Scaling D2C revenue 3× with predictive inventory and AI-led marketing

Snicsnac had built genuine product-market fit for their flagship range — consumer demand was there. The operational problem was that their inventory planning was entirely reactive. Buyers placed purchase orders based on gut feel and the previous month's sell-through data, with no model for seasonality, promotional uplift, or competitive events. The result was a 40% stockout rate on their top 10 SKUs during peak seasons, including Diwali 2023, when they ran out of their bestselling Masala Makhana SKU within six hours of a promotional push — losing an estimated ₹18 lakh in that single window.

On the demand side, wastage on slow-moving SKUs was running at 20% — product being written off at cost every quarter. Buyers had no signal on which products were losing momentum until it was too late to reroute purchase orders. The gap between supply decisions and consumption data was 30-45 days due to manual reporting cycles from distribution partners.

Their digital marketing was equally uncoordinated. Meta and Google ad accounts were running on manual bidding with no conversion tracking beyond last click. The marketing team had no visibility into which channels drove actual repeat purchases versus one-time buyers. Three different agencies had managed various parts of the account over two years, leaving behind fragmented campaign structures, inconsistent naming conventions, and no unified attribution model.

We started with a full data audit — connecting Shopify, Meta Ads, Google Ads, Amazon Seller Central, and their warehouse management system into a unified first-party customer data platform built on Segment and BigQuery. For the first time, Snicsnac had a single view of a customer from first touchpoint through to lifetime purchase value, segmented by acquisition channel, SKU preference, and purchase frequency.

The demand forecasting model was built on an LSTM architecture trained on 18 months of order history. We enriched the training data with external signals: weather patterns (relevant for seasonal snack preferences), regional festival calendars, competitor pricing data scraped weekly, and influencer content scheduling. The model was validated against held-out data from the previous festive season, achieving a mean absolute percentage error of 8.3% at SKU-week level — compared to the 35%+ error in the existing manual process.

The marketing rebuild started with server-side conversion tracking via Facebook CAPI and Google Enhanced Conversions, recovering an estimated 30-40% of conversion signals that iOS 14+ tracking restrictions had eliminated. Ad accounts were restructured with value-based bidding strategies seeded with first-party purchase value data. Audience strategy was rebuilt around customer lifetime value tiers rather than demographic proxies.

The demand forecasting model now runs every Sunday night, generating SKU-level reorder recommendations for the coming four weeks across every distribution channel. Buyers receive a structured report with recommended purchase quantities, flagged stockout risks, and SKUs showing deceleration signals that suggest reducing forward orders. The model's recommendations have been followed by the buying team for six consecutive months, with stockout incidents falling 67% versus the prior year period.

On the marketing side, the rebuilt account structure with value-based bidding achieved a 2.4× improvement in blended ROAS within the first eight weeks. The custom attribution model — giving weighted credit across touchpoints based on incrementality analysis rather than last click — revealed that branded search was converting customers originally acquired by mid-funnel content, shifting budget allocation significantly toward content and Meta prospecting.

A loyalty and retention programme was built on top of the CDP, segmenting customers into high-value, at-risk, and lapsed cohorts and triggering personalised campaigns via email, WhatsApp, and SMS. The programme focused acquisition spend on high-LTV customer profiles identified from the first-party data. Taken together, these interventions delivered 3× D2C revenue within six months — while blended customer acquisition cost fell 19%.