AI Lessons from SA’s 2025 Tech Power Players

South Africa’s online economy hit a different tempo in 2025: fibre consolidation finally moved from rumour to reality, mobile operators started behaving like software companies, and a 17-year IP dispute reminded everyone that “digital innovation” still lives and dies by contracts.

That’s why TechCentral’s South African Newsmakers of 2025 isn’t just a leadership list. Read through it with an AI in e-commerce and digital services lens and you’ll see the real story: the businesses winning aren’t the ones “doing AI”. They’re the ones fixing the plumbing—connectivity, platforms, data access, and governance—so AI can actually drive revenue, reduce costs, and improve customer experience.

Below are the most useful takeaways for founders, e-commerce managers, and digital service leaders heading into 2026—grounded in the same forces behind the year’s biggest names.

The real AI advantage in South Africa is infrastructure

AI outcomes in e-commerce are constrained by three basics: network quality, cloud proximity, and cost-to-serve. If those are shaky, your “smart” features become expensive features.

TechCentral’s list highlights how telecoms leadership shaped those basics. When Telkom integrates mobile, fibre, and data centres under a “OneTelkom” operating model, it’s not just a corporate restructure. It’s a direct enabler of lower-latency customer journeys (faster app experiences), cheaper data distribution (better conversion on mobile), and more predictable uptime (fewer abandoned carts and failed payments).

Here’s the practical link for e-commerce and digital services:

• Fibre reach and pricing influence how many customers will complete a video-heavy product journey (think rich PDPs, live shopping, AR try-ons).
• Mobile network competitiveness determines whether your acquisition strategy can rely on mobile-first experiences without losing users to load times.
• Local data centres reduce latency for personalization, fraud scoring, and real-time recommendations.

If you’re budgeting for AI, don’t start with a model. Start by asking: Is our customer experience bottlenecked by connectivity, hosting location, or data costs? Fix that first.

What to do next (operator-agnostic)

1. Measure mobile conversion by network conditions (2G/3G/4G/5G and low-signal scenarios). AI-driven UX doesn’t help if pages don’t load.
2. Move key AI workloads closer to users (edge-friendly caching, local-region hosting where viable).
3. Treat uptime as a growth lever, not an IT KPI—your AI cannot personalize a session that never starts.

Telecoms consolidation changes the economics of AI-driven CX

Vodacom’s progress on major transactions (including fibre) points to a broader shift: South Africa is moving toward fewer, larger infrastructure plays.

That’s not inherently good or bad for businesses—but it does change how you plan AI-driven customer experience:

• Larger fibre footprints can improve last-mile reliability for customers working and shopping from home.
• Consolidation can also reduce pricing pressure, which affects your ability to offer data-heavy experiences.

For e-commerce teams, the AI implication is simple: your cost to run modern customer experience depends on network market structure.

If you’re building AI-heavy journeys—like conversational commerce, personalized video, or dynamic bundling—plan for two scenarios:

• Scenario A (costs fall): you expand rich content, add more real-time personalization, and test video-led conversion.
• Scenario B (costs rise): you optimize for compression, offline-first UX, and lighter-weight on-device AI.

A useful rule: if your AI feature needs a perfect connection, it’s not ready for the South African mass market.

“Asset-light” models aren’t just finance—they’re AI strategy

Cell C’s turnaround and shift to an asset-light approach (paired with its role as an MVNO host) is a signal of a bigger operating trend: separating ownership from capability.

In practice, that mirrors what’s happening in AI adoption:

• You don’t need to own infrastructure to deliver great digital services.
• You do need tight control over data flows, SLAs, and customer experience.

For many South African retailers and digital platforms, the winning AI stack in 2026 will look less like “build everything” and more like:

• Buy: commodity tooling (product recommendation modules, email send platforms, analytics)
• Partner: connectivity/payment/logistics enablers
• Build: the differentiator (your customer data layer, your decision rules, your merchandising intelligence)

A practical blueprint for e-commerce AI without enterprise budgets

• Phase 1 (0–30 days): Fix tracking, event schemas, product taxonomy, and identity stitching.
• Phase 2 (30–90 days): Deploy AI for service deflection (returns, delivery status), search, and basic personalization.
• Phase 3 (90–180 days): Add margin-aware recommendations, churn prediction, and fraud/abuse scoring.

If you skip Phase 1, Phase 2 becomes a demo. It might look impressive, but it won’t compound.

Media platform battles are teaching retailers a lesson about personalization

The Canal+ / MultiChoice deal drama isn’t only about broadcasting. It’s about platform economics: who owns the subscriber relationship, who controls distribution, and who can afford content.

Retailers and digital service firms face a parallel reality:

• Paid acquisition is volatile.
• Marketplaces and social platforms can change algorithms overnight.
• Consumers are trained to expect “for you” feeds everywhere.

That means your AI strategy needs to prioritize first-party data and retention.

What this means for your e-commerce roadmap

If I had to pick one stance for 2026: stop treating personalization as a nice-to-have. It’s your defence against rising acquisition costs.

Focus on:

• Search that understands intent, not just keywords (synonyms, local language patterns, common misspellings).
• Lifecycle messaging triggered by behaviour (browse abandonment, repeat category interest, price sensitivity).
• Offer governance so your AI doesn’t discount your business into the ground.

A common failure mode: teams deploy “personalized offers” without margin controls, then wonder why revenue grew but profit didn’t.

IP fights (like “please call me”) are a warning for AI teams

Nkosana Makate’s long legal battle over the “please call me” concept is a sharp reminder: innovation value leaks when ownership is unclear.

As more South African businesses use generative AI for content, product imagery, customer comms, and internal code, the IP questions get messier, not simpler:

• Who owns prompts created by staff?
• Can you train models on customer chats under POPIA constraints?
• Are you allowed to reuse supplier images to generate variants?
• What happens when an agency builds an AI workflow on your behalf?

A minimal AI governance checklist (POPIA-friendly)

• Data classification: label what’s personal, sensitive, and non-personal.
• Model boundaries: define what data can never enter third-party tools.
• Content provenance: record whether an asset is human-made, AI-assisted, or AI-generated.
• Contract clauses: specify ownership of workflows, prompts, outputs, and fine-tuning datasets.
• Human review gates: especially for pricing, credit decisions, and customer dispute handling.

This isn’t red tape. It’s how you prevent your future “AI advantage” from becoming an avoidable legal cost.

The best AI use cases in SA e-commerce are still the “boring” ones

Most companies get this wrong: they start with flashy generative AI and ignore the operational AI that actually improves unit economics.

Based on the market signals behind 2025’s biggest tech leaders—cost pressure, consolidation, and competition—here are the use cases I’d prioritize for South African e-commerce and digital services:

1) Customer service automation that protects experience

Aim for a measurable reduction in tickets without annoying customers.

• Delivery tracking answers
• Returns eligibility and status
• Order edits and cancellations
• Product compatibility checks

2) Fraud and abuse prevention for digital payments

South Africa’s digital payments ecosystem keeps growing, and so does fraud pressure. AI scoring should focus on:

• Account takeover signals
• Voucher and promo abuse
• High-risk device/session behaviour
• Suspicious delivery address patterns

3) Merchandising intelligence (margin-aware)

This is where AI pays rent:

• Demand forecasting by region and season
• Stockout risk alerts
• Substitution recommendations when inventory is tight
• Markdown suggestions with profit guardrails

4) Personalization that’s measurable, not vibes

Tie it to conversion, AOV, repeat purchase, and returns rate.

• Next-best product
• Next-best category
• Bundles based on attach rates
• Size/fit guidance to reduce returns

A 2026 plan: build the “AI-ready” digital service layer

If you’re leading e-commerce or a digital service in South Africa, the lesson from 2025’s newsmakers is consistent: execution beats announcements. The businesses that will win aren’t the ones that talk about AI the most—they’re the ones that operationalize it across connectivity, data, and customer experience.

Here’s a simple, field-tested way to structure your next two quarters:

1. Stabilize the journey: uptime, speed, payments, returns.
2. Centralize data: events, product catalog, customer identity, consent.
3. Deploy targeted AI: service, search, fraud, merchandising.
4. Add governance: ownership, POPIA controls, approval workflows.
5. Scale what works: more channels, more segments, better measurement.

The South African market is competitive, price-sensitive, and mobile-first. That’s exactly why AI matters here—but only when it’s grounded in the realities of networks, platforms, and trust.

If this post is part of your planning for the new year, the next step is straightforward: pick one AI initiative that reduces cost-to-serve and one that increases retention, then measure them aggressively forH, you’re already ahead of most teams.

What would change in your business if your top 20% of customers bought just one extra time in 2026—because your digital experience finally felt personal?