Research agenda 2026/27: seven questions for South African commerce

1. Why publish an agenda

Research that decides its questions after seeing the data finds patterns that are not there. Publishing the agenda first commits us to the questions, and lets merchants, founders and researchers tell us where we are wrong before we start. It also invites collaboration: several of these questions are better answered with partners than alone.

2. Question 1: AI for commerce

What if every product page wrote, tested and improved itself? Most online stores write product pages by hand, test almost nothing and guess at what sells. AI can do a lot of that work, but only when it knows the store, the shopper and how South Africans actually search and speak.

Hypotheses to test:

• Can AI write product pages that sell as well as a good copywriter, in the store’s own voice?
• Can a small store run tests all year round without a data team?
• Will South African shoppers use an AI shopping assistant, and in which languages?

Methods. Controlled experiments on live product pages (the original page against an AI-written version), with conversion rate as the one number, run for long enough to cover at least one payday cycle.

What would count as an answer. AI-written pages that beat the originals on conversion, across more than one store and category.

3. Question 2: Checkout & payments

Why does paying still lose one shopper in every few? Every step between “add to cart” and “paid” costs shoppers: long forms, delivery costs shown too late, card payments that fail at the bank and one-time PINs that never arrive. Each fix is small. Together they decide whether a store makes money.

Hypotheses to test:

• Does showing the delivery cost on the product page lift completed orders?
• Can one saved checkout work across many small stores, the way it does on big platforms?
• How many failed card payments could be saved by offering a second way to pay on the spot?

Methods. Funnel analysis of where payments fail, interviews with merchants and shoppers, and small tests of changes such as showing delivery costs earlier or offering a second way to pay after a failed card.

What would count as an answer. A change that measurably lifts completed orders, and a clear cost per saved order.

4. Question 3: Retail media

How does a local brand buy shelf space on a screen as easily as in a store? Big retailers now sell advertising on their apps, websites and in-store screens. Smaller brands find it hard to buy, and smaller retailers have no easy way to sell it, even though their shoppers are just as valuable.

Hypotheses to test:

• Can a group of independent stores sell ad space together, as one network?
• What would a self-serve way to book in-store screen time look like for a small brand?
• Can we show that an ad sold a product, not only that someone saw it?

Methods. Interviews with brands and independent retailers, a test that sells screen and app space across a small group of stores, and a measure of sales in store, not only views.

What would count as an answer. Brands paying again after a first campaign, based on sales they can see.

5. Question 4: Social & chat commerce

What does a store look like when it lives inside WhatsApp? Plenty of South Africans already buy through WhatsApp, Instagram and TikTok messages. The tools behind those chats are thin: no live stock, no easy way to pay, no order history, and sellers juggling it all on one phone.

Hypotheses to test:

• Can a whole store, from catalogue to payment to tracking, live inside one chat?
• Does a short video from a real seller sell better than a studio shoot?
• How do sellers keep track of orders that arrive across five different apps?

Methods. Diary studies with sellers who trade in chat apps, and a prototype that takes a small catalogue, orders and payment inside one chat, measured by orders completed without a person stepping in.

What would count as an answer. Sellers taking most orders through the prototype, and shoppers coming back to buy again in chat.

6. Question 5: Commerce data

What would a small store do with the data only big retailers have? Big retailers know what sells, where and when, and plan around paydays, school terms and holidays. Small stores sit on the same kind of data in their tills and online stores, and rarely have the time or tools to use it.

Hypotheses to test:

• Which three numbers would change how a small store buys stock?
• Can shared, anonymous data from many small stores tell them more than one big retailer knows?
• Can we forecast demand for a single store around paydays, school terms and public holidays?

Methods. Working with the till and store data of a small number of merchants to test which few numbers change buying decisions, and whether demand around paydays and holidays can be forecast for a single store.

What would count as an answer. Merchants changing what they buy because of the numbers, and fewer stock-outs of their best sellers.

7. Question 6: Township retail

How do we bring spaza shops the tools that chains take for granted? Spaza shops and informal traders are where many South Africans shop every day. They often run on cash, a notebook and a trip to the wholesaler, without the ordering, credit and stock tools that chains rely on.

Hypotheses to test:

• Can ordering from a wholesaler be as easy as sending a voice note?
• What would stock credit look like if it were based on a shop’s real sales?
• Can a spaza shop earn more as a pick-up point for online orders?

Methods. Field interviews with spaza owners and wholesalers, observation of restocking trips, and a voice-note ordering prototype measured by repeat orders after the first month.

What would count as an answer. Shop owners ordering again, unprompted, after the first month.

8. Question 7: Commerce infrastructure

Who builds the warehouses, cold rooms and pick-up points African commerce still needs? Much of what holds South African commerce back is physical: deliveries that fail, stock that spoils without cold storage, stores that close in power cuts, and too few places to shop where people live. AI can’t copy a well-placed warehouse. We test these ideas small, with shared or rented assets, before anything is bought.

Hypotheses to test:

• Can a network of spaza shops and small stores work as pick-up points that cut failed deliveries?
• Would small food brands pay for shared cold storage by the shelf, instead of renting a whole room?
• Can backup power sold as a monthly service keep small retailers trading through outages?

Methods. Pilots with shared or rented assets before any purchase: pick-up points in existing stores, shared cold storage by the shelf, and backup power sold monthly, each measured by use and willingness to pay.

What would count as an answer. Merchants paying for shared assets at a price that would cover owning them.

9. How results will be reported

Each experiment is run under the four-gate method (AVWP 2026/01) and reported whatever its outcome, as a short research note: the question, the bet, the test, the number and the call. Where merchants take part, their data is used only for the study, under POPIA, and reported only in aggregate or with their permission.

References

1. Artive Ventures Lab (2026). The four-gate method: testing commerce ideas before building them. Artive Ventures Working Paper AVWP 2026/01. Johannesburg: Artive Ventures.