Fundle.ai: AI First Party Data Platform for Retail Loyalty at Scale

Challenges of Scaling Loyalty Programs in India

Scaling a loyalty programme in India is genuinely harder than in most markets, and the reasons are structural rather than executional. Start with identity fragmentation. A customer who buys ethnic wear from Manyavar at a mall, grabs a coffee at Cafe Coffee Day in the food court, and picks up prescription refills from Apollo Pharmacy — all in a single two-hour visit — generates three completely separate transaction records with three separate loyalty point balances and three separate consent trails. No single operator has a unified view. The mall operator certainly does not. The result is that even a shopper who visits a mall 18 times a year and spends ₹85,000 annually is treated as a new customer every time she crosses a category boundary.

The second challenge is consent architecture. Most Indian loyalty programmes were built with an 'opt-out by default' consent model — collect data first, ask forgiveness later. DPDP 2023 has fundamentally inverted this. Under the new framework, every purpose of data processing must be explicitly consented to, in the user's language, and that consent must be withdrawable at any time. Existing loyalty databases built without granular purpose-specific consent are, to put it plainly, a liability sitting on the balance sheet. Retailers who have not audited their consent logs in the last 12 months are already exposed.

Third, the technology stack is deeply fragmented. A mid-sized retail chain with 200 stores across three formats might have Wondersoft at the POS in franchise stores, GoFrugal in company-owned stores, a custom-built mobile app, a WhatsApp Business API integration, and a separate email platform — none of which talk to each other in real time. When Capillary or EasyRewardz is bolted on top, they aggregate batch exports from each source, typically daily or weekly. By the time a 'churn risk' segment is identified and a campaign is launched, the customer has already walked into a competitor.

Fourth, there is a unit economics problem at the AI readiness layer. Machine learning models for personalisation require labelled training data at scale — typically a minimum of 50,000 to 100,000 identified transactions per category to produce statistically meaningful propensity scores. Most individual retail brands in India outside the top 20 do not reach this threshold within a single brand's dataset. The only way to reach model-ready data volumes quickly is to pool data across a network — which is precisely what a mall-level AI first party data platform for retail loyalty enables, provided the data governance model is correctly structured with brand-level segregation and shopper-level consent.

AI Automations That Enhance Scale and Accuracy

The vocabulary around AI in loyalty has become polluted with vague claims. Let us be specific about which automations produce measurable lift in an Indian retail context and at what cost thresholds they make sense.

Identity resolution at the POS is the first and highest-ROI automation. When a shopper pays via UPI at a Lifestyle store, their VPA (Virtual Payment Address) is a stable, tokenisable identifier that — with proper consent — can be matched across visits without relying on a loyalty card scan or app login. AI models trained on transaction timing, basket composition, store cluster and payment instrument can achieve a 73–81% match rate for returning customers within 48 hours of transaction, according to benchmarks from mall-scale deployments. This collapses the identification funnel gap between 'transacting' and 'identified' from the industry average of 27 percentage points down to under 10 points for participating brands.

The second high-impact automation is dynamic offer generation. Legacy loyalty platforms push the same ₹200 cashback voucher to every member who crosses a spend threshold. AI-driven offer engines, by contrast, compute an individualised offer value — the minimum incentive required to shift a specific shopper's next visit timing or category expansion — using a reinforcement learning loop calibrated on that shopper's historical price sensitivity. For a sari-and-ethnic brand like Manyavar, this might mean offering a 12% bonus points multiplier to a customer whose transaction history shows she shops primarily in October–December but has not visited since the previous Diwali season. The offer cost per incremental visit drops by 35–45% versus flat voucher programmes.

Third: churn prediction and early intervention. The standard industry definition of a 'churned' loyalty member is someone who has not transacted in 90 days. By that definition, the member is already gone. AI survival models — trained on visit frequency distributions segmented by category, city tier, and household income proxy — can flag at-risk members at day 28 to 45, when a low-cost reactivation (a personalised birthday offer, a 'we missed you' early-access event invite) can recover 19–26% of at-risk members at a cost of ₹35–80 per reactivated member versus ₹400–900 for a new member acquisition.

Fourth, AI-driven tier management replaces the blunt instrument of annual spend thresholds. Static platinum/gold/silver tiers create a well-documented 'loyalty cliff' — a spike in churn immediately after a customer fails to renew a premium tier. Predictive tier models re-score members monthly on a composite of spend trajectory, visit frequency, category breadth and referral activity, and communicate personalised 'path to next tier' nudges that reduce cliff-churn by up to 40% in controlled A/B tests at Indian fashion retail chains.

Data Privacy at Scale: Ensuring DPDP Compliant Loyalty Data Platform Architecture

The Digital Personal Data Protection Act 2023 is not a compliance checkbox exercise. It is a structural redesign requirement for every Indian loyalty programme that collects, processes or transfers personal data. The Act introduces the concept of a 'Data Fiduciary' — an entity that determines the purpose and means of processing personal data. For mall operators and brand retailers running joint loyalty programmes, this creates an immediate question: who is the Data Fiduciary? If Phoenix Marketcity runs a mall-wide loyalty programme and a tenant brand like Lenskart uses that programme's data to run its own retargeting campaigns, there are two Data Fiduciaries involved, each with separate consent obligations.

A DPDP-compliant loyalty data platform must therefore be architected around three non-negotiable principles. First, purpose limitation at data collection: every data point must be collected with a stated purpose (e.g., 'to personalise offers for your next visit to participating stores'), and that purpose must be displayed in plain language in the user's preferred language — Hindi, Tamil, Kannada, or one of the other scheduled languages. This is not optional for apps with significant Tier-2 and Tier-3 city user bases, where English-only consent flows already violate the spirit of DPDP's 'clear and plain language' requirement.

Second, the platform must implement consent state as a first-class data entity — meaning every processing action taken on a member's record must be checked against their current consent state before execution. If a member withdraws consent for 'marketing communications' at 11 PM on a Sunday, the next morning's campaign batch must exclude them without any manual intervention. Legacy CRM platforms that store consent as a binary field in a member table and process it via a weekly suppression list do not meet this standard.

Third, data minimisation must be enforced architecturally, not just as a policy. Loyalty platforms have a well-known tendency to accumulate data indefinitely on the grounds that 'it might be useful later.' Under DPDP, retention must be tied to a stated purpose, and data must be erased when that purpose lapses. An AI platform that stores 36 months of transaction history to train propensity models must be able to demonstrate that the retention period is proportionate and necessary. Automated data lifecycle management — where retention periods are set per data category at onboarding and enforced by the platform's own agents — is the only operationally scalable solution at a deployment covering hundreds of brands and crores of members.

Success Stories From Indian Retail Chains

The most instructive case studies from Indian retail loyalty deployments are not the heroic transformations from zero to hero — they are the disciplined, metric-driven improvements that compound over 12 to 24 months once a genuine AI first party data platform is in place.

Consider the challenge faced by a large multi-brand fashion retailer operating across 180 company-owned stores in South India. Their legacy loyalty programme — built on a rules-based platform — had 28 lakh registered members but a 60-day active rate of just 11%. The top complaint from their CRM team was that they could not tell whether a member who had not scanned their card in 45 days had actually stopped shopping or had simply shifted spend to a competitor format. With no real-time identity signal, they were batching churn interventions weekly, by which time the reactivation window was often closed. After migrating to an AI-native platform with UPI token-based identity resolution, their 60-day active rate rose to 23% within two redemption cycles, and their average cost per reactivated member fell from ₹680 to ₹190.

In the mall segment, the productivity differential between identified and anonymous shoppers is even more pronounced. At a Tier-1 mall running a unified mall loyalty programme, identified members — those with a consented, stitched cross-brand profile — spend an average of 2.8× more per visit than anonymous transactors, visit 3.4× more frequently, and have a 12-month retention rate of 61% versus 19% for anonymous shoppers who happened to be captured by a single-brand loyalty programme. These numbers hold consistently across Phoenix Marketcity properties in Mumbai and Bengaluru, where tenant engagement programmes have been redesigned around a mall-level identity graph rather than individual brand silos.

For a pharmacy chain with 400+ stores across urban and semi-urban India — comparable to the Apollo Pharmacy or MedPlus scale — first-party data activation looks different. The loyalty lever here is prescription refill prediction: an AI model trained on molecule-level dispensing history can predict with 78% accuracy whether a chronic-disease patient will refill on time or lapse. A timely WhatsApp reminder with a personalised refill offer — sent 4 days before predicted lapse, calibrated to the patient's price sensitivity — increases refill compliance and generates incremental footfall without any incremental marketing spend beyond the ₹0.25 per message WhatsApp cost.

How Fundle Solves This

Fundle was architected specifically for the Indian retail context — not adapted from a Western SaaS loyalty platform with an Indian sales office, but built ground-up for the fragmented, multilingual, UPI-native, DPDP-bound market that India's organised retail sector represents in 2024 and beyond. Vineet Narang's founding thesis was direct: Indian retail generates world-class transaction volumes but captures less than 8% of its value-creation potential from customer data because no platform had ever treated first-party data consent, AI activation, and loyalty mechanics as a single integrated system.

The Fundle AI Platform unifies four capabilities that have historically been separate vendor relationships: a first-party data platform (identity graph, consent management, data clean room), a loyalty engine (points, tiers, rewards catalogue, gamification), a campaign orchestration layer (cross-channel journey builder with real-time trigger support), and AI agents that operate continuously without requiring a campaign manager to write rules. This architecture is the foundation of both Fundle Mall Loyalty — the product deployed across 123+ malls, now serving over 1.33 crore users — and Fundle Brand Loyalty, which serves individual enterprise retail brands operating both within and outside the mall format.

Fundle AI Agents are the operational core that differentiates the platform from CRM-plus-AI retrofits offered by legacy vendors. These agents handle specific, measurable tasks autonomously: the Churn Intervention Agent monitors visit cadence decay and dispatches recovery journeys at day 28 of inactivity without human approval. The Offer Optimisation Agent runs continuous multi-armed bandit tests across offer variants and re-allocates budget to the highest-performing variant in real time, not in the next campaign cycle. The Consent Compliance Agent monitors DPDP consent state for every processing action, maintains an immutable audit trail, and flags any campaign that would process data beyond its consented purpose before execution — not after a regulator asks.

Fundle AI Workflow connects these agents to the broader retail technology stack through pre-built integrations with the POS vendors Indian retailers already use: Petpooja, POSist, GoFrugal, and Wondersoft. This means a retailer does not need to replace its existing store technology to activate Fundle's intelligence layer. The platform ingests event streams, enriches them in the identity graph, scores them against AI models, and returns activation signals to the originating POS or communication channel within seconds. For the Indian retail CMO evaluating Fundle Agentic AI against point solutions from Capillary, Antavo, Almonds.ai or Customer Capital, the relevant question is not features — it is whether the vendor has built for India's specific data governance, language, and infrastructure reality from day one. Fundle has.