MWG Advisory · Method
A reusable rubric for health and fitness apps. It scores whether a product actually changes the user's state — and whether it holds their attention while doing it.
The core test
Does the product make the user measurably better — and make that improvement legible and worth returning to, at a friction cost the user will keep paying?
A retentive loop creates value only when both things are true: the user's state improves — more capability, less pain, better wellbeing — and the loop is engaging and low-friction enough that they keep running it.
One correction this forces: dependence on prompts may fall over time, but engagement with the loop should not. A working loop keeps producing new targets and may evolve habits, so the user keeps having a reason to run it. Graduating the user out of the product is not the goal; graduating them out of one objective and into the next one is.
Games did not win retention by adding points. They connected a clear objective, appropriately scaled challenge, legible feedback, and a reason to attempt the next cycle. Health and fitness apps often copy the visible layer — streaks, badges, notifications — and leave the loop underneath incomplete.
The Loop Audit tests two things at once: whether the product changes the user's state, and whether it holds attention while doing it. The unit of analysis is not a feature. It is the complete cycle from “what is true?” to “what should I do next?” — and the friction the user pays to run it.
Seven stages for scoring; four for publishing. Diagnosis maps to Diagnose. Assignment covers Assign and Enable. Verification covers Verify and Coach. Progress covers Progress and Re-enter.
The value compounds only when the output of one stage changes the next. A dashboard that does not alter the prescription is reporting. A plan that ignores performance is content. A streak that rewards opening the app is attendance, not progress.
Rate each dimension from 0 to 5, multiply by its weight, and total to 100. Keep both the weighted score and the raw profile: two products can earn 72 for entirely different reasons.
| Dimension | Weight | The test | Failure mode |
|---|---|---|---|
| Diagnosis | 10% | Which input actually alters the plan? | An absent or long intake that produces a generic home screen |
| Assignment | 15% | Is today's choice obvious, and the reason legible? | A library or dashboard posing as a plan |
| Action enablement | 10% | Can the user start in 30 seconds and finish unaided? | High production value, poor fit to time or ability |
| Verification | 15% | Does the product know the behavior occurred, and how well? | Logging activity instead of measuring the target |
| Coaching & adaptation | 15% | Can the user point to evidence that yesterday changed today? | Static tips, generic praise, hidden logic |
| Progression & mastery | 15% | What can the user do now that they could not before? | Streak length as the only durable progress |
| Re-entry & cadence | 10% | What happens after three missed days? | Push frequency substituted for relevance |
| Trust, agency & safety | 10% | Can users override, export, recover, understand? | Incentives that reward anxiety or overtraining |
| Score | Meaning | Observable standard |
|---|---|---|
| 0 | Absent | The product does not attempt the job. |
| 1 | Cosmetic | A label or generic prompt exists, using no meaningful context. |
| 2 | Functional | The job works, largely through static rules or manual input. |
| 3 | Contextual | The experience responds to goals, state, or recent behavior. |
| 4 | Adaptive | The system changes difficulty, dose, timing or guidance from observed outcomes. |
| 5 | Compounding | The loop improves both the user and its own model; progress survives novelty. |
| Total | Interpretation | What it usually means |
|---|---|---|
| 85–100 | Compounding loop | Evidence changes action; mastery and trust deepen with use. |
| 70–84 | Strong but leaky | A credible loop with one or two material breaks. |
| 50–69 | Useful tool | Delivers value, but retention relies on user motivation. |
| 30–49 | Feature stack | Several jobs exist without causal connection. |
| 0–29 | Engagement theater | The visible mechanics exceed the behavior-change value. |
Minimum evidence standard
Do not infer adaptation from changing content alone. Require a visible or testable connection between user state or performance and the next prescription. Mark every score O, I or X — observed in use, inferred from behavior or documentation, or not testable in the observation window. Only observed scores get published as firm conclusions.
Friction suppresses improvement and attention at the same time, and it hides from the eight dimensions because it isn't a job the product does — it's a cost the user pays. Score it separately: for each stage, rate what the user pays to run it — time, equipment, another person, cognitive load, emotional cost — from 0 (negligible) to 3 (blocks the stage for most users).
Measurement friction deserves particular attention. An objective test that needs a second person, or a ten-minute setup, will risk being run rarely — which caps verification and progression however good the instrument is. A product can hold the best measurement in its category and still show the user nothing, because the measurement is too expensive to repeat.
Report the ledger beside the score. Two products with the same weighted total behave very differently when one costs three minutes a day and the other costs an afternoon.
| Metric | Operational definition | Why it matters |
|---|---|---|
| Activation | First personalized action completed | Replaces signup or first open |
| Loop completion | Diagnose → action → evidence → next plan | Separates use from value |
| Adaptive yield | Share of cycles where evidence changes the next action | Tests whether personalization is real |
| Progress velocity | Change in outcome per completed cycle | Connects retention to efficacy |
| Recovery rate | Return after lapse without downgrade or churn | Measures humane re-entry |
| Friction cost | Median time and inputs per completed cycle | The tax that suppresses both axes |
| Trust | Use of correction, override, export, explanation | Surfaces agency and confidence |
The sharpest closing move
Don't end a teardown with “add more gamification.” End with the missing causal link: use [evidence] to change [assignment], explain [why], and make [progress] visible at [cadence].