Product case study
Payment Signal
A payment investigation product that turns transaction records into a reviewable decision.
The problem
A fall in approvals says what happened, not why.
Payment teams still need to find the bank, payment method, and failure behind the change. I designed one investigation that checks the data, tests the leading explanation, recommends the next investigation, and compares later payments with the earlier rate.
Scope: completed card authorisation incidents. Live feeds, fraud decisions, automated fixes, and UPI are outside this study.
Independent study using labelled sample data. No bank, merchant, or customer data was used.
How it works
One path from file to action
- 01Validate
Reject incomplete or unreliable payment files.
- 02Compare
Find where failures rose above the earlier rate.
- 03Test
Check whether one failure signal explains the change.
- 04Recommend
Show the evidence and the next investigation.
- 05Check later payments
Compare the affected payments with their earlier rate.
How the explanation was found
Three comparisons pointed to Bank A's online OTP connection.
Its approval rate fell from 93% to 68%. It produced 1,000 of the 1,040 additional failures.
Every timed out payment failed, while payments that completed OTP stayed stable at about 94% approval.
Other payment combinations stayed stable. The next 2,000 Bank A online payments reached 92.5%, compared with 93% earlier.
If the same 1,000 OTP timeout failures were prevented, additional failures would fall from 1,040 to 40 and approval would rise from 83.5% to 91.8%. The estimate uses generated data and is not a measured business result.
Eleven generated decision and file safety cases pass twenty-eight automated checks. The prototype can also compare anonymised historical incidents with decisions verified by a payment specialist. Historical accuracy has not been measured because no reviewed incident data was used.
Working prototype
See the analysis step by step
Review the result, supporting evidence, decision rules, and tests. You can also upload a payment file or reviewed historical incidents.
Next