BRENNEN

Industries / Retail, hospitality and personal services / Retail and e-commerce

Buying and assortment

Buying is the merchant's work of choosing which products a shop sells (its assortment), finding suppliers, setting up each item's details and deciding how much to order.

5

Milestones achieved, from 1928 to 2020.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1928

    A shop's own brand

    Marks & Spencer registered the trademark St Michael in 1928 and at first gave it only to its best-quality goods. An own brand, also called a private label, is a product a supplier makes to the shop's specification and the shop sells under its own name. By 1958 every item M&S sold carried the St Michael name; M&S phased the brand out in 2000.

    LessonWith an own brand, the buyer specifies the product instead of only choosing it, and the shop answers for its quality.

    archive.marksandspencer.com

    Credit

    Marks & Spencer

  2. 1985

    Supplier and store share data

    Procter & Gamble and Walmart began a partnership in which the two firms swapped orders and stock figures by electronic data interchange (EDI), standard messages sent from one company's computers to another's. P&G watched Walmart's stock and shipped before Walmart placed an order, a method called continuous replenishment. Two business professors writing in MIT Sloan Management Review in 1995 reported that P&G's on-time deliveries to Walmart "improved significantly, while inventory turnover increased dramatically."

    LessonBoth firms gained once the supplier could see what the store was selling.

    sloanreview.mit.eduscdigest.com

    Credit

    Procter & Gamble and Walmart

  3. 1992

    Suppliers see store sales

    Walmart deployed Retail Link, a system that showed suppliers the sales trends and stock levels of their own products in Walmart stores. In 1996 Walmart made Retail Link and EDI available over the internet.

    cio.comscdigest.com

    Credit

    Walmart

  4. 2015

    Bad item data empties shelves

    Target opened stores in Canada in 2013. Canadian Business, as relayed by MPR News, traced the trouble to "data entry missteps" in its item records. Shipping containers could not be packed correctly, some items did not fit on the shelves, and shoppers met "rows of empty shelves" while Target's distribution centers were so full that it rented extra storage. In January 2015 Target filed to close all 133 Canadian stores, which employed about 17,600 people, and expected about $5.4 billion in pre-tax losses.

    LessonEvery order a merchant places depends on correct item records, such as size and case quantity.

    mprnews.orgcorporate.target.com

    Credit

    Target Corporation

  5. 2020

    Forecasting contest on Walmart data

    The M5 competition, run from the University of Nicosia between March and June 2020, asked entrants to forecast 28 days of unit sales for 42,840 series of Walmart sales data, most of them with many days of zero sales. Close to 6,000 people took part. The winning entry used LightGBM, a machine-learning method that adds up many small decision trees. By the organizers' published scores, it averaged 0.520 on their error measure (lower is better), against 0.671 for the best of their 24 standard benchmark methods, a form of exponential smoothing.

    LessonThe organizers' untuned machine-learning benchmarks scored worse than exponential smoothing, so the kind of method alone did not decide the result.

    unic.ac.cygithub.com

    Credit

    the M5 organizers, led by Spyros Makridakis; Walmart, which supplied the data

Watch for

  1. AI drafts own-brand fashion

    Walmart says its Trend-to-Product tool reads trend data and uses generative AI to draft mood boards, which designers and merchants then refine. It says this "shortens the traditional production timeline for Walmart fashion by as much as 18 weeks."

    Achieved whenan outside party with access to the data confirms the shorter lead times across Walmart's fashion lines.

    corporate.walmart.comcorporate.walmart.com

    Claimed by

    Walmart, April and May 2025

  2. Paid supplier data lifts sales

    Walmart Data Ventures says suppliers who pay for Scintilla, its data and analytics service, saw "total omni sales increasing by 15%" (sales in stores and online together) when it compared October 2024 to September 2025 with the year before.

    Achieved whenan independent analysis confirms the effect and rules out that faster-growing suppliers are simply the ones that buy the data.

    corporate.walmart.com

    Claimed by

    Walmart Data Ventures, October 2025

Open

  1. 01

    Store food audits from video

    A tool that scores a food store from a phone video of its aisles using NEMS-S (Nutrition Environment Measures Survey in Stores), the standard form that public-health researchers use to record whether a store stocks healthy options, at what price and in what condition. The aim is a score as good as a trained person's, so that every store in a city can be audited each year.

    The brief
    Credit · unclaimed
  2. 02

    Check green claims on shelves

    An open checker that reads product packs and product pages and flags claims banned in the EU since 27 September 2026: generic environmental claims such as "eco" or "green" that the seller cannot back with recognized excellent environmental performance, and sustainability labels that come neither from a certification scheme nor from a public authority. Buyers could run it before an item goes on sale.

    The brief
    Credit · unclaimed

Open 01 · the brief

Store food audits from video

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A tool that scores a food store from a phone video of its aisles using NEMS-S (Nutrition Environment Measures Survey in Stores), the standard form that public-health researchers use to record whether a store stocks healthy options, at what price and in what condition. The aim is a score as good as a trained person's, so that every store in a city can be audited each year.

Why it is still open

NEMS-S is an observational form filled in by trained raters who walk the store. Its developers tested it in 85 stores around Atlanta and found that two raters agreed closely on availability (kappa 0.84 to 1.00, where 1 is perfect agreement). A search for studies that score NEMS-S from photos or video found none.

europepmc.org

Built with what exists

a phone camera; vision-language models that read video frames, such as Gemini or Claude; product facts and photos from Open Food Facts, a free public food database; Python for scoring.

  1. First step

    Learn the NEMS-S form, audit three stores by hand while filming each aisle, and use those as the first test set.

  2. Done looks like

    In 30 stores audited both by trained raters and by the tool, the tool's availability scores agree with the raters' at kappa 0.8 or better, and its price readings are within 5% for at least 90% of items. The videos, scores and code are published.

Open 02 · the brief

Check green claims on shelves

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open checker that reads product packs and product pages and flags claims banned in the EU since 27 September 2026: generic environmental claims such as "eco" or "green" that the seller cannot back with recognized excellent environmental performance, and sustainability labels that come neither from a certification scheme nor from a public authority. Buyers could run it before an item goes on sale.

Why it is still open

In a 2020 check of websites, the European Commission and national consumer authorities examined 344 green claims that looked doubtful and had reason to believe that in 42% of cases the claims were "exaggerated, false or deceptive." The new rules (Directive (EU) 2024/825) apply from 27 September 2026 "without a transition period", and they also cover goods already made or on shelves before that date. A search found compliance services but no published checker tested against expert labels.

ec.europa.eumayerbrown.com

Built with what exists

a phone camera and product pages from EU online shops; vision-language models to read pack text; Label Studio, an open-source labeling tool; scikit-learn for a simple baseline; the text of Directive (EU) 2024/825 as the rulebook.

  1. First step

    Collect the claims on 100 cleaning and personal-care products from one supermarket's website and label them by hand against the directive.

  2. Done looks like

    1,000 claims from at least 300 products sold in three EU countries, each labeled "allowed" or "banned" by two people with consumer-law training. On 300 held-out claims the checker finds at least 90% of the banned ones, and at least 80% of what it flags is banned. Data and code are published.