An onboarding teardown pairs each step in your signup funnel with recent users’ accounts of what happened there. The funnel shows where behaviour changed. Interviews help you form and test explanations for why.
That distinction matters. Imagine this illustrative funnel—not Maren customer data:
- 100 people create an account;
- 71 confirm their email;
- 58 reach the product;
- 31 create a first project;
- 14 connect the data the product needs;
- 9 invite a teammate and reach the account-level activation milestone.
The numbers tell you where to look. They do not tell you whether people misunderstood the promise, lacked permission to connect data, hit a defect, postponed the task, or never intended to use the product. Nielsen Norman Group describes analytics as behavioural data and interviews as attitudinal data; mismatches between the two can be especially revealing.
This template shows how to investigate those mismatches with up to 20 recent signups, turn the evidence into a teardown table, and choose a change you can measure.
What an onboarding teardown is
A conventional teardown walks through a first-run experience screen by screen and comments on the design. This version adds evidence from the people who encountered the flow.
For every meaningful step, capture four things:
- what the product asks the person to do;
- what the funnel records;
- what recent users remember doing and expecting;
- what other evidence would confirm or challenge their explanation.
It is not a usability test. You are reconstructing an event that has already happened, so memory is imperfect. It is not a replacement for analytics or session replay either. Interviews can reveal intent, expectations, constraints, and workarounds; they cannot reliably establish how often something happened across the whole population.
If you need to understand an activation milestone rather than the full flow, start with the B2B SaaS activation interview guide.
Why plan for 20 interviews?
Twenty is a planning cap, not a universal sample-size rule and not a promise of representativeness.
The useful idea is to sample across the funnel instead of interviewing only successful users. Your first recruiting grid might cross outcome with the segment most likely to change the experience:
| Activated | Stalled | |
|---|---|---|
| Primary role or acquisition path | 4–5 | 4–5 |
| Secondary role or acquisition path | 4–5 | 4–5 |
That gives you room for roughly four or five conversations in each cell. If role does not matter, use device, integration path, company size, or another difference tied to your decision. Do not create segments merely to fill a grid.
Recruitment will rarely land perfectly. Treat 20 as the maximum you are prepared to analyse, then review patterns after each small batch. Stop a segment when additional conversations no longer change the themes relevant to your decision; the guide on knowing when you have talked to enough users explains a defensible stopping rule.
The stalled group is essential. People who activated are usually easier to reach, but interviewing only them turns the study into an account of the route that worked. Deliberately include people who stopped at different steps, and record how many invitations each group needed. A low response from stalled users is itself a sampling limitation, not evidence about why they stalled. See the silent-majority problem for ways to handle that gap.
Recruit while the first session is still recent
Interview memory gets less reliable as the event recedes. Nielsen Norman Group lists recall as a limitation of interviews, while Product Talk recommends grounding questions in specific instances of past behaviour.
Recruit as soon as your workflow and consent basis allow. “Within a week” can be a useful operational target for a small team, but it is not a research-backed forgetting threshold. Record the delay between signup and interview so you can see whether older accounts provide thinner recollections.
Keep outreach neutral. Tell people you are trying to understand their first session, that the conversation is not a support or sales call, how the material will be used, and whether comments will be attributed. Do not imply that silence means consent, and do not recruit contacts you are not authorised to contact.
For a practical recruiting workflow, use the guide to recruiting participants for user interviews.
Run a 20–30 minute walk-back
Start before the first product screen:
Take me back to the day you signed up. What were you trying to get done?
Then follow the participant’s actual route. Use current screenshots or the live product as memory cues, but acknowledge when the interface has changed since their session.
For each step, ask:
- What did you expect to happen here?
- What did you do next?
- What, if anything, slowed you down?
- What did you try when that happened?
- Who else, if anyone, needed to be involved?
- What made you continue or stop at that point?
Ask for observable detail. “Was onboarding easy?” invites a summary judgment. “What happened after you clicked Continue?” gives you an event you can compare with logs. When an answer becomes general—“I always skip tours”—bring it back to that session: “What did you do with this tour that day?”
Do not explain the interface during the walk-back. A misunderstanding is evidence. Note it, ask what the participant expected instead, and correct it only after the research portion if support is needed.
Use a consistent core guide across participants so the evidence is comparable, while leaving room to probe unexpected details. The guides on avoiding leading questions and taking interview notes can help.
Build the teardown table
Create one row per meaningful step. Keep the funnel metric and interview evidence in separate columns so a vivid quotation cannot masquerade as prevalence.
The following is an illustrative example, not customer research or a claim about typical onboarding:
| Step | Intended job | Funnel signal | Interview evidence from 20 | Competing explanation | Next check |
|---|---|---|---|---|---|
| Landing page CTA | Start an informed trial | 100 accounts | 6 said they signed up mainly to learn what the product did | The landing page may withhold needed detail | Test clearer pre-signup explanation; watch qualified signup and activation |
| Signup form | Create and qualify an account | 100 completions | 5 questioned fields; 3 said an answer was approximate | Friction may come from length, wording, or data sensitivity | Remove one nonessential field and compare completion |
| Email confirmation | Verify contact details | 71 confirmations | 4 nonconfirmers recalled switching apps on mobile and not returning | Delivery, device switching, or low intent could each explain the loss | Inspect delivery and device data; test a clearer return path |
| Welcome tour | Teach the core workflow | 54 early closes | 11 recalled closing it; none recalled a card unaided | The tour may interrupt rather than teach | Move guidance to the first relevant task |
| Empty dashboard | Start the first project | 31 projects | 9 did not map “project” to their work | Vocabulary and missing example content may both matter | Test a concrete example and task-based label |
| Connect data | Reach first value | 14 connections | 7 needed permission or approval from someone else | The blocker may be organisational, not interface-level | Offer a preview or collaborator handoff before connection |
| Invite teammate | Activate the account | 9 invitations | 5 wanted personal proof of value first | The invitation may arrive too early for this segment | Test the sequence at account level |
Attach a source to every piece of interview evidence: participant ID, interview date, and transcript or note location. Also record disconfirming evidence. If six people blame the confirmation step and two completed it easily on the same device, both belong in the table.
What to inspect at each signup step
1. The promise and the decision to sign up
Samuel Hulick recommends learning what the person hoped to get and what tipped the scales on that particular day. Separate urgent jobs from curiosity. They enter the same funnel with different chances of reaching value.
Also check whether the CTA matched intent. In one small Nielsen Norman Group usability study, six of eight participants clicked “Get Started” while looking for information and were pushed into signup. That is not a benchmark for SaaS. It is a reminder that an apparent conversion can conceal a promise problem.
2. The form and verification step
Ask what each unusual field meant to the participant and whether they could answer accurately. Baymard’s 2024 e-commerce research found an average of 11.3 checkout fields where it estimated most checkouts needed eight. Checkout is not SaaS signup, so use the finding as a design prompt, not an expected effect size.
For verification, keep several hypotheses open: message delivery, device switching, perceived risk, effort, loss of urgency, and simple distraction. Do not jump from “confirmation has the largest drop” to “remove confirmation.” Security, compliance, and account-recovery requirements still apply.
3. Tours, checklists, and empty states
In a controlled Nielsen Norman Group study of 70 people using four relatively simple iOS apps, tutorial readers and skippers had similar task-success rates: 91% versus 94%, a non-significant difference. Tutorial readers rated the tasks as harder. That result does not prove every tour fails. It does justify testing whether guidance works at the moment of need instead.
Ask people to recall a tour before showing it. Then ask what they needed on the first real screen. Nielsen Norman Group recommends using empty states to explain what belongs there and offer a clear next action.
4. Integrations, permissions, and teammates
A B2B product often depends on something one user cannot provide alone: admin approval, data access, a colleague’s participation, or a security review. Ask who owned the missing permission, what the participant would have needed to ask them, and what proof of value existed before that request.
Activation may therefore be an account-level milestone. Rachitsky and Timen give “10 items created and 2+ active editors by week four” as an example of an account-level activation definition. Use your own retention-linked behaviour, not someone else’s milestone.
Turn findings into one measurable change
Do not rank findings by quote count alone. For each candidate problem, assess:
- reach: how many funnel records could be exposed to it;
- evidence: whether interviews, behavioural data, and support signals agree;
- severity: whether it blocks value or merely annoys;
- confidence: whether alternative explanations remain plausible;
- reversibility: how cheaply you can test the change.
Choose the smallest change that tests the explanation. If people do not understand “project,” try a task-based label before rebuilding the dashboard. If users need admin permission to connect data, test a preview or handoff before redesigning the integration.
Then measure the same funnel and interview a fresh sample. Superhuman reports that simplifying its first self-serve screen coincided with activation moving from 40% to 50%. Treat that as one company’s first-person case study, not a forecast for yours. The reusable lesson is to connect a diagnosed problem, a specific change, and a measured outcome.
Use Maren for the interview layer
Maren can run the conversational part of this workflow asynchronously. Create a project around one onboarding decision, invite recent users by email or share a participant link, and let Maren ask adaptive follow-ups. Participants do not need an account. You can review transcripts and individual summaries, then synthesise patterns across completed interviews.
Bring your own funnel data and sampling logic. Maren helps collect and organise what participants say; it does not turn 20 interviews into a representative measurement of every signup. See how Maren runs AI user interviews.