Seven Billion AnalyticsContact Us
Client logo

Retail & CG

AI SKU Recommender for a Snacks FMCG Maker

AI SKU Recommender for a Snacks FMCG Maker

15%

15%

Increase in distributor SKU penetration

>75%

>75%

Precision of the next-best-SKU model

₹32 Cr

White-space revenue pipeline identified

Seven Billion helped us build and deploy an AI system in weeks that solved our key retail distribution bottlenecks.

Ankur Gupta

Head IT

DFM Foods (Crax)

Client logo

About

A snacks FMCG maker lifted distributor SKU penetration 15% with a next-best-SKU model above 75% precision and found a ₹32 crore white-space pipeline.

Industry

Retail & CG

Company size

1,000 – 5,000 employees

Founded

1985

Business Context

Business context and operating model

A snacks portfolio with more distributors than visibility The client, an Indian snack food manufacturer, manufactures and distributes its snack portfolio through a network spanning modern trade, general trade, and regional wholesale channels. Two years of sell-in and sell-through data existed across the business, but it lived in fourteen regional systems that never spoke to each other. With no unified view, every distributor was sold the same core range regardless of what their local outlet mix, income band, or competitive set could actually move. Fast movers ran out in strong markets while slow movers piled up in others — a pattern worth an estimated ₹30 crore a year in mismatched stock.

Problem Definition & Framework

Current State

Failure Modes

Decision Criteria

KPI Baseline

Problem definition and framework

Reps could only sell what they already knew Sales reps default to habit. Without a systematic way to surface what a specific distributor was under-buying relative to comparable distributors nearby, the sales conversation stayed anchored to the same five or six hero SKUs on every visit. The result was two compounding problems: stock-outs on fast-moving SKUs in markets that could have sold more, and dead stock on slower SKUs sitting in warehouses that never had the demand profile to justify the allocation in the first place. Neither the sales team nor category management had the data infrastructure to see this before it showed up as a P&L variance.

  • Decision framed: Which SKU should each distributor be offered next, based on outlet mix, category affinity, seasonality, and peer behavior?

  • Framework used: Distributor-SKU affinity modelling and white-space segmentation

  • Baseline mapped: current process, failure modes, handoffs, data owners, and decision latency.

  • Success criteria tied to business KPIs, adoption behavior, and operating cadence changes.

How it works

Before / After

var(--variable-hWXsjrXo2)

Architecture

var(--variable-bl988V6Ac)

Process Flow

Data Pipeline

Results

var(--variable-hSfiUIv6W)

Delivery Timeline

Engagement Method & Process Diagram

01 Discovery

02 Data Audit

03 Baseline

04 Prototype

05 Pilot

06 Production

07 KPI Monitoring

Engagement method and process diagram

  • Process diagram: Regional sales systems -> unified distributor/SKU schema -> embedding model -> next-best-SKU ranking -> Power BI field workflow -> adoption and precision tracking

  • Engagement method: Data audit across 14 regional silos; distributor/SKU feature design; model backtesting; field pilot with sales reps; boardroom analytics rollout; weekly precision and adoption review.

  • Delivery pattern: discovery, data audit, baseline measurement, prototype, pilot workflow, production rollout, and KPI monitoring.

An embedding model that profiles every distributor and product We unified two years of sales data across the fourteen silos into a single schema, then trained a deep-learning embedding model that represents every distributor and every SKU in the same latent space — capturing outlet mix, price band, seasonality, and category affinity. Each morning the model scores every distributor-SKU pair and serves a ranked next-best-SKU list through a Power BI app connected directly to SAP HANA, so reps walk into every call with a specific, data-backed recommendation rather than a generic order sheet. On top of the recommender we deployed a natural-language analytics layer so brand and finance teams could ask direct questions — "why is our lead brand underperforming in the Northeast this quarter" — and get an answer with root-cause drivers and competitive context in under thirty seconds, work that previously took an analyst three to five days of manual data pulls.

Architecture, Results & Measurement

Solution architecture, results, and measurement

15-point penetration gain and a ₹32 crore pipeline Distributor SKU penetration rose 15% within two quarters, with the largest gains among mid-tier distributors who had never been shown their own white space before. The recommender's precision held above 75% in live testing, and field adoption reached 93% — reps kept using it because the recommendations were specific enough to act on, not generic category-level guidance. The unified data layer surfaced ₹32 crore in previously invisible white-space revenue across the distributor network. On the analytics side, brand-performance queries that used to take three to five days of analyst time now return in under thirty seconds, freeing the category team to spend their time acting on root causes instead of assembling the data to find them.

  • Measured outcome: 15%: Increase in distributor SKU penetration

  • Measured outcome: >75%: Precision of the next-best-SKU model

  • Measured outcome: ₹32 Cr: White-space revenue pipeline identified

  • Measurement loop: compare weekly operating behavior against baseline, surface exceptions, and review model/workflow drift with business owners.

The recommender does not replace our reps' judgment — it removes the guesswork before the conversation even starts. They walk in already knowing what that distributor is missing.

Head of IT, snacks FMCG maker

KEEP READING

See How Other Teams are

Winning with Seven Billion

See How Other Teams are

Winning with Seven Billion

Explore more studies on AI, analytics, and enterprise intelligence.

The lowest-risk way to find out if AI is right for your business.

Phase 0 is a two-week discovery that tells you exactly which problems AI can solve, what it will take to build, and what it will cost. No obligation beyond it.

ABOUT Seven Billion

Seven Billion is an Applied AI company — a team of data scientists and AI engineers who build and deploy AI systems that run in production.

OFFICE

World Tech 83, First Floor, IT City, Mohali 140603

Intelligence that delivers starts here.

Whether you are mapping your first AI use case or scaling AI across the enterprise, we will help you cut through the noise and build something that actually ships.

ABOUT Seven Billion

Seven Billion is an Applied AI company — a team of data scientists and AI engineers who build and deploy AI systems that run in production.

OFFICE

World Tech 83, First Floor, IT City, Mohali 140603

The lowest-risk way to find out if AI is right for your business.

Whether you are mapping your first AI use case or scaling AI across the enterprise, we will help you cut through the noise and build something that actually ships.

ABOUT Seven Billion

Seven Billion is an Applied AI company — a team of data scientists and AI engineers who build and deploy AI systems that run in production.

OFFICE

World Tech 83, First Floor, IT City, Mohali 140603