Decision Workbook #
Purpose. This workbook turns a public case or a practical problem into a recorded behavior choice, a realistic test, and a next decision that matches the evidence.
Choose a packet: M-PESA, Instagram, or GOV.UK Verify. Keep your initial response before opening its worked analysis. Recognizing the company does not prevent practice; explain your choice using the packet’s evidence.
Download the printable workbook or editable Word workbook. The fields below are the same eight sections. Copy them into a document if you prefer to work there.
Suggested standalone time: 40 minutes. The section times are instructional estimates, not measured completion times. A workshop host can spread the sections across a longer discussion. Complete one row or short paragraph per prompt; expand only where the decision needs it.
Record before starting: your name or team label, date, case, source cutoff, and the practical decision you are making. Mark entries as public fact, reported experience, instructor analysis, your inference, or fictional assumption. A later published account of an earlier event remains a later source.
The examples throughout use an invented shift-handover service. Its conditions, thresholds, and choices are teaching examples. They are not company data or a completed study.
1. Define the outcome #
Suggested time: 4 minutes.
| Record | Your entry |
|---|---|
| Useful outcome and whose problem it resolves | |
| Observable action that might contribute | |
| Feature or service that might enable it | |
| Reason the action could produce the outcome | |
| What would show that this link is wrong or incomplete |
Worked example. The outcome is that incoming staff know who owns unfinished work. The action is a prepared handover acknowledged by the incoming lead. The feature is a shared form. Acknowledgement without correct task ownership could satisfy an interface metric while failing the outcome.
Check: Can another person distinguish what people do from what you build and what you hope improves?
2. Compare candidate actions #
Suggested time: 5 minutes.
Write each candidate as actor + observable action + context/opportunity + deadline + support. Include the current workaround and a route that reduces user work.
| Candidate | Full action statement | Expected benefit | Work removed or added; for whom |
|---|---|---|---|
| A | |||
| B | |||
| C, including a less-user-work challenger | |||
| Current route or workaround |
Worked example. A: both leads exchange a written handover. B: they hold a brief discussion and record unresolved items. C: existing task records generate a draft that the leads check and acknowledge. C reduces writing but adds dependence on correct records and review. These are alternative service routes, not observed preferences.
Check: Are the candidates materially different actions? Renaming the button or changing its color is a change within a route.
3. Map actors and support #
Suggested time: 5 minutes.
| Step and actor | Required action or handoff | Prerequisite | Completion evidence | Fallback if it fails |
|---|---|---|---|---|
Add rows until the useful service is complete. Include off-platform work, people who authorize access, recipients, and staff who repair errors.
| Support question | Your entry |
|---|---|
| Normal support the intended service will supply | |
| Exceptional help a trial might accidentally supply | |
| Staff work and cost, including failed attempts | |
| Peak demand, available capacity, and possible waiting | |
| Evidence or assumption supporting those capacity estimates |
Worked example. Updating a task record, writing the handover, resolving an unclear owner, and acknowledging receipt require different work. A researcher who repairs every unclear owner creates exceptional support. A planned supervisor review can be normal support if its time and capacity are included in the service.
Check: Have you removed work, or transferred it to another person or system?
4. Compare the evidence #
Suggested time: 5 minutes.
| Candidate | Supporting observation and source | Contradictory evidence | Unknown | Finding that would change your ranking |
|---|---|---|---|---|
| A | ||||
| B | ||||
| C |
For each serious candidate, add a short note about Dispositional Fit (recurring tendencies and preferences), Capability Fit (abilities, skills, knowledge), and Context Fit (external social and physical conditions). State which segment and window each note describes.
Worked example. If a case packet shows inconsistent source records, a generated handover draft has an unresolved input-quality requirement. That does not establish that people dislike the draft or lack the skill to review it. Observe those questions separately.
Check: Can you justify the provisional selection without a total score? Optional BFA ratings organize judgments. They are not probabilities, validated cutoffs, or proof of fit. Leave a gap as unknown rather than assigning a middle score.
Record your provisional choice: chosen candidate, strongest reason, best alternative, and the most important unresolved assumption.
5. Specify the next test #
Suggested time: 6 minutes.
| Test field | Your entry |
|---|---|
| Decision the test will change | |
| Population, recruitment route, exclusions, and relevant contexts | |
| Action and full normal-support service to test | |
| Comparator and assignment method, if estimating an effect | |
| Observation method, window, and repeat-opportunity rule | |
| Sample and precision rationale, including grouped/repeated records | |
| Completion, quality, cost, and capacity rules with reasons | |
| Missing-data rule and condition for an inconclusive result | |
| Date agreed; any later changes and their reasons |
Worked example. A handover service might require at least 70% completion because its assumed fallback capacity covers only the remaining 30%. That is an invented operating rationale, which must itself be checked. A small pilot above that threshold still needs adequate precision, normal support, and a quality check before it could justify expansion.
Choose a diagnostic observation when you need to locate a break. Choose an appropriate comparison when you need to estimate the effect of a service change. Neither a staged demonstration nor a few convenient volunteers establishes a general population rate.
Check: Would a different result change the decision? If every possible result leads to the same action, explain why this is the next useful test.
6. Define the measures #
Suggested time: 5 minutes.
| Measure | Numerator | Denominator and unit | Window | Data source; unknown/missing rule |
|---|---|---|---|---|
| Reach and applicable opportunity | ||||
| Attempt | ||||
| Completion and quality | ||||
| Repeat completion given another opportunity | ||||
| Useful outcome | ||||
| Staff time, total cost, and capacity |
Worked example. A saved handover form starts an attempt. A usable handover acknowledged by the next lead completes the action. Count team-handover opportunities, and identify repeat completion among first-completing teams that have another applicable handover. A missing acknowledgement record is unknown if the recording failed; a verified lack of acknowledgement by the deadline is a noncompletion.
Keep yes, no, not applicable, and unknown distinct. Report the unresolved count beside each rate. A zero denominator gives an undefined rate. A rate among attempts needs the broader opportunity rate beside it.
Check: Would excluding unreachable people or missing records make the result look better? Show what the exclusions remove and how plausible unresolved outcomes affect the decision.
7. Diagnose the result #
Suggested time: 5 minutes.
For a case with no observed test results, write anticipated interpretations here. Label them as a proposed analysis. Do not invent a successful trial to finish the workbook.
| Result or scenario | Plausible explanations | What this does not establish | Evidence that distinguishes the explanations |
|---|---|---|---|
Worked example. Forms are saved but handovers are not acknowledged. Possible causes include shift overlap, unclear responsibility, a notification problem, or a recording failure. The result alone does not prove that written handovers are the wrong behavior. Compare record checks and observation of the handoff before choosing a repair or alternative.
Check: Separate a treatment that did not help from a behavior that cannot work, and a completed action from a useful outcome. Consider uncertainty, alternate explanations, and who is missing from the data.
8. Record the next decision #
Suggested time: 5 minutes.
Choose one: proceed narrowly, change support, reselect, stop, or inconclusive. A proposed test is a complete next action when the evidence cannot yet support rollout.
| Decision record | Your entry |
|---|---|
| Selected next decision and specific action | |
| Evidence and operating rule that justify it | |
| Population, context, support, and scale limits | |
| Remaining uncertainty and best rival explanation | |
| Owner, resource limit, and review point | |
| Finding that would reverse or reopen the decision |
Worked example. Change support: investigate acknowledgement timing and test a scheduled overlap before extending the written route. Preserve the discussion route as an alternative. This is a proposed response to an invented problem, not evidence that the repair will work.
Final check: Can a reader follow the evidence to your decision, identify what remains unknown, and carry out the next action without asking you to supply missing assumptions?
Compare and revise #
Keep your original response, then open the selected case’s worked analysis. Record one defensible difference and one revision. A good response need not reproduce the historical outcome. It must use the available evidence, expose assumptions, and make a practical next step clear.
Choose your worked analysis: M-PESA, Instagram, or GOV.UK Verify.
Use the measurement lab to check your denominators and failure diagnosis. The teaching notes explain source labels, adaptation, and review of learner responses.
Original workbook text and exercises: Jason Hreha, version 1, 2026-09-05, CC BY 4.0. Linked external sources retain their own rights. This is a teaching resource; no participant study or independent validation of the method is claimed.