Define completion as an outcome
Completion should mean the customer reached the first real result, not that they ticked every item on a list. For a reporting product it might be the first report sent to a client. For a service it might be the first approved deliverable. If you have not named this milestone, do that first using How to design customer onboarding.
Then count: of customers who started onboarding in a given month, how many reached the milestone within the target window, such as 30 days.
Map the steps and count the drop
List every step from signing to first result, in order. Next to each one, put the number of customers who reached it. Ten steps, 100 customers: perhaps 96, 90, 88, 55, 52, 50, 49, 30, 28, 27.
The biggest gap is the one to work on. Here, going from 88 to 55 is a bigger loss than all the other steps combined. Most onboardings have one cliff, not a gentle slope.
Find out why people stop
Numbers say where. They do not say why. Contact five to ten customers who stalled at the cliff, within a week, with a short plain message asking what got in the way. Offer a ten minute call.
Sort the answers into four buckets: they did not understand what to do, they lacked something they needed from their own side, the task took more time than they had, or they lost interest. Each one needs a different fix.
Remove the dependency you control
Customers often stall because they are waiting on something. Access to a system, a file, a decision from a colleague. Ask for these things earlier, in the sales process or in a pre-kickoff checklist, so the customer arrives ready.
If the step needs data from them, give a template. If it needs a login, give a three line instruction with a screenshot. Reduce the work they do, not the content you deliver.
Offer help at the cliff
Put a person or a prompt exactly where people stop. If 30 per cent of customers stall at step four, send a message when someone has been on step three for three days: "Most people get stuck on the next step. Here is a five minute walkthrough, or I can do it with you on a call."
In-app prompts from a tool such as Userpilot or Intercom can do this when your product has a login. For a service, a scheduled check-in on day five works the same way.
Cut or reorder steps
Ask of every step whether the first result truly needs it. Move optional setup, such as branding or extra integrations, to after the milestone. Put the step that gives the quickest visible win first. People who see something working in the first ten minutes tend to keep going.
Track by cohort
Compare each month's starters with the previous month's. Do not compare this month's overall rate with last year's, because customer mix changes. A cohort table with the step counts above, one row per start month, shows at once whether the cliff moved.
Give each change a date, and compare only cohorts that started after it.
Common mistakes
- Measuring logins instead of the outcome.
- Adding more tutorials when the real blocker is something on the customer's side.
- Trying to fix every step at once.
- Making all customers follow the same path when a small segment needs something different.
- Declaring victory after one good week.
How you know it works
The step that used to lose half your customers now loses far fewer, and the next cohort shows it too. Time to first result gets shorter. Fewer new customers cancel in the first 90 days. When the support questions in week two change from "how do I start?" to "how do I do more?", onboarding is doing its job.
