Fresh Ordering by Algorithm: What Heinen’s 23 Stores Actually Reported

How CodoraTech is funded: CodoraTech is supported by advertising and, in some articles, by affiliate links. Where an article contains affiliate links, we say so at the top of that article.

Heinen’s Fine Foods runs 23 supermarkets around Cleveland and Chicago. In August 2020 it announced that it had deployed Afresh Technologies’ AI-powered ordering and merchandising platform across all of them, in the fresh departments where spoilage is decided daily. It remains one of the few named deployments at genuinely mid-sized scale rather than at a national chain. It is also a useful lesson in reading evidence, because the percentages that travel with the story came from the software vendor, and the grocer itself published no number at all.

Key facts

• Heinen’s Fine Foods: 23 stores across the Cleveland and Chicago areas; Afresh deployment announced 18 August 2020
• Afresh Technologies: San Francisco company selling forecasting and ordering software for fresh departments
• Up to 50 percent less in-store food waste and 3 percent incremental sales growth are Afresh’s claims, not audited findings
• Heinen’s published no percentage of its own
• 8.8 percent of U.S. small businesses reported using AI, per the SBA Office of Advocacy, September 2025

Who deployed what, and when

Grocery Dive reported on 18 August 2020 that Heinen’s had put Afresh’s platform into every one of its 23 stores. Afresh, based in San Francisco, sells demand forecasting and ordering software aimed specifically at fresh categories, where a buying decision made on Tuesday shows up as spoilage on Friday. The choice of a family-owned regional chain matters here: most named AI deployments in grocery come from companies with thousands of locations and a corporate data science function, which makes them poor guides for a business running under thirty stores.

Greg Sotka, Heinen’s director of category management and merchandising, was quoted in that report saying: “Our product is fresher and more in-stock and we’ve significantly reduced shrink.” The grocer said the improvement came out of its own internal testing. What it did not do was attach a figure to any of it. There is no published Heinen’s percentage for waste, for shrink or for sales, and any article that supplies one has taken it from somewhere else.

The 50 percent figure belongs to the vendor

Two numbers follow this deployment wherever it is retold: up to a 50 percent reduction in in-store food waste, and 3 percent incremental sales growth. Both are Afresh’s own claims, reported as the vendor’s claims by Grocery Dive. No third-party audit of either figure appears in the reporting reviewed for this article. The hedge “up to” is the vendor’s, and dropping it changes a ceiling into an expectation.

Afresh has separately said, as of March 2026, that its software has prevented more than 200 million pounds of food loss on a projected annual basis. That is again a supplier figure, unaudited, and the word projected is doing considerable work inside it. What can be established from outside the company is commercial traction rather than outcome: Albertsons Companies rolled the platform out to more than 2,000 stores, and later extended it from produce into meat and seafood. Continued purchasing tells you customers find the software worth renewing. It is not the same evidence as a measured waste reduction.

• Afresh’s claim, reported by Grocery Dive: up to 50 percent reduction in in-store food waste
• Afresh’s claim, reported by Grocery Dive: 3 percent incremental sales growth
• Afresh’s claim, March 2026: more than 200 million pounds of food loss prevented on a projected annual basis
• Heinen’s own published outcome figure: none
• Independent audit of any of the above located in this research: none

Store managers override the model, and their targets tell them to

The most useful counter-evidence comes from research by ReFED published in May 2026 and reported by The Packer. It found that store managers routinely disregard accurate AI ordering suggestions, because internal performance metrics punish an empty shelf far more heavily than they punish backroom spoilage. The incentive to over-order survives the arrival of software that recommends otherwise, and it is a management problem rather than a modelling one.

The same research notes that AI ordering requires workflow restructuring that, in its phrasing, “legacy chains aren’t equipped to support,” alongside cultural resistance to trusting predictive systems. Afresh chief executive Matt Schwartz describes the intended operational change this way: “Before Afresh, you see a very full backroom…Afterward…the floor will be full, but we’ll see much leaner backrooms.” That is the supplier characterising the effect of its own product, and it reads as a description of the target state rather than a measurement of it.

One figure in that reporting is often lifted out of context: a projected 2.7 million dollars in annual savings at a U.S. value grocer. The grocer is not named. An unnamed company cannot serve as a case study for anybody, and the number belongs in an article only as an illustration explicitly labelled as such.

Small firms are about a year behind, on the government’s own count

For context on how unusual the Heinen’s decision was in 2020, the U.S. Small Business Administration Office of Advocacy published a research spotlight titled “AI in Business: Small Firms Closing In” on 24 September 2025, written by regulatory economist Robert Press. It reported that 8.8 percent of small businesses, defined as those with fewer than 250 employees, said they were using AI, against 11.1 percent of large businesses. Six months earlier the small-business figure had been 6.3 percent, which is why Advocacy characterises small firms as only a year behind larger ones. The underlying data is the Census Bureau’s Business Trends and Outlook Survey, covering September 2023 to August 2025.

The same spotlight found the average small business using AI runs about 2.0 use cases against 2.1 for large firms, so the depth of adoption is close even where the rate is not. The gaps sit in specific technologies.

• Robotic process automation: 16.7 percentage point gap between small and large firms
• Data analytics: 9.9 percentage point gap
• Chatbots: 7.0 percentage point gap
• Census Bureau reporting separately confirms firms with 20 or more employees are the heaviest AI users

Five questions to put to a fresh-ordering supplier

1. Who produced the headline percentage, the vendor or someone independent of it, and has anyone audited it?
2. What baseline is the reduction measured against, over what period, and across how many stores?
3. Which named customers will discuss the deployment, and what figures did those customers publish themselves?
4. What happens to the claimed result when a store manager overrides the ordering suggestion, and how often does that occur in existing accounts?
5. What workflow, staffing and performance-metric changes does the software assume, and who inside the business is responsible for making them?

Sources: Grocery Dive · The Packer · U.S. Small Business Administration Office of Advocacy · Grocery Dive · U.S. Census Bureau