The Business Challenge
The business situation the organisation faced before the project began.
A cashback percentage tells a shopper little about what they will actually save. A family might earn ₹40 on groceries and still spend more overall because the same basket costs less elsewhere. Minimum spends, excluded items and delayed payouts make the comparison harder. A local retailer faces a related question: will funding the offer bring a shopper back often enough to cover its cost?
The project began as an idea for a bank-issued credit card secured by a fixed deposit. The immediate business challenge was to identify which part of that idea could be tested locally, within a small budget, before entering an issuer partnership.
Why AI Was the Right Solution
Why AI was the right approach — and what alternatives were considered.
Merchant terms mix ordinary language with precise conditions. A Telugu explanation should sound natural to a shopper while preserving the difference between “₹40 per order” and “up to ₹40 per month.” AI is worth testing for the first draft because a reviewer can compare it directly with the approved source.
For a small catalogue, a human-written template may already be quicker. The pilot would compare manual drafting with AI drafting plus review on equivalent offers. The decision to use AI would depend on total time to an accurate, approved version, including corrections. Reward eligibility, amounts and payouts would follow fixed rules.
How the Solution Was Delivered
Discovery, design, development, testing and rollout — the journey, not the tooling.
- 1Reviewed published secured-card, cashback and retail loyalty offers to identify the alternatives a shopper already has.
- 2Mapped one grocery purchase from reading the offer through payment, merchant confirmation, returns and reward settlement.
- 3Created concept screens showing the eligibility rules, expected payout date and reward status.
- 4Modelled the same order from the shopper’s, merchant’s and platform’s perspectives.
- 5Defined how a bilingual reviewer would verify an AI-drafted offer against the merchant’s approved terms.
- 6Prepared a ₹25,000 validation plan with interview targets, pilot measures and criteria for expanding.
Key Technical & Architecture Decisions
Architecture, model selection, workflow and trade-offs.
The proposed demo uses a mobile-friendly web interface, Spring Boot and PostgreSQL. A shop-specific offer code links a purchase to the published terms without requiring access to a shopper’s card statement. A reward ledger records confirmations, reversals and payouts. Duplicate confirmation of the same order should leave the balance unchanged. A failed payout should remain visible until reconciled.
For translation, each draft would be linked to a specific version of the merchant-approved terms. A bilingual reviewer would verify amounts, caps, exclusions, dates and return conditions before publication. A change to the source would send the translation back for review. The model would receive offer text only. Model selection remains open pending a comparison of translation quality, review time and cost.
Challenges & How They Were Solved
Obstacles hit along the way and how they were overcome.
The first design challenge was identifying who would fund the reward. The original idea explored earnings from UPI transactions. The revised model ties each reward to a merchant commission on a confirmed eligible purchase. A pilot merchant can therefore assess the proposal against an actual invoice before agreeing to participate.
A second challenge was defining what “cashback earned” means when an order can still be returned. The design introduced distinct pending, confirmed, paid and reversed states. It also assigned the merchant a confirmation role and gave the shopper a way to raise a missing-reward query.
A third challenge was keeping translated promises accurate. The proposed review process checks every financial condition against its source. A discrepancy in a cap, exclusion or payout date would block publication until corrected.
User Adoption & Change Management
How users responded — training, change management and feedback loops.
Customer and merchant adoption remain to be tested. The first round would involve 30 prospective shoppers and 10 merchants. Shoppers would use a sample basket and explain in their own words what qualifies, how much they expect to receive and when it should arrive. Merchants would work through one invoice and identify a reward they could afford to fund.
Subject to signed terms, the next step would be a 50-user pilot across three shops. Cashiers would receive a short demonstration and a shop-specific offer code. The team would review missing claims, questions and repeat purchases weekly. Repeated confusion about the same condition would trigger a change to the wording or counter display.
Business Outcomes & Impact
The measurable outcomes recorded on this AI Case Study — with the story behind them.
Customer Experience and Quality Improvement are the intended categories. Select them if the form accepts design objectives. If it reserves these selections for achieved outcomes, leave them unselected until the pilot is measured. Measurable Business Outcome The financial model established a break-even test for the pilot. In the base scenario, a merchant contributes ₹60 on a ₹1,000 eligible order. After a ₹40 shopper reward and ₹8 in variable costs, ₹12 remains before acquisition costs, overhead and tax. At an assumed ₹180 to acquire one shopper, the programme would need 15 comparable confirmed orders from that shopper to recover the acquisition cost. The proposed pilot targets are to pay at least 95% of validated rewards by the promised date and to see at least 30% of first-month buyers make a confirmed purchase in month two. Actual savings, repeat purchases, revenue and hours saved remain unmeasured.
The story behind the numbers
The study changed the order of investment. The original card concept depended on a bank partnership. Mapping the shopping journey showed that the team could first test the main customer promise with a willing merchant and the payment methods shoppers already use. That gave the venture a smaller, answerable first question: will someone return to the shop after receiving the promised reward?
The basket comparison also changed what the pilot should report. In one illustration, a shopper pays ₹1,000 and earns ₹40 cashback. The same goods cost ₹980 elsewhere, and reaching the offer costs an extra ₹15. The shopper’s actual saving is ₹5. That distinction informed the decision to measure comparable basket prices and extra costs alongside cashback paid.
Lessons Learned
What surprised the team, what worked well, and what would be done differently.
- Compare the final basket cost with what the shopper would normally pay.
- Agree who funds each reward, how returns affect it and when it becomes payable.
- Measure repeat purchases and contribution for the original group of shoppers, including those who stop buying.
- Test whether shoppers understand the Telugu terms by asking them to explain an example in their own words.
- Evaluate AI on the accuracy and total cost of approved copy, including human review.
- Use observed pilot results before committing to paid acquisition or a bank integration.
Future Opportunities
Where this solution could go next.
Build a demo with simulated orders and test duplicate confirmations, partial returns and failed payouts. Then evaluate Telugu offer drafts against bilingual human-reviewed examples. Record the errors that could change a shopper’s decision and the time needed to correct them.
If a live merchant pilot demonstrates timely payouts, positive shopper savings and repeat purchases, expand the catalogue. An issuer-approved referral or secured-card phase could follow. Its first customer materials would explain the deposit commitment, total bill due, fees and closure process before an applicant proceeds to the bank.
Professional Reflections
Guidance for another builder tackling a similar problem.
Start with one real basket and one merchant invoice. A basket reveals whether the shopper saves money. An invoice reveals whether the merchant can afford the promise. Together they expose assumptions that a polished screen or headline cashback rate can conceal.
For the AI component, ask a bilingual reviewer to list the errors that would change a purchase decision before choosing a model. A fluent translation that drops a minimum spend or changes a payout date fails the customer. Measure the time to publish a trustworthy offer, including review and correction.