The problem with your retention data

Two users leave.
One is a success.

A person who stops using your app because they got better, and a person who stops because it frustrated them, produce the identical signal in every engagement dashboard ever built. Declining usage, then nothing.

If a real share of your churn is the first kind, a retention initiative will not just waste its budget. It will add friction trying to hold onto people who are finished.

The four trajectories Derived taxonomy

Low friction · Stays

Satisfied Persister

The outcome the product was designed for.

Low friction · Leaves

Satisfied Abandoner

The graduation effect. Left because it worked. A success that reads as churn.

High friction · Stays

Friction-Tolerant Persister

Habit offsets real friction. Retained, but fragile.

High friction · Leaves

Frustrated Abandoner

The genuine product failure, and the only one your dashboard can catch.

In the underlying survey data, users who discontinued after their condition improved scored higher on enablers and satisfaction than users who left frustrated. Opposite outcomes, indistinguishable metrics.

Measured, not inferred

This is not a
model's opinion.

Any general-purpose AI can review your screens and tell you a button is confusing. It is reasoning from design heuristics, and design heuristics are opinions that sound confident.

Every construct in a Friction Audit corresponds to a survey item whose relationship to abandonment was measured in a large-scale, two-stage study. The scores are positions on validated scales, not impressions.

Stage 01

170,127 reviews

Quantitative text analysis across the mental health app category, surfacing what actually drives people out in their own words, at scale, outside a lab.

Stage 02

314 real users

A survey testing each construct against reported behavior, so every score in your report traces back to a measured relationship rather than a plausible-sounding rule of thumb.

The proof

It found things a model wouldn't

Satisfaction does not reliably predict retention here. Friction keeps a direct effect on leaving even after satisfaction is controlled for. Both contradict the intuitive assumption, which is precisely why heuristic reasoning never arrives at them.

The distinction matters most when the answers disagree. A heuristic engine will tell you satisfied users stay. The data says some of your most satisfied users are the ones leaving, because they finished.

Who this is for

Three situations where this pays for itself.

01

Grant-funded intervention teams

NIH, SBIR, SAMHSA and foundation-funded projects where engagement is an outcome your renewal depends on, and the evaluation budget is already approved.

Trigger: writing an evaluation plan, or a mid-project engagement problem you need documented before the report is due.

Can be named in your application

02

Behavioral health product teams

You're about to commit a quarter of engineering time to retention and you want independent evidence about where to point it, defensible to a board that will ask.

Trigger: a board cycle, a fundraise, a retention cliff, or a new head of product taking stock.

Pre-launch or live

03

Agencies, payers and funders

You are choosing between apps to recommend, procure or fund, and the clearinghouses that used to help with that decision are no longer maintained.

Trigger: a vendor selection, a portfolio review, or a recommendation you will have to defend.

Multi-app reviews available

What you get

Four deliverables.
Nothing padded.

A report nobody finishes has no value, so length is capped on purpose. The prioritized fix list is the part clients actually use, and everything else exists to make it defensible.

Deliverable 01

The report, 12 to 18 pages

One-page executive summary, written so a board member can read only that page
Scoring against all 10 assessable inhibitor and enabler constructs
Trajectory-mix estimate across the four quadrants, with reasoning
Prioritized fix list, ranked by expected impact against implementation cost
An explicit limitations section

Deliverables 02 to 04

Scoring appendix, readout, follow-up

Scoring appendix giving every construct, the evidence used, and the score, so you can disagree with a judgment and see exactly what produced it
90-minute readout call with your product team, recorded
One round of written follow-up within 30 days

Timeline: three weeks from the day access is granted.

See it before you buy it

A real sample,
gaps and all.

A complete 13-page Friction Audit of a well-known meditation and mental health app, scored entirely from publicly available material. No proprietary access was used and none was requested.

It scores 8 of 10 constructs and deliberately withholds two, because they cannot be judged honestly from outside an app. Every gap is marked, with the evidence that would close it.

Read the sample report

PDF, 13 pages. Opens in a new tab, no email required.

What it costs

Three tiers.

Two options, because there are only two useful shapes of this work. For comparison, a full-service UX research agency study runs $15,000 to $40,000 and takes 8 to 12 weeks. One engineering sprint pointed at the wrong fix costs more than either option here.

Focused

$6,500

One flow: onboarding, paywall, or check-in cadence.

8 to 10 page report
60-minute readout
2 weeks

Standard · most chosen

$12,000

Full app, all 10 assessable constructs.

12 to 18 page report
Full scoring appendix
90-minute readout
30-day written follow-up
3 weeks

Terms: 50% on signature, 50% on delivery. Audits are delivered personally, by me. You are not buying agency capacity and you will not be handed to an account manager.

Scope

What it isn't.

Stated up front so you can rule it out quickly if it's the wrong instrument. That saves us both a call.

Not a clinical validation or efficacy review
Not a regulatory, FDA or HIPAA compliance assessment
Not a security or privacy audit
Not a safety review of crisis-response features
Not a prediction of your actual retention rate. This is an assessment grounded in measured relationships, and the difference is stated plainly in every report
Does not assess app stability or crash frequency. Those need crash analytics and usage logs, which I do not collect and do not offer at any price
No implementation work. Recommendations only

The evidence behind it

Where the taxonomy comes from.

170,127 App reviews analyzed
with topic modeling

314 Users surveyed
on abandonment

12 Constructs, each tied
to a measured item

The framework comes from a large-scale, two-stage study of why people leave mental health apps, recognized for distinguished research contribution and supported by four peer-reviewed publications. Every construct maps to a survey item whose relationship to abandonment was measured, not assumed.

See the research The Evidence Library

Fair questions

Objections, answered.

Good, you should. Usability testing tells you whether people can complete a task. It does not tell you whether they will still be opening the app in week six. Those are different failures with different causes, and the second one is what determines whether your intervention works.

It is derived from 170,127 reviews across the mental health app category and a survey of 314 users, published and peer-reviewed. It is scoped to young adults and every report says so. If your population is meaningfully different, I will tell you on the fit call, before you pay me.

You can, and you will get generic heuristics. What you will not get is scoring against constructs whose relationship to abandonment has been measured in real users, or the distinction between churn that means failure and churn that means success. That distinction is the entire point, and no general-purpose tool makes it.

It is less than a third of a full agency study, and less than a single engineering sprint pointed in the wrong direction. The useful question is what you are planning to spend on retention work next quarter. If the answer is a quarter of engineering time, this is roughly 3% of that, spent on knowing where to point it.

That is the best time. A complete Figma prototype is enough to score most constructs, and changing a paywall position in a prototype costs nothing compared with changing it after launch.

Yes, and it is one of the better uses of it. I can provide a short scope description and budget justification you can paste directly into an evaluation plan, and be named as a consultant or subcontractor. Ask on the call and I will send it the same day.

Next step

Twenty minutes.
No pitch deck.

Tell me what you're building and where the engagement problem shows up. If a Friction Audit is the wrong instrument, I will say so on the call rather than sell you one.

Bring your retention numbers if you have them. You do not need to share anything confidential to have a useful conversation.

Book a 20-minute fit call

Prefer to write first? Send a message instead.