How Fynd helped Reliance Smart Bazaar cut forecasting error from 55% to 41%

A stock target set at the state level treats every store the same - a busy Mumbai street corner shop and one two towns away both told to hold the same stock, as if their shelves and customers were the same. They are not and that gap caused forecasting errors over 55%.

Granary replaces one-size-fits-all targets with store-specific numbers. Already live in an 11-store Mumbai pilot, it’s cutting errors nearly in half.

The challenge: One number, stretched across a thousand stores

Grocery is tough to plan for. Margins are slim, many items perish, and chains like Smart Bazaar and Smart Point manage about end numbers of SKUs across thousands of stores - a huge planning challenge no manual process can handle well.

One number, applied everywhere

Stock targets were set at the state level, ignoring each store’s unique shelves and customers.

Forecasting error above 55%

State-level targets did not match store realities, causing big errors.

No consistent reasoning behind decisions

What to keep, discount or remove was decided manually, with no clear logic.

No system to catch outliers early

Thousands of SKUs need early alerts for zero sales, negative inventory or high days-on-hand to avoid waste.

The solution: A number built for the store it's actually feeding

Granary is a smart planning and assortment platform made for groceries. It handles forecasting, restocking, and category optimization for Smart Bazaar and Smart Point. Live in an 11-store Mumbai pilot across ten product segments, its forecasting engine has already cut prediction errors nearly in half.

ML forecasting engine

Forecasts 12,000 SKUs in 4,000 stores from a massive daily dataset, lowering errors from 55%+ to 41%.

Store-level targets

Stock goals are set per store automatically, replacing the old state-level approach.

Every removal explained

A nine-reason system like inventory costs or quality issues makes delisting decisions clear and auditable.

The platform: From raw data to a decision on the shelf

Granary works in layers, from raw data to smart decisions:

Data & ML

Sales, inventory, supplier and loyalty data feed ML pipelines refreshed daily.

Agentic layer (Cortex)

The decision engine for assortment, forecasting, rules and restocking.

Command center

Dashboards and alerts for category managers, buyers and store teams, with audit trails for decisions.

3D store twin roadmap

A digital copy of the store to test assortment or layout changes before applying them.

The impact

• Forecasting error cut from 55%+ to 41%, roughly halved in an 11-store pilot.

• 12,000 SKUs forecast daily across 4,000 stores from a 48-million-row dataset.

• Live in 11 Mumbai stores across ten grocery and personal care segments.

• A nine-reason classification system for delisting replaces guesswork with audit trails.

A number built for the store it's actually feeding

Stock targets stretched across every store in a state never fit a busy street shop with different customers than a store two towns away. Granary replaced this guesswork with store-specific targets and gave managers clear reasons for every decision. Forecasting error is down nearly a third and the platform is just beginning, with a 3D store twin coming soon to test shelf changes before they happen.

Still planning stock at the state level, not the store level?

Book a demo with us today