Telecoms Policy Clarity: Fuel for AI Commerce in SA

South Africa’s online economy has a quiet dependency that most e-commerce teams only notice when it breaks: predictable, fairly priced connectivity. When the rules are unclear, investors hesitate, networks expand slower, and “digital-first” businesses end up building plans around coverage gaps and cost spikes instead of customers.

That’s why communications minister Solly Malatsi’s recent policy directive to Icasa on BEE and equity equivalent investment programmes (EEIPs) matters beyond the telecoms industry. The telecoms body ACT supports the directive’s intent—but warns the real risk sits in execution, especially around transparency and enforcement. If you’re building AI-powered digital services in South Africa, that warning should land.

This post is part of our series on How AI Is Powering E-commerce and Digital Services in South Africa. The thesis here is simple: AI adoption isn’t only about models and data. It’s also about whether the country’s connectivity market is structured to expand reliably—and whether new entrants are held to consistent obligations.

Why telecoms regulation determines how fast AI e-commerce grows

Answer first: AI in e-commerce scales when latency drops, coverage expands, and unit economics for data improve—and that’s heavily shaped by telecoms policy.

AI-powered retail is more network-hungry than many people assume. It’s not just a “website and payment gateway” world anymore. South African retailers and digital service providers are increasingly running:

• Real-time personalisation (recommendations, dynamic landing pages)
• Conversational commerce (chat assistants on web, WhatsApp, and in-app)
• Fraud and risk scoring (especially for card-not-present transactions)
• Forecasting and replenishment (supply chain planning, route optimisation)
• Creative automation (product copy, campaign variants, imagery workflows)

Even when the AI compute happens in a cloud region or local data centre, your customer experience still depends on the “last mile” and on mobile network performance. If connectivity is uneven, your funnel becomes uneven.

The under-discussed cost: uncertainty

Most businesses can budget for higher data costs if they’re stable. What hurts is regulatory uncertainty that changes who pays what (and when), or creates uneven obligations between incumbents and newcomers. The moment network investment slows or gets redirected, the downstream effect hits merchants: slower rollouts, weaker rural coverage, and higher costs per transaction for data-heavy experiences.

What Malatsi’s directive changes (and what it doesn’t)

Answer first: The directive pushes toward regulatory parity—but Icasa isn’t automatically bound to implement it, and the “how” will decide whether it improves the market or adds confusion.

ACT’s view, as expressed publicly by CEO Nomvuyiso Batyi, is not opposition to the directive itself. The concern is that EEIPs in other sectors have often been opaque—big numbers announced, little detail on outcomes, and limited enforcement when targets aren’t met.

Here’s the core point that matters for digital services: rules that look fair on paper can still be unfair in practice if monitoring is inconsistent.

Regulatory parity: why incumbents keep talking about it

Regulatory parity is basically this: if you operate in the same market, you shouldn’t get wildly different obligations for:

• licensing conditions
• spectrum and licence fees
• contributions to sector funds tied to universal access
• coverage and service obligations

ACT’s members include major operators and network players. Their argument is that multinationals or new entrants using EEIPs should not be able to sidestep obligations that incumbents carry.

For e-commerce and AI platforms, parity matters because it shapes whether competition leads to more coverage and better pricing (good), or whether it leads to regulatory gaming that discourages long-term infrastructure investment (bad).

Icasa’s independence is a feature—and a bottleneck

Icasa is an independent statutory body. It must consider policy directives, but it isn’t obliged to implement them as-is.

That independence protects the regulator from political swings. But it also means timelines are real-world slow. Even ACT’s CEO has pointed out that the directive itself took months to finalise; regulatory amendments can take longer.

If you’re planning product launches for 2026—especially AI features that increase bandwidth use—assume multi-quarter regulatory timelines, not weeks.

EEIP transparency is not a side issue—it’s the whole issue

Answer first: If EEIPs are part of the route to market, they need measurable outcomes, audits, and clear oversight—otherwise they create uncertainty that scares off investment.

ACT’s strongest warning is about transparency and monitoring. The criticism is blunt: in some sectors, you rarely hear about organisations “missing their target” because no one can confidently verify performance.

From a business perspective, this isn’t an ideological debate. It’s an operational one.

What “good” EEIP oversight looks like

If regulators accept EEIPs as an alternative mechanism, then the market needs a system that’s as legible as other licence conditions.

A workable oversight approach typically includes:

1. Standardised reporting (same templates, same definitions, same timing)
2. Independent audits (not just self-reported marketing claims)
3. Outcome metrics (jobs created, SMEs supported, infrastructure enabled—measured consistently)
4. Public transparency (enough disclosure to build trust without exposing trade secrets)
5. Enforcement pathways (clear consequences when targets aren’t achieved)

Here’s the line I come back to: If it can’t be audited, it can’t be regulated.

Don’t dump EEIP policing onto Icasa by default

There’s another important point: it may be unreasonable to expect Icasa to become the primary EEIP enforcement body if that function doesn’t sit cleanly inside its statutory mandate.

If the oversight structure is fragmented—one body sets targets, another issues licences, a third has limited funding—execution breaks down. And when execution breaks down, businesses delay big bets: fibre builds, rural coverage expansions, new network technologies, and the data-centre adjacency that AI services increasingly rely on.

What this means for AI-powered e-commerce teams in 2026

Answer first: Expect AI features to keep pushing you toward “always-on” connectivity, but design your growth plan around compliance, resilience, and cost control—not hype.

This isn’t just policy wonk territory. It affects your roadmap, your margins, and your customer experience.

1) Design AI experiences for uneven network reality

Even if regulation improves, South Africa will still have variability across regions and income segments. That’s normal. Smart teams build AI features that degrade gracefully.

Practical patterns I’ve found work:

• Hybrid experiences: do lightweight personalisation on-device or at the edge; reserve heavy calls for high-intent steps (checkout, account pages)
• Compression-first media: treat images and short video as a performance budget, not a branding afterthought
• Asynchronous AI: for customer support, allow “we’re working on it” flows where responses can take 10–30 seconds without feeling broken
• Offline-friendly journeys: especially for catalog browsing, order tracking, and saved carts

2) Treat connectivity cost as a model input

AI programmes fail quietly when costs are separated: marketing owns conversion, tech owns API spend, finance owns connectivity, and nobody owns unit economics.

Build one combined view:

• cost per AI interaction (chat, recommendation refresh, fraud check)
• incremental conversion impact
• incremental data usage and performance changes

If regulation increases competition and lowers costs, you’ll see it in your margins. If uncertainty slows network investment, you’ll see it in your bounce rates.

3) Watch “parity” as a signal for investment cycles

When obligations are consistent across players, investment is easier to justify. When they aren’t, capital tends to wait.

For e-commerce leaders, that translates into a simple planning rule:

• If the market is moving toward clear parity and enforceable transparency, you can be bolder with AI features that rely on richer media, real-time interactions, and broader geographic reach.
• If the market moves toward opaque exemptions and unclear enforcement, prioritise resilience and cost containment.

4) Don’t ignore universal access obligations

Universal access isn’t abstract. It’s new customers.

As networks expand coverage and affordability, e-commerce platforms see growth in:

• first-time online buyers
• prepaid and mobile-first shoppers
• township and rural logistics innovation

That’s where AI can do real work: language-aware support, fraud controls that don’t over-block, and recommendations tuned for affordability and delivery constraints.

The lead-worthy takeaway: AI growth needs rule clarity, not noise

Answer first: South Africa’s AI commerce opportunity expands fastest when telecoms rules are transparent, enforceable, and consistent—especially where EEIPs are used.

ACT’s support for Malatsi’s directive, paired with its execution warning, is a useful reality check. You can’t build a thriving AI-driven digital services sector on compliance ambiguity. If EEIPs are part of the framework, they need the same kind of visibility and accountability that operators face under licensing conditions.

If you’re building AI into your online retail, fintech, logistics, or customer support stack for 2026, plan for two tracks at once: push your AI roadmap forward, and track telecoms policy signals that affect connectivity, cost, and coverage. The teams that win won’t just have better models. They’ll have better operating assumptions.

If your organisation is planning AI-driven personalisation, conversational commerce, or fraud automation and you want a clear view of the infrastructure and compliance dependencies, map your AI use cases against connectivity realities before you scale. What would you build differently if network performance was your tightest constraint?