The product changed when the customer's interpretation became visible

In 2022 we tested a product presentation using a fictional renovation budget of ₹11.6 lakh and an ₹8 lakh loan. The difference was ₹3.6 lakh to be funded by the household under the example's stated terms. Fourteen of forty participants read the prominent loan presentation as covering the whole project. They had not necessarily made an arithmetic mistake. The design had made the company's approved amount the central answer while leaving the wider work budget and customer contribution much less visible.

The delivery sponsor wanted to retain the launch date and add a short note. Branch colleagues argued that an employee would explain the difference during the actual conversation. I asked the team to test the product without that helpful employee supplying the missing context. The household needed to understand the relationship before deciding what work it could undertake. A clear loan amount was insufficient if the overall presentation implied that the renovation itself was fully funded.

We spent ₹31 lakh and delayed the affected release by six weeks. The revised account showed the work budget, financed portion and customer contribution together, with the source and limitations of the cost estimate. Credit and technical owners retained their assessments; the display did not approve more money or certify the project could be completed for the estimate. A new thirty-six-person test left four participants still requiring assistance, so the supported route remained part of the release plan.

The result was a clearer proposition with additional cost, delay and remaining explanation work. The two research groups were different, so I did not turn the counts into a controlled causal claim about every borrower's understanding. I would bring the whole household task into the first prototype next time. The company had been precise about its own contribution while allowing the customer to infer a wider completion. Product leadership needed to correct that relationship, not simply make the existing number more prominent.

I supported a preferred-contractor panel in 2023 because it appeared to help households begin their renovation and create relevant introductions to our lending service. The business case projected ₹18 crore of annual sourced loan volume. In research with thirty-eight prospective users, fourteen understood the preferred designation as assurance from the lender about the quality of the contractor's work. We did not supply a service capable of supporting that interpretation. A lending relationship and a technical review for our own purpose did not make us the household's building contractor.

The commercial sponsor wanted a qualification beneath the list. Some colleagues felt that users should recognise that the contractor remained a separate business. I accepted responsibility for the proposition I had promoted and asked whether the product's central presentation was making the qualification implausible. The preferred label and ranking were doing more work than the explanatory text. I would not defend that arrangement by treating the user's inference as an unreasonable failure to read.

We withdrew the ranked referral proposition and spent ₹24 lakh on the resulting product and communication changes. Households remained responsible for selecting their own contractors, with the lender's technical and credit requirements explained separately. The company gave up the ₹18 crore projected annual sourced volume associated with that proposal. It was a forecast opportunity, not a realised loss or an amount of loans cancelled after approval. No completed contractor service was transferred to the product team.

The decision made the offer narrower and removed a feature I had expected to be commercially useful. It also prevented our account of the service from depending on a distinction the main design had undermined. I would test the implied responsibility before developing the acquisition case in future. A useful-looking directory can change what a customer believes the organisation stands behind. If the company cannot support that promise, the product needs a different design or a smaller scope, not merely a stronger argument that the qualification was technically present.

A 2024 prototype offered a Cancel remaining loan control for customers who no longer needed the undrawn portion of a staged renovation facility. In thirty scripted user sessions, twelve participants interpreted it as ending the whole loan, including money already borrowed. The intended action was narrower: a request concerning future draw availability, subject to the applicable terms and authorised processing. Existing repayment obligations would continue. The product's convenient wording had collapsed those different consequences.

The team wanted a short control because the longer explanation appeared difficult on a phone. Servicing colleagues wanted fewer ambiguous requests arriving through branches, while legal and credit owners insisted that the action's actual scope remain clear. I recommended redesigning the path around a request to stop future draws, showing the outstanding balance and continuing servicing separately. Confirmation needed to describe the request received and any further assessment, without implying either automatic approval or forgiveness of the existing debt.

The change cost ₹37 lakh and delayed the feature by four weeks. We reviewed the full sequence, including withdrawal before submission, the pending state and the explanation after authorised processing. A repeat thirty-person rehearsal left four participants still requiring help, and we kept an assisted route. The research was a bounded product exercise, not evidence of twelve live accounts whose debt had been wrongly cancelled. No contractual obligation changed simply because the interface wording changed.

The consequence was a longer path with more explanation and continued human support. That was less elegant in a short demonstration, but more faithful to the decision the customer was making. I would name the object of a cancellation early enough that the team cannot use one simple word for several different endings. A product can make a request easy without making its consequences vague. The responsible owners must still decide the actual terms and action; the customer's understanding should not depend on discovering that limit after pressing the control.

In 2025 we reviewed eighty-four self-employed applications in a product trial. Twenty-nine applicants could not accurately represent their circumstances in the required constant monthly-salary field. Some had seasonal receipts; others drew money from a business at uneven intervals. The field encouraged either an invented regular amount or a stalled application. Neither was a useful substitute for the evidence underwriting needed. The finding concerned the representation of income, not a conclusion that those twenty-nine applicants qualified for a loan.

The delivery team preferred a free-text box to preserve the schedule. Underwriters wanted a usable record rather than an explanation separated from supporting documents. I worked with them on a twelve-month evidence calendar and an assisted path that could distinguish business receipts, relevant costs and the income assessment they were authorised to make. The product did not calculate eligibility from a crude annual total or lower the standard because the ordinary field was inconvenient.

The work cost ₹42 lakh and took nine additional weeks. The operating plan also added two underwriting-support analysts at ₹14 lakh annual expense, stated separately from the one-off product programme. Nineteen of the twenty-nine applicants completed the required evidence route during the reviewed period; ten still needed further material. Completion meant the file could proceed to the appropriate assessment, not that a loan had been approved or would perform well. The wider trial population remained eighty-four.

The decision made the service more expensive and the standard journey less uniform. It also stopped the product from asking customers to distort their circumstances to fit the screen. I would involve underwriting and assistance capacity when the evidence model is first designed, rather than treat both as exceptions after a simple form has been approved. A different route can preserve the same professional standard. Its success needs to be reported through the task actually completed, with the people still outside that result remaining visible.

A 2025 review examined 180 accounts within ten days of full repayment. The specified post-repayment tasks were complete for 118. Sixty-two remained in process: 54 involved document collection and eight involved verification of charge-related updates, with no overlap between those groups. Our portal locked the lending account when its balance reached zero, leaving those customers to follow the remaining work through other service channels. The financial event was real, but it did not mean every task the household expected from the lender had finished.

The platform team favoured the existing closure rule because it simplified account access. Operations still had responsibility for completing the work and answering customers through branches and support. I recommended retaining an appropriately controlled service view after repayment, with clear task status, responsible contacts and the next action. Legal and operating owners retained the actual document-release and charge-related duties. The product change could not extend any applicable deadline or replace the action with a reassuring status label.

The programme cost ₹28 lakh and took five weeks. Access controls distinguished the continuing service record from a facility available for new borrowing. A later 160-account review recorded 125 with the specified tasks complete and 35 still in process. The populations and timing differed, so I did not describe the change in proportions as a causal improvement produced by the portal. Nor did the sample alone establish compliance across the loan book. It showed the continuing work that the product needed to represent.

The result was a longer customer relationship in the service design, even after the commercial balance had ended. That required support capacity and a more careful account-state model. I would ask what remains owed to the customer before using zero balance as a universal closure rule. The lender should complete its obligations through the responsible teams, while the product helps the household understand and follow them. A completed financial milestone should make the remaining task clearer, not remove the customer's most familiar route to it.