The investor demo was supposed to be clean.
Sales wanted polished charts, a rehearsed product flow and a recording that showed repayment growth without the mess of exceptions.
Aisha requested a live cohort.
“We don't demo edge cases to investors,” the sales head said.
“The data room is frozen because of an edge case. A recording does not answer the question.”
Investor counsel, Ms. Shah, asked the useful version.
“Can you show a normal payment and a recycled-number path side by side using synthetic data?”
Wekesa said yes.
The first synthetic customer made a normal repayment. Source funds moved, settlement matched and the dashboard updated.
“That's the product,” sales said.
Then the legacy test account received a recycled-number activation event. Because the historical closure code had not migrated, the system reused the old customer key.
“So the phone number becomes identity continuity?” Ms. Shah asked.
“On this legacy path,” Wekesa said. “Not on correctly migrated closures.”
The loan engine opened an eligible profile.
Auto-reserve produced a repayment journal without a customer action.
The KPI increased.
Silence spread across the livestream.
Sales pushed back. “Reserve is real company-controlled liquidity. It isn't imaginary money.”
“Correct,” Aisha said. “The problem is describing the reserve journal as borrower repayment and allowing a recycled number to inherit an old identity.”
Victor called it a known bug.
Ms. Shah asked how many historical profiles were exposed.
“We do not have a final confirmed-harm count,” Aisha said.
Sales had estimates. Aisha refused to present risk population as confirmed cases.
Finance then compared the reported delinquent-cohort repayment rate with the same cohort excluding reserve-generated journals.
The company-wide difference was not catastrophic. In the delinquent segment used heavily in the valuation model, it mattered more.
Ms. Shah asked the denominator and calculation method. Finance provided both so the investor could rerun the result.
She also asked what was not proven.
Aisha listed it: not every auto-reserve entry was improper; there was no evidence the telecom reseller participated in manipulation; the final historical harm count remained open.
That restraint made the hard findings harder to dismiss.
They ran a second synthetic case without the recycled-number bug. Reserve classification still affected the KPI, but no dead identity appeared. The demo separated identity design from reporting treatment.
Finally, they turned on the new independent checker requirement. The emergency journal could not run until the second approver accepted it.
“Remediation works on this path,” Wekesa said.
“Good,” Ms. Shah replied. “Now quantify history.”
Sales asked to rerun a prerecorded clean dataset.
Aisha refused.
“The live test already showed normal and disputed behavior. Removing the second path changes the question.”
No real customer data entered the investor package. Mama Atieno's family was not used as a prop. The data room received synthetic test inputs, configuration, before-and-after KPI calculations and the limited audit findings.
At the end, investor counsel issued the instruction that sales had hoped to avoid.
The investor analyst then requested a reconciliation range rather than a single estimated exposure number. Finance divided the population into potential identity collisions, profiles with new loans, reserve-linked reporting entries and customers or agents with confirmed harm. The largest number was not presented as the loss figure. Ms. Shah said the categories were more useful than a dramatic total because each would affect due diligence differently. Victor called the range unnecessarily cautious. Aisha replied that caution worked in both directions: it prevented the company from understating confirmed cases and prevented reviewers from calling every legacy record a victim.
After the main session, Ms. Shah asked for one more controlled demonstration using a genuine-style customer repayment arriving during the same minute as a reserve journal. The two entries looked similar on the top dashboard but separated cleanly in the raw source classification. That visual comparison made the reporting problem understandable without discussing any real customer. Sales finally agreed that the acquisition deck needed a note distinguishing borrower-initiated payments from reserve-supported stabilization. It was not the wording Aisha would have chosen, but it forced the metric to tell readers that two mechanisms sat beneath one headline number.
**Freeze acquisition data room for affected repayment-KPI history pending reconciliation.**