05 / CONTROL
Assumptions and risks
What must be true, what can break and how the pilot contains each risk.
01
Core assumptions
| Assumption | Why it matters | How the pilot tests it |
|---|---|---|
| Members will pass the exit reader consistently | Exit capture depends on passive or near-passive behaviour | Observe eligible exits and compare with captured events |
| Existing entry identifiers can be matched to exit identifiers | Completed visits require one member identity across both events | Validate identity mapping before site launch |
| Visit frequency decline is operationally useful | This is the primary signal for intervention | Compare recent behaviour, flags and later outcomes |
| Session duration decline may add signal value | This is the distinctive pilot hypothesis | Test incremental value beyond frequency, without claiming causality |
| Gym teams will review and act on reports | Value depends on operator behaviour, not insight alone | Measure report open and action rates |
| Operators will record outcomes accurately | Closed-loop evidence requires action data | Audit samples and simplify required fields |
| A fair control group can be maintained | Retention impact needs a credible comparison | Pre-register assignment and exclusions |
| Membership revenue inputs are available | Commercial value needs a transparent estimate | Agree revenue source and calculation before evaluation |
02
Risk register
| Risk | Likelihood | Impact | Early indicator | Mitigation | Owner |
|---|---|---|---|---|---|
| Members bypass or miss the reader | Medium | High | Low exit capture accuracy | Improve placement, cues and reader range | Technical lead |
| Duplicate or accidental scans | Medium | Medium | Implausibly short visits or repeated events | Apply de-duplication and minimum visit rules | Data lead |
| Long or overnight open visits | High | Medium | High unmatched entry count | Auto-close exceptions separately, never treat as measured duration | Data lead |
| Entry and exit clocks drift | Low | High | Negative or implausible durations | Synchronise clocks and monitor drift | Technical lead |
| Member IDs do not match across systems | Medium | High | High unmatched event rate | Build tested mapping and exception queue | Technical lead |
| Staff do not open reports | Medium | High | Open rate below target | Use nominated owners, reminders and short reports | Site owner |
| Staff action is inconsistent | High | High | Action rate below target | Limit weekly volume and define action service level | Site owner |
| Outcome recording is incomplete | High | High | Missing action or disposition fields | Use required fields and weekly audits | Pilot lead |
| Intervention and control groups differ materially | Medium | High | Baseline imbalance | Match or randomise, then report residual differences | Analysis lead |
| Seasonality distorts behaviour | Medium | Medium | Site-wide frequency shifts | Use same-site controls and calendar annotations | Analysis lead |
| Cancellation status arrives late | Medium | Medium | Outcome lag in reporting | Define status cut-off and refresh final analysis | Gym sponsor |
| Privacy notice is unclear | Low | High | Complaints or opt-outs | Use plain-language notice and documented data roles | Pilot lead |
| Member outreach feels intrusive | Medium | High | Complaints or negative responses | Provide service-led scripts and contact limits | Site owner |
| Session duration is treated as proven | Medium | High | Sales or staff claims exceed evidence | Label as hypothesis in product and reporting | Taango lead |
| Retained revenue is overstated | Medium | High | Estimate ignores control or horizon | Use pre-agreed formula and sensitivity range | Analysis lead |
03
Analysis safeguards
- Pre-register thresholds, windows, exclusions and group assignment.
- Separate measured facts from hypotheses and estimates.
- Report confidence intervals or uncertainty ranges where sample size allows.
- Do not infer causation from simple before-and-after comparisons.
- Keep data-quality failures visible in the scorecard.
- Lock the final dataset before reviewing the primary outcome.
04
Privacy and member safeguards
- Collect only data required for the pilot questions.
- Provide clear notice before collection begins.
- Define controller, processor and gym responsibilities.
- Restrict access by role and site.
- Set retention and deletion periods before launch.
- Provide a practical route for questions, corrections and opt-out where applicable.
- Do not use access data for unrelated marketing without a separate lawful basis.