The challenge
A Bangalore fashion marketplace running 1,400 partner brands and 9 lakh monthly orders had a 38% return rate — the standard apparel curse. Half were genuine size mismatches, but a measurable slice were serial fraudsters returning worn or swapped items. Manual fraud review was a 6-person team that caught maybe 20% of cases. Influencer outreach was being run on spreadsheets and most creators ghosted.
How we deployed
- Trained a size-recommendation model on 14 months of return-reason data, brand fit notes and historical sizing — surfaced as a WhatsApp + on-PDP chatbot.
- Built a return-fraud risk score combining buyer history, return frequency, payment method and pin-code clustering.
- High-risk returns auto-routed to a video-evidence intake bot before the pickup was even scheduled.
- Replaced creator spreadsheets with an AI outreach engine — segmented by category, follower band and engagement rate, ran personalised first-touch on Instagram DM.
- Closed the loop — creator-driven sales attributed back via UTM and rate cards auto-adjusted on performance.
What changed
- Overall return rate dropped from 38% to 22% in 6 months.
- Rs 2.3 crore in fraudulent return claims blocked annually — paid for the deployment 11× over.
- Creator outreach reach grew 220% as the AI ran 6,000 first-touches a month vs the team's previous 400.
- AOV lifted 17% as size confidence pushed shoppers into 2+ item baskets.
- Manual fraud review team redeployed to seller quality audits.
"Size returns were the tax we paid for being a fashion marketplace. Now we charge it back to the fraudsters and the genuine buyers actually get clothes that fit."
— Head of CX · Fashion Marketplace