Fundle.ai: AI-Powered Customer Loyalty Agents: A Fundle Case Study

Client Profile: Indian Retail Brand Running at Scale

The brand in this case study is a multi-category lifestyle retailer operating 60+ stores across Tier-1 and Tier-2 cities — think of the profile of a mid-sized Lifestyle International or a growing Pantaloons franchisee operating under its own regional brand equity. Annual GMV at the time of engagement was approximately ₹420 crore. The loyalty database stood at 1.4 million registered members, of whom roughly 310,000 had transacted at least once in the trailing 12 months — a 22% active ratio that is, depressingly, above the Indian retail median.

The CRM team consisted of four people: a CRM manager, two campaign executives, and a data analyst. Their martech stack was a combination of a legacy points engine, a WhatsApp Business API aggregator, and email via a mid-tier ESP. Reporting was manual, pulled weekly from the POS system — a Wondersoft deployment — into Excel. There was no real-time visibility into customer journeys, no predictive churn scoring, and no mechanism to personalize offers at an individual level. Segments were built monthly, campaigns were fired fortnightly, and results were reviewed the following month. The feedback loop was effectively six weeks long.

The retail environment this brand operates in is acutely competitive. Within a 5-km radius of its top-performing store, it faces direct competition from Reliance Trends, FabIndia, and a Manyavar franchise. All three operate loyalty programs. All three have deeper technology budgets. The brand's competitive moat had historically been its staff relationships and its curated regional merchandise — neither of which a WhatsApp OTP journey can communicate adequately.

This is the profile of thousands of Indian retailers: meaningful scale, constrained CRM resources, a loyalty database that is more liability than asset, and a growing awareness that the next phase of growth must come from existing customers, not just footfall acquisition. It was this recognition that brought them to the Fundle AI Platform.

Challenges Faced Before AI Loyalty Agents Entered the Picture

The brand's CRM head articulated the problem with unusual clarity during onboarding: 'We have a loyalty program that rewards past behavior. We need one that changes future behavior.' That single sentence captures the structural flaw in most Indian retail loyalty deployments.

Challenge one was the segment-campaign-report treadmill. The team was spending 60% of its working hours building audience segments, QA-ing campaign logic, and writing post-campaign reports. This left almost no bandwidth for strategy, creative experimentation, or — critically — responding to intra-week customer signals. A customer who purchased ethnic wear on a Tuesday and browsed occasion jewelry on Thursday was getting the same 'double points on footwear' push as every other member in her tier. The relevance gap was enormous.

Challenge two was churn blindness. Without a predictive model, the team had no mechanism to identify which of their 310,000 active members were drifting toward inactivity. They were running a 'win-back' campaign once a quarter targeted at anyone who hadn't transacted in 90 days — by which point, in Indian retail, the customer has already formed a new habit elsewhere. Industry data from Capillary's Indian retail benchmarks suggests the optimal intervention window for a drifting apparel customer is between days 28 and 42 post-last-purchase. The brand was intervening at day 90+.

Challenge three was offer economics. Because personalization was impossible at scale, the brand defaulted to uniform discount offers — typically 10-15% off vouchers — to drive re-engagement. The margin erosion was significant: approximately ₹1.8 crore per year in discount redemptions from customers who self-reported they would have repurchased anyway. This 'always-on discount' pattern also trained the customer base to wait for offers before transacting, suppressing full-price purchases.

Challenge four was channel chaos. Customers were receiving WhatsApp messages from the brand's aggregator, email from the ESP, and push notifications from a separate app — often carrying inconsistent offer values for the same promotion. There was no orchestration layer. One customer received a 10% offer via email and a 15% offer via WhatsApp for the same SKU category within 48 hours. The brand only discovered this because the customer complained at the store counter.

Implementation of Fundle's AI Loyalty Platform: The Five-Week Sprint

The implementation of the Fundle AI Platform was structured as a five-week activation sprint, not a six-month IT project. This distinction matters enormously in Indian retail, where seasonal calendars — Navratri, Diwali, end-of-season sales — mean a loyalty platform that isn't live in eight weeks is a loyalty platform that misses the year's highest-ROI windows.

Week one focused entirely on data plumbing. The Fundle team connected to the brand's Wondersoft POS via a pre-built API connector, ingested 36 months of transaction history, and constructed the initial customer loyalty graph. This graph maps not just purchase frequency and average order value but also category transition patterns — for example, identifying that a customer who buys ethnic occasionwear in Q4 has a 67% probability of purchasing accessories within 45 days if prompted correctly.

Week two was the RFM and churn model calibration. Fundle AI Agents ingested the historical data and generated a live churn probability score for each of the 310,000 active members. The output was immediately actionable: 47,000 members were flagged as 'at-risk' (churn probability >65% within 30 days), and 12,000 were flagged as 'high-value at-risk' — customers in the top two spend deciles showing early disengagement signals. These 12,000 became the first cohort for the AI agent's re-engagement workflow.

Weeks three and four involved deploying Fundle AI Workflow across three journeys simultaneously: a churn-prevention journey for the at-risk cohort, a tier-upgrade nudge journey for members within 800 points of the next tier, and a post-purchase cross-sell journey for customers who had made their first purchase in a new category. Each journey was configured with guardrails — maximum discount depth, channel frequency caps, and escalation rules that routed high-value complaints to a human CRM agent within four hours.

Week five was go-live and hypercare. The Fundle Agentic AI system went into autonomous operation, with the CRM team shifting from campaign builders to journey reviewers. Their weekly workflow changed from 'build this week's campaign' to 'review which journeys the AI optimized, approve any new journey variants it proposed, and monitor guardrail breach alerts.' The four-person team recovered approximately 14 hours per week of strategic capacity.

Results and Key Performance Improvements After Six Months

A partner brand saw 25% uplift in customer retention within 6 months of adopting Fundle AI loyalty agents. That headline number, striking as it is, understates the breadth of improvement across the program's KPI stack.

Retention, measured as the percentage of active members who transacted at least once in a rolling 90-day window, moved from 22% to 27.5% — the 25% relative uplift. In absolute terms, that is approximately 17,000 additional customers retained per quarter. At an average transaction value of ₹2,400 per visit and an average of 2.1 visits per retained customer per quarter, that translates to roughly ₹8.6 crore in incremental quarterly revenue directly attributable to the AI loyalty agents.

Average transaction value for members engaged by the AI-driven cross-sell journey increased by 18%, from ₹2,200 to ₹2,596. This was not driven by discounting — the AI calibrated offers at an average of 6.2% off, compared to the brand's previous blunt-force 12-15% discount campaigns. The improvement came from relevance: customers receiving a cross-sell nudge for a category they had already browsed converted at 3.4x the rate of customers receiving a generic offer.

Churn among the 'high-value at-risk' cohort of 12,000 customers dropped from a projected 68% to an actual 41% — a 27-percentage-point save rate. Given that each high-value customer generates approximately ₹18,000 in annual revenue, saving 3,240 customers from churn represents a retained revenue pool of approximately ₹5.8 crore annually. The cost of the AI-driven intervention for these customers — primarily WhatsApp messages and personalized tier-upgrade offers — was under ₹14 lakh for the six-month period.

Operationally, the CRM team's campaign build time dropped by 65%. Before Fundle, the team was spending roughly 22 hours per week on campaign construction and reporting. After the Fundle AI Workflow deployment, that dropped to 8 hours — with the reclaimed time redirected to journey strategy, creative development, and store-level CRM activation. Three months into autonomous operation, the team had launched seven new journey variants, more than they had managed in the previous 18 months combined.

How Fundle Solves This

The architecture behind the results described in this case study is not a black box — it is a deliberate design philosophy embedded across the entire Fundle AI Platform. Vineet Narang's founding thesis was that Indian retail CRM teams are not failing because of lack of effort; they are failing because they are using deterministic tools to solve a probabilistic problem. Customer behavior is not a decision tree. It is a dynamic, context-sensitive signal stream. The technology managing it must be equally dynamic.

Fundle Loyalty is the foundational layer: a real-time points, tiers, and rewards engine that connects to any POS system in India's fragmented retail tech landscape — Wondersoft, GoFrugal, POSist, Petpooja, and custom-built enterprise stacks alike. Unlike legacy platforms where tier upgrades run in nightly batch jobs, Fundle Loyalty processes transactions in under 200 milliseconds, ensuring that the moment a customer crosses a tier threshold, every downstream system — app, WhatsApp, store associate interface — reflects it instantly.

Fundle AI Agents sit on top of this foundation as the autonomous decision-making layer. Each agent is a purpose-built AI workflow configured for a specific loyalty job: the Churn Prevention Agent monitors RFM drift and fires the minimum-effective intervention at the optimal window. The Cross-Sell Agent tracks category transition probabilities and triggers personalized bundle or upgrade offers with AI-calibrated offer values. The Tier Nudge Agent monitors members approaching tier boundaries and deploys gamified point-gap messaging without requiring any campaign manager input. These are not simple if-then automations — they are Fundle AI Workflow orchestrations that reason across customer history, brand business rules, and real-time behavioral signals simultaneously.

For mall operators managing multi-brand loyalty programs — whether at a Phoenix Marketcity or a regional mall in Pune — Fundle Mall Loyalty adds a cross-brand intelligence layer. A customer who spends ₹3,200 at Manyavar and ₹1,800 at an anchor food court within the same mall visit generates a composite behavioral signal that individual brand CRM systems would never see. Fundle's mall graph surfaces these cross-brand patterns and enables the mall's marketing director to run program-level retention interventions that individual brand loyalty programs structurally cannot. Fundle Brand Loyalty, in parallel, gives each tenant brand within the mall its own segmentation and campaign control — governed by guardrails set at the mall program level — so there is no channel conflict between brands competing for the same customer's attention on the same afternoon.

The Fundle Agentic AI system's most operationally significant feature, as the case study brand discovered, is the weekly 'decisions digest' — a plain-language audit trail of every autonomous action the agents took, why they took it, and what the outcome was. This transparency is not cosmetic. It is the mechanism by which CRM teams build trust in AI recommendations, identify edge cases where the model needs recalibration, and maintain regulatory accountability in an environment where India's DPDP Act is beginning to impose meaningful data governance requirements on loyalty program operators.