AI in retail: from efficiency tool to force multiplier

AI in retail: from efficiency tool to force multiplier

WHEN Catherine Lückhoff speaks about artificial intelligence in retail, she’s quick to dispel the fairy dust expectations. “Everybody wants to know how they can have the fairy dust, but no one’s really putting in the work,” says the co-founder and CEO of data modernisation firm 20fifty.

Her observation cuts to the heart of a fundamental shift happening across South African retail: AI is no longer a futuristic concept but a practical tool reshaping everything from internal operations to customer engagement. The question isn’t whether retailers should adopt AI, but how strategically they’ll deploy it.

Two categories of impact

Lückhoff identifies two distinct categories where AI acts as a force multiplier. The first focuses inward: optimising store layouts, logistics, predictive modelling for stock management, and understanding cross-shopping behaviours. Dynamic pricing, already visible when retailers discount bakery goods at day’s end, represents just the beginning of what’s possible.

The second category is external-facing, centred on hyper-personalisation. However, this requires solving fundamental data challenges first. “You might have two different profiles for the same customer,” Lückhoff explains, describing scenarios where a family shares accounts across online shopping, store credit, and cash purchases. “How do you build up a visual of who the customer is, and then how can you hyper-personalise on top of that?”

Real-world implementation

South African retailers are already pioneering innovative applications. Shoprite Checkers, according to Lückhoff, is piloting AI agents at point-of-sale systems that assist cashiers with edge cases in vernacular languages. When a customer needs to pay a DStv account – not a daily transaction for most tellers – the system provides real-time training support, speeding up checkout and improving the customer experience.

Pepkor is using predictive models for lay-bye customers, factoring in variables like travel distance to stores. The insight: customers living far from physical locations are less likely to complete payments due to transport costs. “Good data analysts probably would have spotted these patterns over time,” Lückhoff notes, “but now suddenly you can see those insights much faster and you can act on them much faster.”

The foundation: Data quality and strategy

Before implementing AI solutions, 20fifty conducts comprehensive assessments: customer personas, pain points, systems integration mapping, data governance frameworks, security protocols, and GDPR compliance. “The technology is just the enabler,” Lückhoff emphasises. “I know I want soup, but I’m certainly not going to use a fork for it.”

The firm is developing a “voice of the customer” tool that aggregates call centre transcripts, product reviews, customer complaints, and public data sources. Rather than traditional sentiment analysis, the system identifies critical issues – whether supply chain, payment, or staff-related – allowing executives across departments to interrogate data through natural language queries.

The ROI challenge

Measuring return on investment remains complex. Old Mutual achieved a 77% efficiency gain on reporting that previously took 22 days monthly, now delivered in real time. For identity management resolution, 20fifty targets 80-90% accuracy, acknowledging the final 10% – particularly cash customers – presents unique challenges.

“It’s very hard upfront to determine what your return on investment is going to be in a space this new,” Lückhoff admits. Success metrics must extend beyond immediate financial returns to consider factors like conversion rates, customer retention, and long-term strategic positioning.

The human element

Perhaps counterintuitively, successful AI implementation requires significant human input. Teams transition from task-based work to strategic thinking and creativity – what Marks & Spencer UK describes as moving from person managers to machine managers.

However, Lückhoff warns about over-reliance: “Junior developers aren’t allowed to use AI at 20fifty because they haven’t built up the experience to critically evaluate outputs.” Recent research suggests developers think they’re 20% faster with AI coding assistance but are actually 19% slower, spending up to 26 minutes correcting AI-generated code.

For South African retail specifically, Lückhoff anticipates AI solutions tailored to local economic realities: optimising for the 60% of customers earning under R10,000 monthly, improving accessibility, and stretching consumer spending power further.

“The genie’s out of the bottle,” she acknowledges. Success will depend on retailers being transparent about AI use, understanding customer pain points, and ensuring benefits flow both ways – not deploying technology for technology’s sake, but solving real problems for real people.