AI for FMCG Distribution — Demand Forecasting, Route Optimisation, Image Recognition

The thesis — productise the data you already have

Every SalesPort client has been generating structured distribution data for 12-60+ months — every order, every dispatch, every retailer visit, every scheme applied, every payment collected, every GPS ping from a salesperson's phone. Across 45 deployments that's 49 Lakh+ orders, 21.64 Crore GPS data points, 17.43 Lakh schemes, ₹2,677 Crore of payment flows. Most of this data has been used for reporting — operational dashboards, monthly MIS, exception alerts.

The thesis behind SalesPort AI is simple: that same data trains excellent demand-forecasting models, route-optimisation solvers, image-recognition shelf auditors, and scheme-ROI engines. The AI doesn't require new data collection. The AI productises what we've been collecting all along.

For existing SalesPort clients, the AI modules are upsells. Pawanshree Dairy's MD said it directly when we ran the first prototype against his data: "We will love to pay extra for this." That conversation became the genesis of this module suite.

Seven AI modules — what each does, what each costs

The Pawanshree wedge

“We will love to pay extra for this.”

Pawanshree is the largest milk procurement deployment on SalesPort — 79,512 farmers, 1,797 VLCs, 140 collection routes, 3.28 Crore collection records, ₹646 Crore of annual procurement. The operational data depth is unusual: 60+ months of per-route per-shift per-SKU collection and distribution history.

We ran a demand-forecasting prototype on six months of their data — XGBoost trained on SKU + route + day-of-week + weather + holiday-calendar features, predicting next-day SKU-level dispatch volume. The model beat their previous manual forecast by 22% on MAPE for fast-moving SKUs. The Pawanshree dashboard — the live mobile view their MD uses to see procurement and dispatch in real-time — added forecasting deltas alongside the actuals.

The MD's response was the single sentence above. That moment converted "we have data" from an internal capability into a productised AI module set. The same pattern — productise the data → upsell to existing clients — is the wedge for all 45 SalesPort deployments.

What 5 years of distribution data actually looks like

The data depth behind the AI modules:

• Orders: 49 Lakh+ orders across 1.96 Crore order line items — SKU + retailer + distributor + scheme + price + quantity per line
• Dispatches: 11.44 Lakh dispatches — vehicle + driver + route + GPS trace + delivery acknowledgment
• GPS data: 21.64 Crore GPS data points from 132+ Flutter apps across 2.3 Lakh daily active users — high-resolution movement traces
• Schemes: 17.43 Lakh schemes auto-applied — slab tier + claim window + retailer + SKU + price impact
• Payments: ₹2,677 Crore collected across distributor wallets, retailer collections, advance receipts — full payment-cycle history
• Procurement: ₹803 Crore of milk procurement across 83,785 farmer accounts — fat/SNF readings, rate slabs, advance deductions, settlement cycles
• Returns: Full returns history — distributor returns, retailer returns, expired pharma stock, damage claims
• Beat compliance: 2.53 Crore field activities logged — beat-plan adherence, missed visits, retailer-side feedback

For each client, the data is per-client isolated (no cross-client model mixing). Models train on a single client's data and serve that client only. That's both an architectural choice (compliance) and a quality choice (FMCG-spice patterns don't predict pharma-cold- chain patterns; per-client models do better than pooled).

Buy vs build — when standalone AI vendors make sense

Standalone AI-for-FMCG vendors (Aforza, Wiz.ai, RELEX, FieldAssist IRIS, o9 Solutions) are legitimate options. They're better than us at three things: model sophistication (research teams larger than ours), cross-company benchmarks (they aggregate patterns across many clients), and standalone deployability (you don't need to be on their core platform to use the AI).

The tradeoff is cost and integration. Standalone enterprise AI runs ₹3-15 Lakh/month per module, plus integration projects that take 6-12 months to land. For top-100 Indian FMCG brands with dedicated data-science teams, this is the right path.

For the SalesPort client segment — mid-market FMCG, regional dairies, agri brands, government cooperatives — the math is different. The data is already flowing through SalesPort. The modules ride on existing infrastructure at +₹8K-40K/month. Time-to-value is weeks, not quarters. The economics work because we're not building a new product; we're productising data the client already owns.

Roadmap — what ships when

• Live now (Q2 2026): Demand Forecasting, Live Sales Analytics Dashboard. Available to all 45 existing SalesPort clients.
• Q3 2026: Route Optimisation, Trade Promotion ROI Engine, WhatsApp Order Bot.
• Q4 2026: Image Recognition (Perfect Store) with starter SKU catalogs for dairy and FMCG.
• Q1 2027: Distributor Credit Scoring with 24-month default-prediction horizon.
• Roadmap candidate: Multi-echelon inventory optimisation, retailer-app AI assistant, predictive maintenance for distributor fleets.

Frequently asked questions