

Diagnosing a funnel

Find which step loses the most recoverable users, split it by plan, platform or source, and check whether the window is doing the work.

Reading retention honestly

Cohort curves by signup week, with the definition attached, so a flattening curve is not confused with a rising average.

Measuring feature adoption

Track who reached a new feature, who used it twice, and whether the second use changed anything downstream.

Testing pricing changes

Segment by plan and usage tier to see who upgrades, who stalls at a limit, and which limits are load-bearing.

Answering the ad-hoc question

Ask in plain English during a standup, see the query the answer came from, and save it as a report when it turns out to matter.

// what it takes

The first week, in three numbers

Every story on this page starts the same way: a small, honest event taxonomy, sent from the SDK on day one. The analysis is the easy part once the events mean something.

Read how the pieces fit together in the product overview, check what it costs on pricing, or start with the quickstart.

Events to start0enough to diagnose onboarding
Minutes to first event0install, key, one call
Weeks to a readable cohort0one to instrument, one to collect

There is no seat count in any of this: pricing is on events, so everyone who needs to see the number can.

// go deeper

The reasoning behind these stories

Each case above rests on a piece of analysis practice we have written up properly.

How to read a funnel without fooling yourself

Why the steepest drop is rarely the biggest opportunity, and how to rank steps by recoverable users instead.

Read the post

Your first activation metric is probably wrong

The method Peakline's story follows: compare early behaviours against long-run retention, then test the candidate.

Read the post

What retention actually measures

Three curves, three questions, and why the flattening point matters far more than the height of day one.

Read the post

Ask the question you have been guessing at

Send six events, wait a week, and see which step of your own onboarding is doing the damage. Free while you are small.

Start free Talk to sales

The companies named on this page are fictional and the numbers are illustrative.