Industry · Food Retail

Retail automation and AI for South African food retailers

Store-level replenishment that knows your shelf lives, fresh markdowns made on data rather than gut feel, and R638 temperature records that keep themselves. Built for independent supermarkets, wholesalers and franchise groups — not only for the chains with their own data science teams.

Discuss your project
R178bn
in general dealer sales in Q1 2026 — nearly half of all South African retail trade
Source: Stats SA, Retail trade sales P6242.1 (March 2026)
64%
of the grocery market held by the top five retailers
Source: Competition Commission, Grocery Retail Market Inquiry (2019)
10.3m tonnes
of edible food lost or wasted in South Africa each year — 45% of the available supply
Source: CSIR briefing note on food loss and waste (May 2021)

Why food retail automation is a data problem before it is an AI problem

Food retail automation in South Africa is usually sold as a forecasting model. In practice the model is the easy part. The hard part is that last night's sales are in the POS, stock on hand is in the ERP, deliveries are on supplier portals and in email, write-offs are on a clipboard in the back of the store — and none of them agree on what a product is called.

That matters more when prices are not doing the work. Stats SA put annual food and non-alcoholic beverage inflation at 1.1% in August 2026, so growth in rand sales has to come from volume, from availability and from what you stop throwing away. The CSIR estimates that 10.3 million tonnes of edible food is lost or wasted in South Africa every year. Much of that happens before food reaches a shelf, but the share that happens in your stores and distribution centre is yours to manage.

The market is also uneven. The Competition Commission's Grocery Retail Market Inquiry found the top five retailers held about 64% of the market, and that rebates paid to national chains for store-level activities are not offered to the wholesalers that serve independent retailers. Independents cannot buy their way to the same terms, but they can run tighter operations with the data they already have.

We build the layer underneath: pipelines that land POS, stock, delivery and supplier data every night and check it before anyone acts on it; one product master; replenishment and markdown proposals that buyers can see, override and explain; and food-safety records that are captured automatically instead of being filled in at the end of a shift. Where it earns its place, AI sits on top — not instead of clean data.

Cloud data platforms for retail dataOur logistics and distribution practiceWhatsApp automation for store and field teams
How it works

How last night's sales become this morning's order

This is the shape of most replenishment automation we would build for a grocer or wholesaler. The pipeline does the counting and the arithmetic; limits your buyers set decide what is ordered on its own; a person decides everything else.

  1. STEP 01

    Sales and stock land overnight

    POS sales, stock on hand, deliveries received and write-offs arrive per store and per SKU, mapped to one product master so a line has one name everywhere.

  2. STEP 02

    The data is checked first

    Negative stock, missing store files and sales spikes are flagged. A store whose feed failed is treated as unknown, not as zero sales — the most common way automated ordering goes wrong.

  3. STEP 03

    Demand forecast per store and SKU

    Expected sales by day, taking in promotions, public holidays and seasonality from your own history, with shelf life setting how far ahead it is worth looking.

  4. STEP 04

    Order proposed

    Forecast minus stock and stock on order, rounded to case packs and limited by shelf space, shelf life and each supplier's lead time and delivery days.

Automated

Routine lines: ordered automatically

Stable, long-life lines inside the tolerances your buyers set go to the supplier or DC without anyone retyping them.

Human decides

Exceptions: sent to a buyer

Promotions, new lines, large variances and short-life fresh go to a buyer or store manager with the forecast, the stock figure and the reason it was flagged.

Every order explainable

Which sales and stock figures it used, which forecast, and who changed it. Overrides are kept and fed back, so the next proposal learns from your buyers rather than ignoring them.

Guides

Food retail automation, explained: how to start and why it is harder than it looks

The questions grocers, wholesalers and franchise operators actually ask us, answered plainly. Each starts with a short answer you can take into a meeting, then the steps or reasons behind it.

01 · How to

How do you reduce food waste in a supermarket using data?

Short answer

Measure waste per store and per line first, then attack it at the order and the markdown. Most fresh waste is ordered, not created on the shelf. Joining sales, deliveries and write-offs shows which lines lose money after waste, so orders and markdown timing can be corrected line by line.

  1. 1

    Record write-offs and markdowns per SKU and store, with a reason — damaged, expired, quality. Without this there is no baseline. (This and the next four points are from our project experience, not a study.)

  2. 2

    Join them to sales and deliveries so each line shows margin after waste, not just sales.

  3. 3

    Find the lines where waste is consistently high and orders consistently exceed sell-through, and correct those orders first.

  4. 4

    Move markdowns earlier and make them data-driven: discount by remaining shelf life and stock left, not at a fixed time.

  5. 5

    Track the result weekly against the baseline, so the saving is measured rather than assumed.

  6. 6

    Keep the scale in mind: the CSIR estimates 10.3 million tonnes of edible food is lost or wasted in South Africa a year — 45% of the available supply.

Source: CSIR briefing note on food loss and waste (May 2021)
02 · How to

How do you automate stock replenishment for an independent supermarket?

Short answer

Get clean nightly sales and stock data from your POS, set order rules per supplier, and let the system propose orders for every line. Start by automatically sending only stable, long-life lines within limits your buyers set, and route promotions, new lines and fresh produce to a person until the forecasts prove themselves.

  1. 1

    Fix stock accuracy first. Automated ordering on wrong stock figures orders the wrong amount faster. (These points are from our project experience.)

  2. 2

    Build one product master that maps POS codes, supplier codes and case sizes.

  3. 3

    Capture each supplier's lead time, delivery days, minimum order and case pack.

  4. 4

    Run proposals in shadow mode for a few weeks and compare them with what buyers actually ordered.

  5. 5

    Switch on automatic sending for the lines where proposals and buyers agree, and widen the limits only as the results hold.

  6. 6

    Record every override with a reason, and review them monthly.

03 · Why

Why is forecasting fresh food harder than forecasting packaged groceries?

Short answer

Because errors in fresh cannot be stored away. An over-order of tinned food sits in the storeroom; an over-order of lettuce ends up in the bin within days. The CSIR's own assumptions put distribution-stage losses at 5.5% for fruit and vegetables and 7% for meat and fish, against 2% for cereals.

  • Shelf life caps how far ahead you can buy, so the forecast has to be right for each delivery, not on average over a month.

  • Sales of fresh lines move with weather, quality on the day and what is on promotion elsewhere in the store. (This and the next two points are from our project experience.)

  • Many fresh items are sold by weight or produced in store, so the sales unit and the order unit differ.

  • Write-off data for fresh is often the least reliable data a store has, which makes the history the model learns from noisier.

  • The CSIR assumes 5.5% of fruit and vegetables and 7% of meat entering the distribution stage is lost, against 2% of cereals.

Source: CSIR briefing note on food loss and waste (May 2021), Table 1
04 · How to

How do you keep R638 temperature and traceability records without paperwork?

Short answer

Put sensors on chillers and freezers, capture receiving and batch details at the back door, and store everything in one searchable system. R638 requires thermometers on cold units but allows monitoring by the best available method, and requires processing and distribution records kept for at least six months after shelf life.

  1. 1

    Fit temperature sensors that report every few minutes, and alert a named person when a unit drifts — not only when someone checks a clipboard.

  2. 2

    At receiving, capture supplier, product, batch or lot, quantity, date and delivery temperature for perishables.

  3. 3

    Link batches to the stores and customers they went to, so a recall can be scoped in minutes.

  4. 4

    Retain records for at least six months after the product's shelf life, as R638 requires.

  5. 5

    Keep staff training records in the same system, since R638 requires them to be available to an inspector.

  6. 6

    Plan for power and connectivity cuts: sensors that buffer readings and upload later avoid gaps in the log.

Source: R638 of 2018, Regulations governing general hygiene requirements for food premises
05 · Why

Why do independent grocers find it harder to compete with the national chains?

Short answer

Scale shows up in buying terms as well as in stores. The Competition Commission found the top five retailers held about 64% of the market, and that rebates paid to national chains for store-level activities are not offered to the wholesalers serving independents. Independents have to make up the difference through operations, and operations run on data.

  • Concentration: the Inquiry put the top five retailers at about 64% of the market.

  • Buyer power: national chains obtain rebates for merchandising, store openings, advertising, shelf space and category management that independent retailers do not receive, even though they carry the same costs.

  • Sites: long-term exclusive leases kept independents out of many malls, and are now being phased out — Spar committed to end its exclusive leases by December 2026.

  • Systems: chains have central forecasting, replenishment and data teams; most independents run on a POS and spreadsheets. (This point is from our project experience, not the Inquiry.)

  • The gap that can be closed is operational: stock accuracy, ordering, fresh waste and supplier claims.

Source: Competition Commission, Grocery Retail Market Inquiry summary of findings (2019)
Challenges We Solve

The hard problems Food Retail faces

Fresh waste and markdowns

Short-life lines are ordered on habit and marked down late. Without store-level sales and write-off data in one place, nobody can see which lines lose money after waste and markdowns.

Data spread across too many systems

POS, ERP, supplier portals, delivery notes and spreadsheets each hold part of the picture, with different product codes. Reporting becomes a weekly reconciliation exercise instead of a morning dashboard.

Supplier terms and claims

Promotional support, rebates and short deliveries are agreed in one place and claimed in another. Unclaimed money and disputed invoices are common wherever this runs on email and spreadsheets.

Food-safety records

R638 expects thermometers on chillers and freezers, traceability, a recall procedure and processing and distribution records kept for at least six months after shelf life. Paper logs are easy to fake and hard to search when an inspector asks.

By line of business

Where automation fits, by type of food business

A wholesaler serving spaza shops and a QSR franchise group have little in common except perishable stock and thin margins. The first project that pays back is different for each.

First win: Replenishment and fresh

Independent supermarkets and buying-group members

Usually a capable POS and a stretched head office. Store-level order proposals and fresh markdown timing come first, because they touch the two biggest controllable costs: stock that does not sell and stock that is not there.

First win: Order capture and invoicing

Food wholesalers and distributors

Orders arrive by phone, email and WhatsApp from hundreds of small customers. Capturing them into the system automatically, checking them against stock and price lists, and matching deliveries to invoices removes a daily manual load and the errors that come with it.

First win: Franchise reporting

QSR and food-service franchises

Head office needs comparable sales, food cost and waste across stores that may run different POS set-ups. A single reporting layer — with the same definitions for every store — is usually worth more than any model.

First win: Shelf life and cold chain

Fresh produce, butcheries and bakeries

Margins live or die on days of shelf life. Batch and date tracking, temperature monitoring and production planning that follows sell-through are where data earns its keep.

Our Solutions

What we build for you

01

Replenishment and Forecasting

Store- and SKU-level demand forecasts that drive order proposals within limits your buyers set, with exceptions routed to people and every override recorded.

02

Retail Data Platform

POS, stock, delivery and supplier data landed nightly into one tested platform with a single product master — feeding dashboards, forecasting and supplier reconciliation from the same numbers.

03

Cold-Chain and Compliance Records

Temperature sensors, receiving checks and batch records captured automatically, with alerts when a unit drifts and records retained and searchable for inspections and recalls.

Regulatory context

The rules and market shifts South African food retailers are working inside

Food retail is not a lightly regulated business. Hygiene, labelling, consumer protection, privacy and competition law all reach into store operations — and several of them are really record-keeping requirements. Here is what is in place as of October 2026, with dates, so you can check rather than take our word for it.

  1. Published November 2019

    Grocery Retail Market Inquiry final report

    The Competition Commission found high concentration, with the top five retailers holding about 64% of the market, and buyer power: national chains obtain rebates for store-level activities such as merchandising, shelf space and category management that are not offered to the wholesalers serving independent retailers. It recommended ending long-term exclusive leases and support for small and independent retailers.

  2. Spar committed to end them by December 2026

    Exclusive leases in shopping centres

    Following the Inquiry, Shoprite, Pick n Pay and Spar entered consent agreements with the Commission. Their exclusive leases covered close to 2,000 malls and convenience centres. Spar's August 2023 agreement commits it to end exclusive leases by December 2026. For independents, more mall sites become contestable — and a new store needs a forecast and a supply plan from day one.

  3. Regulations of 22 June 2018, in force

    R638: hygiene requirements for food premises

    Food may not be handled on premises without a valid certificate of acceptability. The person in charge must keep training records, maintain a traceability system and recall procedure, and keep processing, production and distribution records for at least six months after the product's shelf life. Chillers and freezers must have thermometers, though monitoring may be done by the best available method — which allows automated sensors.

  4. Deadline extended to 28 February 2025

    Registration of spaza shops and food-handling outlets

    After recurring food-borne illness incidents, government directed in November 2024 that spaza shops and other food-handling facilities register with their municipalities. For wholesalers and distributors supplying that channel, customer records, batch traceability and recall reach now matter commercially as well as legally.

  5. Draft published 21 April 2023 — not final as of mid-2026

    Food labelling: R146 in force, R3337 still a draft

    R146 of 2010 remains the labelling law. Draft R3337 would repeal it and add front-of-pack warning labels; as drafted, date markings could not be removed or altered and stickers would not be allowed for them, which affects in-store relabelling. The Department of Health has not published a final version, so treat timing as unconfirmed — but keep product data in a form that can be reprinted, not retyped.

  6. In force

    Consumer Protection Act: product liability and recall

    Section 61 makes the producer, importer, distributor or retailer liable for harm caused by unsafe goods or a product defect, irrespective of negligence. Section 60 sets up recall systems. A retailer that cannot say quickly which stores received which batch carries that risk without the means to limit it.

  7. In force

    POPIA and loyalty marketing

    Section 69 prohibits direct marketing by electronic communication unless the person has consented or is an existing customer — and then only for your own similar products, with a free and simple way to object on every message. Personalised promotions from loyalty data have to be built around consent and opt-out records, not added afterwards.

Where we start

Engagements we run in Food Retail

01

Store-level order proposals

Nightly proposals per store and SKU from sales, stock, case packs and supplier lead times — routine lines sent automatically within limits, exceptions routed to the buyer with the reason.

02

Fresh markdown timing

Markdown suggestions based on remaining shelf life and stock left, pushed to the department manager's phone in the afternoon rather than discovered at closing.

03

Wholesale order capture

Customer orders arriving by WhatsApp, email and phone captured into the order system, checked against stock and price lists, and confirmed back to the customer — for distributors serving hundreds of small outlets.

04

Supplier invoice and claims reconciliation

Delivery notes, invoices and agreed promotional support matched automatically, so short deliveries, price differences and unclaimed rebates surface as a list rather than as month-end surprises.

05

Cold-chain monitoring and R638 records

Sensors on chillers, freezers and delivery vehicles with alerts to a named person, and temperature, receiving and batch records kept searchable for inspections and recalls.

06

Franchise and multi-store reporting

One reporting layer across stores and POS set-ups, with agreed definitions for sales, food cost, waste and labour — so a franchisor compares like with like by 7am instead of waiting for spreadsheets.

Key terms

Food Retail terms, in plain language

General dealers
Stats SA's category for retail in non-specialised stores, including those where food, beverages and tobacco predominate — in practice, the supermarket and hypermarket chains. It is the largest category in the monthly retail trade sales release.
On-shelf availability (OSA)
The share of time, or of lines, for which a product is actually on the shelf when a shopper looks for it. It differs from stock on hand: stock in the storeroom does not count.
Shrinkage
Stock that is lost between receiving and sale — through waste, damage, theft or administrative error. Fresh departments usually carry most of the waste component.
Markdown
A price reduction on stock nearing the end of its shelf life, used to sell it before it must be written off.
FEFO (first expired, first out)
Stock rotation by expiry date rather than arrival date, so the shortest-life stock is sold or used first.
Certificate of acceptability
The certificate a food premises in South Africa must hold under regulation 3 of R638 before food may be handled there. It is issued by the local authority's environmental health practitioners.
Exclusive lease
A lease clause that stops a shopping centre landlord from letting space to a competing grocery retailer. The Competition Commission's Grocery Retail Market Inquiry recommended they be phased out.
Case pack
The fixed number of units a supplier ships in one case. Order proposals must round to it, which is why a forecast of seven units can become an order of twelve.
FAQ

Food Retail questions

From our blog

Resilient data pipelines built for South Africa's realities5 data quality pitfalls silently killing your workflow automationInternal WhatsApp automation: beyond customer servicePOPIA-compliant WhatsApp automation

Sources & further reading

Related Services
Cloud Data PlatformsDashboards & ReportingWorkflow Automation

Ready to transform your Food Retail data strategy?

Let's discuss the specific challenges your business faces and design a solution that delivers real impact.

Book a free AI assessmentView our services