Consumer & Retail

Build a stronger business as your brand grows.

Give commercial and operations teams a clearer view of demand, inventory and profitability across products and channels.

Opportunity landscape

Where AI can help.

Consumer businesses make thousands of small commercial decisions: what to buy, where to hold it, how to price and promote it, and how to serve the customer afterwards. AI helps where those decisions depend on patterns across products, locations and channels that are hard to hold in a spreadsheet.

Priority applicationWhat it improves
Demand forecasting and replenishmentWhat to buy and when, by product, location and horizon.
Inventory and allocation optimisationWhere stock sits across warehouses, stores and channels, and how to rebalance it.
Product and channel profitabilityWhich products and channels earn their cost once returns, fulfilment and discounts are included.

These are examples of work that can be scoped. The right starting point depends on your data, systems and priorities.

Priority applications

What we can build, in detail.

01 / 03

Demand forecasting and replenishment

Business impact

Improves purchasing, replenishment and allocation decisions, reducing both stockouts and excess stock.

Suitable situations

Businesses with many SKUs across locations or channels, seasonal or promotional demand, and long or variable supplier lead times.

Full application detail
What we build
Forecasts by SKU, location and planning horizon, with uncertainty ranges, an exceptions view for planners, and replenishment recommendations delivered into planning or ERP workflows.
Required inputs
Sales, stock availability, promotions, prices, returns and lead times at SKU and location level, with enough history to see seasonality, plus product attributes for new items.
Technical approach
Hierarchical forecasting that keeps SKU, location and category forecasts consistent. Methods for intermittent demand and for new products with limited history. Correction for stockouts, which hide true demand in observed sales. Rolling backtests against baseline methods.
Workflow integration
Forecasts and recommended quantities flow into the planning tool or ERP. Planners review exceptions and override with a recorded reason.
Human responsibility
Planners own the plan and approve purchase quantities. Commercial teams own promotions and pricing inputs. Overrides are tracked and reviewed.
Measurement
Forecast bias and error against a baseline, alongside inventory levels, stockouts, service levels and working capital.
Limitations
Short history, volatile demand, unrecorded promotions and missing stock data limit accuracy. Forecast uncertainty is shown, not hidden.
Adjacent opportunities
Inventory and allocation optimisation, which depends on the forecast. Profitability analysis.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

02 / 03

Inventory and allocation optimisation

Business impact

Decides how much stock to hold, where, and when to move it, to balance availability against cost and cash.

Suitable situations

Multi-location or multi-channel businesses where stock sits in the wrong place, or where transfers and markdowns are decided by hand.

Full application detail
What we build
An optimisation service that turns forecasts and constraints into stocking, transfer and allocation recommendations, with scenario comparison.
Required inputs
Forecasts, current stock by location, lead times, costs of holding, transfer and stockout, minimum order quantities and capacity limits.
Technical approach
Simulation of demand and lead-time uncertainty, constrained optimisation, and comparison of policies against current practice on historical data.
Workflow integration
Recommendations are queued for planners and written to the purchasing or transfer systems after approval.
Human responsibility
Planners approve transfers and buys. Finance owns working-capital targets and the cost assumptions.
Measurement
Availability and fill rate, inventory days, markdown and write-off levels, working capital, and the gap to a baseline policy in backtests.
Limitations
Results depend on trusted cost inputs and stock accuracy. Operational limits on transfers can cap the benefit.
Adjacent opportunities
Demand forecasting as the input. Logistics and supply chain applications for the movement itself.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

03 / 03

Product and channel profitability

Business impact

Shows which products, channels and customers earn their cost once returns, fulfilment, payment fees and discounts are included.

Suitable situations

Brands selling through several marketplaces, stores and direct channels, where margin is reported at revenue level only.

Full application detail
What we build
A reconciled data model across orders, payments, returns and costs, margin views by product and channel, and an assistant that answers margin questions with the underlying records.
Required inputs
Orders, payments and settlements, returns, fulfilment and marketplace fees, product costs, discounts and promotions.
Technical approach
Entity resolution across channels, reconciliation of orders to payments, cost allocation rules agreed with finance, and checks that totals tie to the ledger.
Workflow integration
Views update on a schedule and are available to commercial and finance teams. Exceptions in reconciliation are routed to finance.
Human responsibility
Finance owns allocation rules and sign-off. Commercial teams own decisions on range, pricing and channel mix.
Measurement
Reconciliation coverage and match rate, time to produce margin reporting, and decisions changed as a result.
Limitations
Allocation of shared costs involves judgement and must be agreed. Gaps in channel data limit coverage.
Adjacent opportunities
Forecasting and inventory decisions that use margin as an input.

Next action

Take the assessment for a first view of where this fits, or contact us to discuss a specific case.

Published deployments

What others have reported.

These are results published by other organisations, shown as their sources report them, with what each figure measures. They are not TJC Labs results and are not a forecast of what a project will deliver. A prospect’s own business case is estimated from its own data.

E-commerce, India

Flipkart

Customer-reported
Problem
Supply chain planning that was quasi-manual and loosely coupled, with teams optimising different objectives as complexity and demand grew.
What was implemented
A forecasting layer, capacity planning and inventory rebalancing. The production forecast is an LSTM sequence model with attention, trained with quantile loss and reconciled across granularities. Capacity planning uses mixed-integer and stochastic programming. Inventory rebalancing is a large mixed-integer program of about 10 million decision variables, solved with a local search heuristic.
Technical approach
Forecasting moved from moving averages through ARIMA and exponential smoothing and an ARIMA and gradient-boosted tree ensemble to deep learning.
Reported result
Weighted MAPE fell from 77.4% to 70.6% in Home, 65.2% to 59.1% in Electronics and 59.8% to 53.3% in Books and General Merchandise, against the earlier ARIMA and tree ensemble. The authors also report a 50% increase in one-day deliveries, a 50% reduction in unhealthy inventory, and a 10% increase in manpower utilisation.
Scope, baseline and period
Forecast test: three business units, over 100,000 SKUs, three years of daily data, a 30-day horizon. The baseline period for the operational figures is not stated.

Reported by: Customer-reported; peer-reviewed in INFORMS Journal on Applied Analytics. Source: INFORMS Journal on Applied Analytics 56(1) (opens in a new tab), 2026 (accepted 15 October 2025).

Fashion e-commerce, Europe

Zalando

Customer-reported
Problem
Deciding what stock partner merchants should hold, when to reorder and where to allocate it across a multi-country warehouse network, under uncertain demand and lead times.
What was implemented
A replenishment engine with weekly 12-week probabilistic demand forecasts from LightGBM quantile models, and an optimiser that uses Monte Carlo simulation, with an extended (R, s, Q) policy.
Technical approach
Optimises the 75th percentile of total cost rather than the mean. The weekly forecasting pipeline processes three years of history for 5 million SKUs in under two hours.
Reported result
In a backtest, GMV was 22.11% higher than human replenishment decisions, and GMV after fulfilment costs 21.95% higher. About 70 to 80% of merchants saw a positive uplift.
Scope, baseline and period
A computational backtest, not a live test: October 2023 to September 2024, about 2 million articles from about 800 merchants, assuming 100% adoption. Merchants keep final authority.

Reported by: Customer-reported; paper in Scientific Reports. Source: Zalando Engineering (opens in a new tab), 15 January 2026.

Grocery and general merchandise retail, United States

Walmart

Company-reported
Problem
A legacy forecasting system based on exponential smoothing that copied prior-year seasonal patterns and ignored factors such as promotions, weather and local demand.
What was implemented
Machine learning forecasting platform with separate model tracks for fast-moving items, long-tail items, event seasonality and new items, and a planner application that tracks forecast accuracy and manual overrides.
Technical approach
Gradient boosting, state space models, random forests and hierarchical techniques, evaluated by backtesting over 52 weeks.
Reported result
About 300 basis points of accuracy improvement in the backtest, with live results in the US launch described as in line. About 500 basis points in international markets.
Scope, baseline and period
Launched for Meat and Produce in June 2018, later extended. Baseline accuracy levels and waste figures are not given.

Reported by: Company-reported. Source: Walmart Global Tech (opens in a new tab), 8 October 2020.

More published examples

Grocery e-commerce

Instacart

An LLM pipeline that generates personalised discovery pages. Retrieval-augmented generation cut generation cost by 15 to 20% against a single-model setup, and a fine-tuned evaluator reduced evaluation cost by over 99% against closed-weight model inference. The post reports that A/B tests were promising and gives no engagement figures.

Reported by: Company-reported. Source: Instacart (opens in a new tab), 26 February 2026.

Quick commerce, India

Zepto

Zepto describes machine learning for warehouse slotting and picker routing, micro-market demand forecasting and last-mile routing. It reports that more than 50% of order volume goes through highly automated distribution centres and that average packing time is about 60 seconds. No baseline is given.

Reported by: Company statements, relayed by Digit. Source: Digit (opens in a new tab), Undated; 2026.

Technical implementation

What shapes the build.

  • Forecast structure. Forecasts must agree across SKU, location and category, and handle intermittent items and new products.
  • Demand under stockouts. Observed sales understate demand when stock runs out, so the model corrects for availability.
  • Integration with planning and ERP. Recommendations are written to purchasing and transfer workflows, with overrides recorded.
  • Evaluation design. Rolling backtests against baseline methods, reported by category and by item type, not only in aggregate.
  • Cadence and compute. Refresh frequency and runtime are sized to the planning cycle and the SKU count.

Readiness and data requirements

What helps you start.

  • Sales history at SKU and location level, with a record of when items were out of stock.
  • Promotion, price, return and lead-time data in a form that can be joined to sales.
  • A single product master, or a mapping between channel catalogues.
  • A named planning owner and the system where purchasing decisions are made.
  • Cost inputs from finance for holding, stockout and fulfilment.

Common questions

Before you ask.

How much sales history do we need?

Enough to see seasonality at SKU and location level, with a record of when items were out of stock. Short history, volatile demand and unrecorded promotions limit accuracy, and there are methods for new products with limited history.

Will forecasts replace our planners?

No. Planners own the plan and approve purchase quantities, and overrides are recorded with a reason.

Can you work across several channels and marketplaces?

Profitability work reconciles orders, payments, returns and costs across channels using entity resolution. Gaps in channel data limit coverage.

Next step

What could AI mean for your business?

Take our free Enterprise AI Adoption Assessment to explore where AI could make a difference. Get two scores, a recommended next move and a full report by email in about three and a half minutes.