AI use case

AI dynamic pricing and markdown optimization for retail

AI that recommends or automatically sets the regular price, promotion or markdown for every item in every store, from demand, cost, competitor prices, inventory and, for perishables, how close the item is to its best before date, so pricing analysts manage exceptions and strategy instead of spreadsheets.

By Len Debets · Last verified 29 September 2026 · 3 public deployments

USD 1.5 million to USD 21.9 million
Indicative value per year
A grocery retailer with 500 stores and USD 4 million in average annual perishable and seasonal sales per store. Worked example, see how it is calculated.

What problem does it solve?

A grocery or general merchandise retailer reprices thousands of items across hundreds of stores every week: regular price changes, promotions, and markdowns on seasonal and perishable stock that will not sell at full price. Done by hand in spreadsheets, this is slow, inconsistent between analysts, and reactive: a marked down shelf is a decision made after the item has already lost value, not before.

Revionics, the pricing platform used by Delhaize America and Rimi Baltic, describes the manual alternative directly: before adopting price optimization software, Delhaize America's pricing analysts managed prices with complex, unwieldy spreadsheets, a slow manual procedure that lacked scalability, and pricing processes and data collection were not standardized, which made it hard to ensure pricing strategies were followed enterprise wide. For perishables, the cost of getting the timing wrong is thrown away stock: at Albert Heijn, Supermarket News reports that prices on chicken and fish are reduced automatically based on their sell by date, with a higher discount for items that need to be sold soonest.

How does it work?

  1. Bring in the signals. Point of sale history, cost and margin, current inventory, competitor prices, and, for markdowns, the sell by or best before date, feed a pricing engine continuously.
  2. Estimate demand at different prices. The engine estimates price elasticity per item and location: how much volume changes as price changes, learned from historical sales, not fixed rules.
  3. Recommend or set the price. For everyday and promotional pricing, the engine proposes a price within the retailer's guardrails for an analyst to approve; for markdowns on electronic shelf labels, it can act automatically within a small set of fixed discount steps, choosing the step and the timing so the item is more likely to sell before it expires.
  4. Push the price to the shelf. The new price reaches paper tags on the next print run, or an electronic shelf label in near real time, so a fast moving markdown clock is actually visible to the shopper.
  5. Feed the outcome back. Sell through, waste and margin by item and store retrain the elasticity estimates, and pricing analysts review exceptions and set the strategy: which categories run automated markdowns, and what the outer guardrails are.
Audience
Employee facing
Autonomy
Supervised agent
Adoption
Early adopters
Channels
Internal tools, Kiosk and branch

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

No public deployment has disclosed a measurable outcome yet.

Value drivers: Revenue growth, Employee productivity, Speed and cycle time.

Indicative value

A grocery retailer with 500 stores and USD 4 million in average annual perishable and seasonal sales per store

USD 1.5 million to USD 21.9 million

Perishable and seasonal margin recovered per year

How this is calculated

Formula: stores * perishableSalesPerStore * unsoldAtFullPriceShare * perishableMarginRate * marginRecovered. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Stores stores, stores500500The reference retailer.
Perishable and seasonal sales per store per year perishableSalesPerStore, USD per store per year3,000,0005,000,000Editorial assumption for a full line grocery store, replace with your own perishable and seasonal category mix.
Share of perishable and seasonal sales sold at a markdown or written off unsoldAtFullPriceShare, fraction of perishable and seasonal sales0.040.1Editorial assumption, replace with your own waste and markdown rate by category.
Gross margin on perishable and seasonal sales perishableMarginRate, fraction of sales value0.250.35Editorial assumption: a typical gross margin range across produce, meat, dairy and general merchandise categories, replace with your own category margin mix.
Share of that lost margin recovered by better timed, better sized markdowns marginRecovered, fraction of the margin at risk in markdown and waste0.10.25Editorial assumption, conservative because none of the three deployments on this page discloses a percentage or currency margin result: Revionics' customer case studies for Delhaize America and Rimi Baltic state results in qualitative terms only ("improved markdowns planning and margins" for Delhaize America, "positive lifts in units and revenue" for Rimi Baltic), and Albert Heijn has not disclosed a percentage or currency figure for its dynamic discounting rollout. A separate Microsoft customer story about a related Albert Heijn initiative (cited on this library's demand forecasting page) reports a weight based saving, "This now saves 250,000 kilos of wasted food per year", which is not a margin or currency figure and is not used in this estimate.

What it leaves out: Gross margin recovered only. It leaves out the cost of the pricing platform, electronic shelf labels and integration, the risk that a wrong markdown clears stock too early at a needless discount, and any change in footfall or basket size from more visible in store discounting.

Who already uses it?

3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

Albert Heijn

Netherlands · Retail and ecommerce · 2019

ScaledGrade B

Albert Heijn, the Ahold Delhaize supermarket chain in the Netherlands, built an in house algorithm that automatically discounts fresh produce, chicken and fish as their best before date approaches, so stock is cleared by the end of the day instead of thrown away. The algorithm was advised by anti waste specialist Wasteless and first tested with Pricer's electronic shelf label hardware in one store in 2019. It is now rolled out to every Albert Heijn store fitted with electronic shelf labels, working with three fixed discount levels (25%, 40% and 70%) that the software chooses per product from sales history, local and seasonal patterns, weather and current stock.

No outcome disclosed.

Rimi Baltic

Latvia · Retail and ecommerce · 2019

ScaledGrade C

Rimi Baltic, the ICA Gruppen owned grocery and convenience retailer operating in Latvia, Lithuania and Estonia, replaced rule based pricing in its ERP system with the Revionics price optimization platform. It rolled out category by category in two steps, first moving its existing pricing rules onto the platform, then switching those categories to price elasticity based pricing, and completed the change across its full product assortment in all three countries in under a year.

No outcome disclosed.

Delhaize America

United States · Retail and ecommerce · 2014

ProductionGrade C

Delhaize America, at the time the US subsidiary of Delhaize Group operating supermarket chains including Food Lion and Hannaford (this record predates the 2016 merger that formed Ahold Delhaize), replaced spreadsheet based pricing with the Revionics price optimization platform across all of its price zones and stores. The retailer moved from rule based pricing to scenario based, shopper centric pricing, then extended the same platform to markdown decisions, so the company could make fact based markdown and clearance decisions, including markdown cadence and depth, aligned with local shopper demand and store level inventories instead of blanket rules.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • At least two years of point of sale history by item and store, including promotions
  • Current cost, margin and inventory position by item and store
  • For perishables, best before or sell by dates at the item and batch level
  • Competitor price data where the retailer competes on price

Systems to integrate

  • Point of sale and pricing master data
  • Inventory and warehouse management systems
  • Electronic shelf labels or the store's price ticket printing system
  • Promotion planning and calendar systems

Complexity: Medium

The pricing logic itself is a mature, well understood optimization problem. The work is integrating clean, current cost, inventory and competitor data per store and item, and, for automated markdowns, electronic shelf labels that can update fast enough to make the discount clock real.

  1. 1

    Start with one lever and one category group

    Pick either everyday pricing or markdowns, and a limited set of categories with clean data and clear ownership, before touching the full assortment.

  2. 2

    Set the guardrails before the model runs

    Agree the price bounds, the fixed markdown steps if any, and which categories or events (a supplier price change, a local competitor promotion) always route to a human.

  3. 3

    Decide what runs automatically and what needs approval

    Automated markdown execution on electronic shelf labels needs the tightest guardrails, since there is no review step before the shopper sees the price; everyday price changes can start as recommendations an analyst approves.

  4. 4

    Pilot in a subset of stores with a holdout

    Run the new pricing in a sample of stores against a matched set that keeps the old process, so sell through, waste and margin can be compared honestly.

  5. 5

    Roll out category by category

    Rimi Baltic moved its existing pricing rules onto the platform first, one category at a time, before switching those categories to elasticity based pricing, which kept the change manageable and let the team catch mistakes in the rules logic early.

  6. 6

    Give analysts the exception queue, not the whole catalogue

    Once trust is established, most items should not need a human look; analysts spend their time on flagged exceptions, strategy and the categories still outside automation.

Guardrails

  • Hard floor and ceiling on any price or discount change, with no override outside approved limits
  • A fixed, published set of markdown steps for automated execution, not an unbounded discount
  • Competitor and cost data source and freshness checked before it can move a price
  • Exceptions (a data outage, an unusual price swing, a new item with no history) routed to a pricing analyst, not applied automatically

KPIs to instrument

  • Sell through and waste by category and store, automated versus a holdout
  • Margin by category, automated versus a holdout
  • Analyst time spent on exceptions versus routine repricing
  • Price change accuracy: the share of published prices matching the system's recommendation

Human in the loop

Pricing analysts own the guardrails, the category rollout plan and every exception the system flags, and review a sample of automated markdown decisions each week. Category managers sign off before a new category moves from recommendation to automated execution.

Common failure modes

A stale signal moves the price
Bad inventory or cost data pushes a wrong price or markdown live. Check data freshness before applying a change, and cap the size of any single move.
Discount steps train shoppers to wait
If the first markdown step always appears at a predictable point, shoppers learn to wait for it. Vary timing within the guardrails and monitor sell through at each step, not only at the end.
No holdout, no honest result
Comparing automated stores to their own past performance overstates the effect, because seasonality and the wider market shift too. Keep a matched holdout for as long as the programme runs.
Automation outruns the guardrails
A category is automated before its guardrails, exception rules and electronic shelf label coverage are actually ready. Gate the move to automation on a checklist, not a launch date.

What are the risks and rules?

EU AI Act

Minimal risk

Setting or recommending retail prices for goods is not listed in Annex III and does not evaluate the creditworthiness, employment, or access to an essential service of a natural person, so this is minimal risk under the EU AI Act. It would need reassessment if a retailer used the same engine to set an individual price per identified customer rather than per product and store, which raises separate consumer protection and non discrimination questions the deployments on this page do not describe.

Rules that apply

Controls to put in place

  • Published, auditable guardrails (price bounds, markdown steps) that automated execution cannot exceed
  • A permanent holdout group of stores or categories for honest measurement
  • Change log of every price and markdown decision, automated or analyst approved, with the data it used
  • Regular review of automated categories for unintended patterns, such as discount timing shoppers can predict and exploit

Frequently asked questions

Does AI dynamic pricing mean charging each shopper a different price?
Not in the deployments on this page. Delhaize America and Albert Heijn price by product and, for markdowns, by store and remaining shelf life. Rimi Baltic's elasticity based pricing runs at the product and category level; its sources do not describe store level pricing. None of the three prices by identified shopper. Individual level pricing raises separate fairness and disclosure questions this page does not cover.
How much does AI markdown optimization actually save?
None of the deployments on this page publishes a percentage or currency result for markdown or margin impact. Revionics' own case studies state results at Delhaize America and Rimi Baltic in qualitative terms only ("improved markdowns planning and margins" for Delhaize America, "positive lifts in units and revenue" for Rimi Baltic), with no percentage or currency figure attached. Albert Heijn has not disclosed a percentage or currency result for its dynamic discounting rollout either. A separate Microsoft customer story about a related Albert Heijn initiative reports that it "now saves 250,000 kilos of wasted food per year", a weight, not a margin or revenue figure, so treat any margin or revenue number you see elsewhere as an industry estimate, not a result from these deployments.
What is the difference between everyday pricing and markdown optimization?
Everyday pricing sets the regular shelf price from demand, cost and competition. Markdown optimization decides how and when to discount stock that will not sell at full price, typically seasonal goods or perishables close to their best before date, which is why Albert Heijn's system works from remaining shelf life rather than from a fixed markdown calendar.

How to cite this page

Blits.ai AI Use Case Library, "AI dynamic pricing and markdown optimization for retail", last verified 29 September 2026, https://www.blits.ai/ai-use-cases/retail-dynamic-pricing-and-markdown-optimization. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 29 September 2026: First published

Related use cases

Retail and ecommerce

AI demand forecasting and automated replenishment for retail

Machine learning that forecasts demand for every item in every store or fulfillment center, day by day, from sales history, promotions, prices, weather and local events, and turns the forecast into automatic store and warehouse orders within limits set by planners, who handle the exceptions.

Deployments
4 public, best grade B
Autonomy
Supervised agent
Travel and hospitality

AI copilot for hotel revenue management

An employee facing AI system that forecasts demand for a hotel or portfolio by date, room type and segment, recommends or automatically adjusts room prices and availability controls within limits a revenue manager sets, and scores group and event enquiries for true profitability, so a small revenue team can run pricing that used to need daily manual adjustment in the property management system.

Deployments
2 public, best grade C
Reported revenue uplift
33%
Hôtel Swexan, vendor claim
ManufacturingRetail and ecommerce

AI trade promotion optimization for consumer goods manufacturers

Machine learning and optimisation that plans and prices trade promotions, discounts, displays and rebates a consumer goods manufacturer runs with retailers, forecasting the volume each promotion mechanic would generate and searching the combinations of timing, depth and mechanic that best meet a revenue, profit or volume target, for a key account manager to review and agree with the retailer.

Deployments
2 public, best grade B
Autonomy
Copilot
Insurance

AI copilot for insurance pricing and actuarial analysis

AI that speeds up the work of pricing and actuarial teams, from automated, transparent risk and demand model building to natural language analysis of rate filings, experience data and reserving diagnostics, while actuaries select the models, sign off the rates and own the professional judgment.

Deployments
5 public, best grade B
Reported productivity gain
5x
Generali France, organization claim