AI in Grocery Retail 2026: What Late Adopters Are Losing

AI in grocery retail is changing how supermarkets order fresh food, set markdowns, staff stores and sell through shopping assistants. Grocers that wait pay in shrink, empty shelves and lost baskets. US independents lost 3.9% of sales to shrink in fiscal 2025, above their 2.2% pre-tax profit. That gap is where AI is paying back first.

Grocery is a 2% business. FMI puts the average US food retailer's net profit at 2.1% of sales in 2025.

On a margin that thin, small fixes move profit a long way. Cutting shrink by a tenth can add almost a fifth to pre-tax profit, as the model further down shows.

When supermarket and superstore owners ask me where AI fits, I ask what shrink cost them last year. For the average US independent, it cost more than they made.

This article puts numbers on that. It covers what changed in 2026, which grocers are banking savings, what waiting costs, and where AI loses money.

• US food retailers averaged 2.1% net profit in 2025 (FMI), so waste and empty shelves decide whether a grocer has a good year.
• Shrink at US independent grocers rose to 3.9% of sales in fiscal 2025, well above their 2.2% pre-tax profit (NGA and FMS Solutions).
• More than two-thirds of US food retailers used AI in early 2026, up from 47% a year earlier, according to FMI.
• For a $500 million grocer, a 10% cut in shrink plus a small fresh sales lift is worth about $2.5 million a year, roughly 23% more pre-tax profit.
• AI agents have started building grocery baskets: Walmart says Sparky users' baskets are 35% larger, and Instacart runs full checkout inside ChatGPT.
• Personalised AI pricing is now a legal risk: Maryland bans surveillance pricing by food retailers from 1 October 2026, and New Jersey follows in 2027.

How Is AI Changing Grocery Retail in 2026?

AI is moving into the jobs that decide grocery margin: ordering, markdowns, labour and the basket itself. Adoption jumped in a single year.

More than two-thirds of food retailers in FMI's 2026 survey said they use AI, against 47% a year before. Generative AI use rose from 43% to 59%. Retailers also doubled technology spending to nearly 2% of sales in 2025.

That money is going into a short list of jobs.

• Fresh ordering. Models set daily orders for produce, meat and bakery from sales history, promotions, weather and shelf life.
• Markdowns. Software decides when and how deeply to discount short-dated food, based on stock left and sell-through.
• Store work. Kroger is building agents that cut dozens of daily store alerts down to the few that matter.
• Distribution. Walmart expected about 65% of its stores to be serviced by automated distribution by the end of fiscal 2026.
• Shopping assistants. Kroger and Walmart let shoppers describe a meal or a budget and get a ready basket. Woolworths is building the same.
• Assortment and loss prevention, which FMI's 2026 report lists among the ways retailers already apply AI.

Ordering leads the list. In Toast's 2026 retail survey, 76% of grocers said they use or plan to use AI for inventory and ordering.

AI in grocery retail

AI in grocery retail is the use of machine learning, computer vision and language models to run grocery operations and shopping. Common uses are demand forecasting and automated ordering, markdown and price optimisation, shelf and inventory monitoring, labour scheduling, and AI shopping assistants that plan meals and build baskets.

Grocery Doppio is a research partnership of Incisiv and Wynshop. It estimates AI could create $136 billion in value for grocers by 2030. Half of it sits in supply chain and logistics.

McKinsey found US grocery sales grew 1.2% in 2025, while prices rose 2.2% and volumes fell 1%.

When shoppers buy fewer units, one grocer's growth mostly comes out of another grocer's basket.

Sources: FMI: The Food Retailing Industry Speaks 2026, McKinsey: The State of Grocery North America 2026, Grocery Doppio: State of AI in Grocery 2024

How Much Money Are Grocers Losing by Not Using AI?

For many grocers, the avoidable losses are bigger than the profit. The clearest data comes from US independents.

The NGA and FMS Solutions study covered 503 independent operators and nearly 2,000 stores. In fiscal 2025, shrink rose to 3.9% of sales. Pre-tax profit was 2.2%.

3.9% of sales lost to shrink at US independent grocers in fiscal 2025, against average pre-tax profit of 2.2% NGA and FMS Solutions, 2026 U.S. Independent Grocers Financial Study

Shrink

Shrink is the value of inventory a grocer pays for but never sells. It covers spoiled and damaged goods, theft, and recording errors, and it is reported as a percentage of sales. In grocery, spoilage of fresh food is the part most directly affected by ordering decisions.

The same study shows the other leaks. Out-of-stocks ran at 6.6%. Inventory turns slowed from 17.8 to 16.1, so more cash sat on shelves and in backrooms.

National waste data tells the same story. ReFED valued US retail surplus food at $30.3 billion in 2024. That is about 3% of food and beverage retail sales.

Across all of retail, IHL Group puts the global cost of out-of-stocks and overstocks at $1.7 trillion a year.

Inside a single store, the money leaks in five places.

• Fresh spoilage. Many stores still order produce, meat, deli and bakery by feel, and the error ends in the bin.
• Empty shelves. A 6.6% out-of-stock rate means roughly one item in 15 is missing when a shopper looks for it.
• Rule-based markdowns. A fixed half-price sticker at closing time ignores how much stock is left and how fast it sells.
• Slow stock turns. Every extra day of inventory ties up cash that a 2% business cannot spare.
• Manual counting and ordering. Department managers spend part of every shift on counts and order sheets.

Most of this is fixable with data grocers already hold. In FMI's 2025 fresh foods report, fresh shrink held steady while center-store shrink rose.

FMI credited better inventory and demand planning tools for fresh food. Yet only 40% of retailers had used such tools in the past year.

Sources: NGA and FMS Solutions: 2026 Independent Grocers Financial Study, ReFED: Food waste by sector, retail, IHL Group: 2026 Inventory Distortion Study, FMI: Four stats on fresh food departments

What Does Waiting on AI Cost a $500 Million Grocer?

About $2.5 million a year in pre-tax profit, on cautious assumptions. That is roughly 23% more profit than the grocer makes today.

I built the model below from published averages rather than vendor promises. Where I use vendor results, I take a fraction of them.

The shrink line does most of the work. Afresh, an AI ordering vendor, reports up to 25% less shrink in the fresh departments it runs.

Afresh is live in more than 12,500 departments across 40 US states. My model assumes far less than its 25%, because not all shrink is spoilage. Theft and recording errors need other fixes.

The sales line is smaller but real. Afresh reports a 3% sales lift, and the model assumes one third of that.

Now stretch it over time. Three years of waiting is about $7.6 million of profit this grocer never books. That is before any shopper switches to a store with fuller shelves.

Your inputs will differ, so rerun the model with your own numbers. You need four figures: sales, pre-tax profit, shrink and fresh sales. Your month-end pack should already hold all four.

Sources: FMI: Food industry facts, Afresh: funding round and customer results, April 2026

Which Grocers Are Already Making Money From AI?

Mostly the biggest chains, and they now report AI results to investors. The table below collects what each has published.

Walmart is furthest ahead. It told investors in May 2026 that Sparky users build baskets 35% larger than other shoppers.

Sparky-attributed sales rose 150% in the quarter to 30 April 2026. Walmart now runs Sparky inside ChatGPT as well as in its own app.

35% larger baskets for Walmart shoppers who use its Sparky AI agent, compared with shoppers who do not Walmart Q1 FY27 earnings call, reported by Modern Retail

Kroger rolled its AI Shopping Assistant out across its websites and apps in July 2026. At Groceryshop in September, chief digital officer Yael Cosset said users were building larger baskets.

He had expected the opposite. Kroger is also training more than 400,000 associates on AI tools.

Sources: Modern Retail: Walmart says AI users build 35% bigger baskets, Kroger: AI strategy at GroceryShop 2026, Supermarket News: Kroger AI assistant drives larger baskets

In the UK, Tesco beat its Save to Invest target in 2025/26 with about £535 million of savings. Tesco calls AI an enabler, from demand forecasting to supply chain risk tools.

Sainsbury's moved its food products onto a machine learning forecasting platform. Food availability rose 190 basis points over four years. The grocer is on track for £1 billion of cost savings by March 2027.

Mid-sized chains buy the same capability as software. Afresh runs fresh ordering for Albertsons Companies, Meijer and Wakefern.

Across retail, IHL Group found sales growth leaders are 482% more likely to call themselves early technology adopters than laggards.

BCG reached a similar result across industries. Its 'future-built' firms, 5% of the sample, show 1.7 times laggards' revenue growth and 1.6 times their EBIT margin.

Savings like these end up on the shelf. Tesco says its programme funds a better customer offer and higher colleague pay.

A rival that cuts waste can afford lower shelf prices. A grocer that cannot has to match them out of a 2% margin.

Grocers ask me where AI should start. I tell them the produce room, because that is where their profit goes in the bin every night.

Sources: Tesco: Preliminary Results 2025/26, Sainsbury's: Preliminary Results 2024/25, IHL Group: The compounding retail AI advantage, BCG: AI leaders outpace laggards

Why Could AI Shopping Agents Take Your Customers?

Because agents now build the basket, and they can only fill it from stores whose data they can read.

Agentic commerce

Agentic commerce is shopping carried out by an AI agent on a customer's behalf. The agent plans, compares products and prices, builds the basket and, with permission, completes the purchase, often inside a chat assistant rather than the retailer's own app or website.

The first agents are already live in grocery. Instacart launched full grocery checkout inside ChatGPT on 8 December 2025. It covers more than 1,800 retailers.

Walmart's Sparky runs in its own app and in ChatGPT. Woolworths is rebuilding its Olive assistant on Google's Gemini to assemble baskets, with the shopper's permission.

Kroger has described where this goes. Cosset laid out three stages: assisted, augmented and autonomous. In the last stage, shoppers share goals and agents handle planning, shopping and fulfilment.

Shoppers are warming up slowly. McKinsey found 12% of European consumers have already used AI somewhere in their grocery journey.

In North America, 51% would use AI to help search for products. Only 20% would accept fully automatic ordering with no review.

That gap buys grocers some time, though not much. Nearly half of the grocery executives McKinsey surveyed expect agents to shape how customers shop.

Meanwhile, Amazon keeps growing in fresh food. It reported over $150 billion of grocery gross sales in 2025. It now delivers fresh groceries same-day in more than 2,300 US cities and towns.

An agent can only recommend what it can see. To keep your products in the basket, it needs five things from you.

1. A clean product catalogue with sizes, ingredients, allergens and images.
2. Store-level stock counts that update several times a day.
3. Current prices and promotions, including loyalty offers.
4. Delivery and collection slots the agent can book.
5. An API or product feed that agents and marketplaces can call.

In the grocers we have worked with, these sat in separate POS, inventory and e-commerce systems. Joining them is a software project, and it is the foundation for everything else here.

Sources: Instacart: app launches in ChatGPT, Retail Dive: Walmart brings Sparky to ChatGPT, McKinsey: The State of Grocery Retail Europe 2026, Amazon: Q1 2026 stores and grocery update

Why Do So Many Grocery AI Projects Fail to Pay Back?

Most stall between the pilot and the P&L. McKinsey's 2026 European grocery report describes this as an AI paradox.

About 90% of retailers are experimenting with AI, yet only a small share see a clear profit impact. Meanwhile, 47% of grocery CEOs rank AI and automation among their top three priorities.

The causes are usually operational, and they repeat.

• The pilot targets the showroom. A shopper chatbot demos well, but it does not touch shrink or availability.
• The item master is dirty. Wrong pack sizes, duplicate codes and missing shelf-life data break any forecast.
• Stock records are wrong. If the system thinks 12 units sit on the shelf and there are none, the model orders nothing.
• Nobody set a control group. Without comparison stores, a good month gets credited to AI and a bad one blamed on it.
• The tool does not talk to the POS or ERP, so staff re-key orders by hand.
• Store teams were never trained, so they override the system and nobody tracks why.

The other quiet loss is software that never connects to your data. Per-store fees keep running after a pilot stalls.

Sources: ESM Magazine: McKinsey on the AI paradox in grocery

Where Can AI Cost a Grocer Money Instead of Saving It?

Pricing is the fastest way to lose money and trust with AI. Instacart found that out in December 2025.

Consumer Reports, Groundwork Collaborative and More Perfect Union had 437 shoppers add the same items to Instacart carts. They used the same stores at the same time, across four cities.

About three-quarters of items showed more than one price. Some shoppers saw prices up to 23% higher. The groups estimated a family could pay up to about $1,200 a year more.

Instacart ended the price tests on 22 December 2025. Lawmakers moved anyway.

Maryland's governor signed the first state ban on surveillance pricing by food retailers on 28 April 2026. It takes effect on 1 October 2026, with fines up to $10,000 per violation.

New Jersey followed on 23 July 2026. Its ban starts on 1 August 2027. A pause on electronic shelf labels in grocery stores starts earlier, on 1 February 2027.

New York already requires businesses to disclose when an algorithm sets a price using personal data.

Sources: Consumer Reports: Instacart stops AI pricing tests, Morgan Lewis: Maryland HB 895, Progressive Grocer: New Jersey pricing and ESL law

What Should a Mid-Sized Grocer or Superstore Do First?

Start with AI ordering in one fresh department, measured against control stores for 12 weeks. Produce is a strong first candidate.

USDA's Economic Research Service estimated supermarket shrink at 12.6% for fresh fruit and 11.6% for fresh vegetables. It published those figures in 2016.

A sequence that works for a regional chain looks like this.

1. Pull last year's shrink, out-of-stock rate and fresh sales by store and department. That is your baseline.
2. Fix the item master and stock accuracy for the pilot department. A forecast starts from the stock count.
3. Run AI ordering in produce for 12 weeks in a handful of stores. Keep similar stores as a control.
4. Add markdown timing for short-dated fresh food, priced by stock and shelf life, and applied to every shopper.
5. Publish a clean product and stock feed that shopping agents and delivery marketplaces can read.
6. Give department managers one daily task list in place of a dozen reports. Measure the time it frees.
7. Scale only what beat the control stores, and drop the rest.

Buy where a proven product fits your POS. AI fresh ordering is an established category, and vendors already serve regional chains.

Build what only you can own: the POS, inventory and loyalty data pipeline, and the product feed agents read.

Leading super shop chain, Bangladesh

Augmex built a retail analytics platform for a multi-store super shop that was losing sales to stockouts of popular items while overstocking others. It joined POS, inventory and loyalty data, added demand forecasting for reorder points, and gave store managers stock and sales dashboards built for non-technical users.

Read the full case study

That project started where this article starts. The data already existed in three systems. The work was joining it and forecasting from it.

SKG Al Kamal Supermarket, Abu Dhabi

Augmex is building the supermarket's e-commerce platform with a 7-person team: real-time stock sync with the POS, scheduled delivery slots with route optimisation, recurring orders and personalised recommendations. The build has been running since 2025.

Read the full case study

Online, the same order of work applies. Stock sync with the POS comes before recommendations, because selling missing stock frustrates the customer.

Sources: USDA ERS: Updated supermarket shrink estimates for fresh foods

How Do the UK, Australia, the UAE and Bangladesh Compare?

UK and Australian majors now report AI results in their accounts. In the UAE and Bangladesh, public evidence is thinner.

• UK. Tesco's savings programme delivered about £535 million in 2025/26, and a Tesco customer assistant rolls out later in 2026. Sainsbury's runs food forecasting on machine learning.
• Australia. Coles began rolling out ChatGPT Enterprise in late 2025, the first major Australian retailer to do so at scale. Woolworths is rebuilding Olive on Google Gemini.
• UAE. Majid Al Futtaim, Carrefour's UAE operator, rolled out AI audience measurement in ten flagship hypermarkets in 2025. Mid-sized supermarkets such as SKG Al Kamal are building their e-commerce stack now.
• Bangladesh. Published data on AI use by super shops is scarce. The chain in our case study kept POS, inventory and loyalty data in separate systems before the analytics build.

The method travels even where the benchmarks differ. Measure shrink and availability first, then fix ordering.

Sources: Coles: AI adoption with OpenAI, Computer Weekly: Woolworths to power Olive with agentic AI, Zawya: Majid Al Futtaim and Advertima AI rollout

Frequently Asked Questions About AI in Grocery Retail

Before your next AI meeting, pull last year's shrink and out-of-stock figures by department. Those two numbers tell you what waiting costs. Augmex can build that model from your own POS data and scope the first workflow with you.