BehavioralStrategy.com teaching pack v1.0
Author: Jason Hreha
Source file: education/mpesa-worked-decision.md
License for original text: CC BY 4.0. External sources retain their rights.
Editable text copy; Markdown headings and tables are retained.

# M-PESA: Worked Decision

**Read after completing the [decision packet](https://behavioralstrategy.com/education/mpesa-decision/).** A strong answer connects evidence to a bounded next investment. Agreement with the eventual historical choice is not the marking rule.

This page contains original instructor reasoning and a proposed [Behavior Market Fit](https://behavioralstrategy.com/glossary/behavior-market-fit/) test. These are not recovered M-PESA plans. No evidence reviewed for this lesson establishes that the team used DRIVE, BMF or BFA, or that this teaching method improves real decisions.

## Response A: Test a narrow transfer route

**Decision.** Allocate the next learning budget to a bounded person-to-person transfer test. Start with people who have an actual transfer need and a reachable intended recipient. Define the target as completing the transfer through to usable money at the other end. Do not commit to a national launch on this evidence.

**Why this is reasonable.** The alternative-use observations in the packet make a transfer service worth investigating. A test can ask whether the recipient gets value without adding a lending institution's loan-record requirements to every transaction. That removes one class of dependency from the proposed service. It does not remove agents, financial controls or customer support.

**Strongest objection.** The pilot does not supply a paid, representative transfer funnel. The earlier source's caution about use beyond repayment should reduce confidence in extrapolation. A handful of resourceful participants might account for the most memorable examples. Recruiting only similar enthusiasts would repeat this problem.

**Treatment of the other options.** Repayment remains plausible because it has a recurring, identifiable occasion and a checkable institutional endpoint. Before expanding it, request evidence that the institution can credit and reconcile payments within an agreed operating budget. An open-ended pilot extension is less attractive than a test with a defined target and reversal rule.

**Fit judgment.** A desire to support a relative makes the transfer goal plausible for a defined segment. It does not show that entering a number, managing a PIN or finding cash is easy. A [Behavior Fit Assessment](https://behavioralstrategy.com/glossary/behavior-fit-assessment/) should carry those unknowns forward. Assigning a high score to each dimension would conceal the test's purpose.

**What changes this decision.** Stop expansion if ordinary support cannot deliver recipient access reliably, or if people with a new genuine need consistently choose the existing route at realistic fees. Reconsider repayment if its complete workflow has materially better repeat completion and support economics for a clearly defined population. The choice is conditional on evidence, not on the attractiveness of the remittance story.

## Response B: Establish delivery capacity first

**Decision.** Hold consumer expansion. Use the next bounded work period to audit complete payment episodes within the existing pilot population and test a sustainable operating routine. Decide on a new target population only after the service can distinguish delivered value from incomplete or rescued transactions.

**Why this is reasonable.** The packet leaves the numerator, opportunity denominator and support cost unresolved. Buying a larger transfer test now could produce another ambiguous result if neither endpoint can be verified. This response gives more weight than Response A to the risk of measuring an operation the team cannot yet support.

**The work to fund.** A service operations owner would match customer, agent, platform and endpoint records for the same episodes. That owner would classify the reason for each incomplete result and identify a normal support routine that can be staffed and costed. For a loan episode, this includes the correct borrower credit. For a cash remittance, it includes recipient access. Keep the populations and endpoints separate in the report.

**A candidate with less user work.** Before adding another borrower step, examine whether record matching can be handled inside the service and institution. Compare total customer and staff work with the current route. This is a proposed design question, not a claim that the historical team had this option ready or that automation would necessarily help.

**Strongest objection.** This approach postpones learning about new transfer customers. An operations review can become an excuse to retain the original product or perfect every edge case. Cap it around the specific missing evidence; do not redesign the whole financial system.

**Exit and reversal.** In this teaching example, allow two ordinary payment cycles to test the defined routine. That duration is an instructor's planning assumption. If records remain irreconcilable or delivery requires exceptional rescue, stop expansion and revise the route. If ordinary staff can reconcile the full chain within a predeclared budget, proceed to the transfer test below. If too few genuine opportunities occur, report insufficient evidence and choose whether one targeted extension is worth its cost.

These responses have the **same business objective**. Response A values faster learning about a new use. Response B values resolving the delivery uncertainty first. Neither assumes an obligation to keep the loan partnership or advance the remittance service. A learner can choose either while making a clear, testable judgment about which uncertainty is most costly.

## A complete proposed transfer test

The following protocol makes Response A concrete. Response B could use it after the operations gate. All amounts of time, counts and thresholds below are **illustrative teaching choices**, not historical M-PESA measurements, universal cutoffs or a study ready to run. No participant testing has been conducted for this lesson.

### Population, opportunity and endpoint

Recruit consenting sender-recipient pairs in one defined transfer route where an ordinary agent service can be provided at both ends. Require a current, independently described reason to send money. Record the existing route and expected need timing before offering the test. Include variation in phone familiarity; report whom the access requirements exclude.

The unit is a **pair with a genuine transfer opportunity**, not an account, a message or a transaction attempt. A new opportunity requires a new underlying need. Retrying the same failed payment does not create another opportunity.

For this initial cash-remittance test, completion requires all of the following:

1. The sender chooses the service after the fee and expected delivery time are explained.
2. The intended amount is correctly transferred and confirmed by the service record.
3. The intended recipient obtains the cash within the promised period, with matched endpoint evidence.
4. No unresolved amount, recipient or balance discrepancy remains for that episode.

The transfer promise is **within one day** for this teaching example. Retaining funds electronically could be a valid user choice in another product definition; record it here as a different endpoint. Do not silently count it as completed cash delivery.

### Support that belongs in the offer

The planned service includes an initial explanation, normal agent assistance, replenishment of cash and electronic balances, a staffed help route and an accountable reconciliation owner. Estimate the cost of that support before the test. Maintain it if it is part of the intended commercial service.

Separate normal assistance from exceptional rescue. A project leader making a special trip, personally funding an agent or handling a customer's credentials would fall outside the planned routine. Log the action, cost and result. An eventual completion can still be recorded, but it does not pass the ordinary-support gate. Never deny necessary help merely to improve the experiment's appearance.

Use the intended fee. If the test must subsidize transactions or devices, disclose the subsidy and limit the claim to subsidized conditions. A reward for completing a transfer changes what the test measures. Compensation for research time should be recorded separately from transaction use.

The packet does not supply a defensible price or support-cost budget. The operator must cost and approve those inputs before running this protocol. Leave commercial viability unresolved until then. A precise completion threshold cannot replace a missing operating budget.

### Observation and denominators

Plan an **eight-week observation period**, with a minimum of **40 pairs with a first verified opportunity** and **20 of those pairs with a later verified opportunity**. These small illustrative counts support an exploratory decision; they do not establish population precision or isolate a causal effect. If the expected need occurs less often, change the window before enrollment and explain why.

| Measure | Numerator | Denominator and interpretation |
|---|---|---|
| First-opportunity completion | Pairs meeting the full cash-delivery endpoint on their first opportunity | All enrolled pairs with a verified first opportunity, including those who choose another route or never attempt the service |
| Repeat-opportunity completion | Pairs meeting the endpoint on their next genuine opportunity | All enrolled pairs with a verified next opportunity, including those whose first episode failed; report previous completers separately |
| Ordinary-support completion | Completed episodes within the declared support routine | All verified opportunities; report first and later opportunities separately |
| Support burden | Total ordinary support time and cost; exceptional rescue reported separately | Show totals and cost per completed episode; avoid making failure costs disappear from the total |
| Access and harm | Failed visits, delayed access, errors, loss and unresolved balances | Show counts against the appropriate opportunities and affected pairs, plus each serious incident |

Distinguish five states: completed; known not completed; opportunity verified but outcome unresolved; no opportunity occurred, verified; and opportunity status unknown. Unknown outcomes remain in a verified-opportunity denominator. People whose need status is unknown must be reported separately from the verified no-opportunity group.

Report a conservative completion rate that counts unresolved outcomes as unconfirmed, and an upper bound that assumes they completed. If that interval crosses a decision threshold, the result is insufficient for that gate. For pairs whose later need is unknown, also show the result if each had an uncompleted opportunity. That is a sensitivity check, not a measured rate. Record both nonuse and failures, and compare the time and total cost with the participant's existing route without claiming a randomized comparison.

### Rules agreed before results

For the exercise, use these provisional rules and challenge their business rationale. A real team would set fees, service promises and resource limits from the intended operation before recruitment.

| Decision | Proposed rule |
|---|---|
| Continue to a larger bounded test | At least 80% first-opportunity and 70% repeat-opportunity full completion; the same thresholds hold within ordinary support; both opportunity minimums are met; the declared support budget is met; no unresolved loss or material balance discrepancy |
| Revise the service | A named failure in access, comprehension or ordinary operations explains a missed gate and has a feasible, costed repair; retest the affected chain |
| Stop expansion | An unresolved loss or material balance discrepancy occurs, or the service cannot meet the defined promise and support budget after the scoped repair; resolve affected customer outcomes before further use |
| Insufficient evidence | Opportunity minimums are missed, need or outcome status is unresolved enough to change a decision, or subsidy and rescue prevent a conclusion about the planned offer |

Passing these gates supports the next bounded investment for this route and offer. It does not establish national demand, financial viability at scale, a welfare effect or comparative validation of BMF.

## Diagnose a weak result before changing the goal

The following examples are hypothetical diagnoses. They are not M-PESA pilot results.

| Observation | Competing explanation to check | Next evidence or action |
|---|---|---|
| Senders attempt; recipients cannot obtain cash | Delivery capacity or location fails | Inspect the endpoint and replenishment records before changing motivation messages |
| People complete only when staff take over | The action requires unsupported skill, or training is poorly designed | Observe which step fails; test a simpler action or costed assistance |
| First completion is strong; later use is low | Later need is absent, fees change the choice, or the first experience disappoints | Separate missing opportunity from refusal; inspect price and earlier outcome |
| Borrowers report paying; loan records remain wrong | The institutional handoff fails | Match the episode to the account record; repair reconciliation before more borrower prompts |
| The service works but costs more to support than it earns | The offer or operating model is not sustainable | Cost the full route; reconsider scope or pricing before expansion |

The [measurement lab](https://behavioralstrategy.com/education/measurement-lab/) provides fictional observations for calculation practice. The [test method](https://behavioralstrategy.com/methodology/validate-behavior-market-fit/) explains how to adapt the design to another setting.

## What happened afterward

Hughes and Lonie report that the launch proposition became **Send Money Home**, with cash conversion, person-to-person transfers and airtime purchasing. Their account also describes subsequent changes to agent controls and balance management. [Evidence BS-0079; see source-notes.txt]

In a tenth-anniversary company retrospective, former Safaricom CEO Michael Joseph describes shifting the loan prototype toward transfers, investing in agent training and launching in March 2007. This is later participant testimony, not evidence available to the imagined 2006 meeting. [Read Joseph's account](https://www.vodafone.com/news/newsroom/digital-society/m-pesa-created).

Later household research addresses a different question. Suri and Jack's 2016 study estimated that increased mobile-money access lifted about 194,000 Kenyan households out of extreme poverty. That is a study estimate from later household evidence, not a count of pilot successes or a result of the proposed test above. [MIT's account of the study](https://news.mit.edu/2016/mobile-money-kenyans-out-poverty-1208). [Evidence BS-0007; see source-notes.txt]

The history illustrates how a product direction and its supporting operation can develop together. It cannot establish that this choice was inevitable, that another bounded learning decision was irrational, or that a present-day framework caused the outcome.

## Compare your answer

Review your workbook without giving yourself credit for predicting the eventual launch. Did you define a complete action, keep the contrary evidence, identify each actor, specify normal support, preserve missing data and make a decision that can change?

Use the [facilitator rubric](https://behavioralstrategy.com/education/teaching-notes/) to assess those features. Then write one change to your original decision and name the evidence or reasoning that caused it. An unchanged answer is acceptable if it now states its assumptions and reversal condition more clearly.

The original teaching material follows the [teaching-pack reuse terms](https://behavioralstrategy.com/education/teaching-notes/). The [student source chronology](https://behavioralstrategy.com/education/mpesa-decision/#source-chronology-and-reading-boundary) identifies the pilot sources. Later company and study sources above are outcome evidence. External publications retain their own rights. Sources checked 5 September 2026.
