Exclusive Q&A: How MLnetworks is Applying Agentic AI to Africa’s Telecom Infrastructure Challenges

Africa’s telecoms infrastructure is under pressure from every direction: expanding coverage demands, energy costs that eat into already-tight margins, and CAPEX decisions that can make or break a network’s future. Against that backdrop, a growing number of operators are asking whether AI can do more than surface insights on a dashboard. Can it actually help them decide and act?

MLnetworks is one of the companies building toward that answer. The firm develops agentic AI systems purpose-designed for telecoms, giving operators and towercos the capability to move from fragmented data visibility to coordinated, intelligent action across network planning, energy management, and operations. With proven deployments across MENA and a growing focus on African markets, MLnetworks is making a clear case that the infrastructure challenges operators face on the continent are not barriers to AI adoption, but precisely the conditions that make it most compelling.

In this exclusive interview, TechAfrica News Founder Akim Benamara speaks with Jawad Maaloum, CEO and Co-founder of MLnetworks , on what that shift looks like in practice, where the real value sits today, and what it will take for agentic AI to become standard across African telecoms.

Everyone is talking about agentic AI right now, so the term is starting to lose meaning. In plain language, what does it actually mean when MLnetworks says it builds agentic systems for telecom, and how is that different from the dashboards operators are already paying for?

When we say agentic AI in telecom, we don’t mean another analytics layer or smarter dashboard.

We mean a system that doesn’t just show what is happening in the network, but actually understands the network state, reasons about it, and takes or proposes actions that improve performance.

Traditional tools stop at “here is a problem: congestion, outage, high energy consumption.” Operators still have to interpret it, decide what matters, and manually coordinate action across teams.

An agentic system closes that loop. It continuously observes the network, identifies issues or opportunities, simulates options, and either recommends or directly triggers actions like parameter changes, energy optimisation, or investment shifts.

So the real difference is: dashboards inform humans. Agentic systems reduce the dependency on humans for every operational decision.

You like to say telecom doesn’t need more dashboards. When you sit with an operator or a towerco, what’s the real problem you see? Is it that they’re short on data, or that nobody can act on it fast enough?

It’s not a lack of data. Telecom operators already have massive amounts of data across network performance, energy, OSS/BSS, and field operations.

The real issue is that this data is fragmented across silos that don’t operate in a connected or operationally aware way.

Each system sees only a slice of reality: one team looks at performance KPIs, another at energy consumption, another at field tickets, and another at planning. But none of them naturally translate into a unified, actionable view of the network.

So even when something is obvious at a system level, like a site that is underperforming and over-consuming energy at the same time, no single team has the full picture, and no system is responsible for turning that into coordinated action.

On top of that, even when insights exist, the execution chain is slow because it has to cross those same silos manually.

“So the real bottleneck is not visibility, it’s the lack of operational coherence between systems and teams, which makes it hard to act quickly and consistently. What we’re solving is exactly that: turning fragmented intelligence into a unified decision layer that is operationally aware across the whole network.”

–Jawad Maaloum, CEO and Co-founder, MLnetworks

Your platform covers both planning and operations. Take us through a real example, from a network investment decision at the top down to something actually changing at a tower site.

It starts at the top. The operator has a CAPEX budget and a network too large to upgrade everywhere. PowerAI ranks the candidate sites by expected return, weighing performance data, demand forecasts, and local constraints. This site rises to the top of the list, with a clear figure attached to the upside and the cost.

Approval is only half the story. Adding capacity means more load on the rectifiers, more draw on the energy system, and more strain on a site that may already run close to its limit. PowerAI runs that check before a truck is dispatched. If the site needs a battery upgrade or a power adjustment to carry the new load, that requirement surfaces now, not after the equipment is installed and the site starts tripping.

So the same platform that recommended the investment also makes sure the site survives it. Planning and operations work off the same picture, in one continuous flow.

On the continent, energy is often the single biggest operational headache: diesel, grid instability, theft. Is that where your system tends to prove itself first, or are you seeing the bigger wins on the investment and CAPEX side?

In most African and MENA markets, energy is the fastest win.

Operators feel it directly in OPEX. Diesel costs, grid instability, and inefficient sites show up on every monthly bill. The platform finds the waste and tightens consumption, so the return is quick and visible.

The larger prize sits further out, in CAPEX efficiency. Once the planning decisions improve, where to invest, what to upgrade, what to leave alone, the operator stops misallocating infrastructure spend. By magnitude, that pool is far bigger than the energy line, though it takes longer to surface.

Energy is the entry point. It proves value early. CAPEX optimisation is where the structural change happens.

You’ve got live deployments across MENA. Those markets share a lot with parts of Africa: tough terrain, mixed grid reliability, cost pressure. What did those deployments teach you that you didn’t expect?

The biggest surprise is how operationally complex seemingly simple decisions really are in telecom.

Technically, the models work well everywhere. But the constraint is almost never AI performance, it’s data fragmentation, organisational silos, and trust in automation.

Another key lesson is that unstable infrastructure environments actually make optimisation more valuable, not less. When energy or grid reliability is inconsistent, even small improvements in site efficiency have outsized impact.

And finally, we learned that operators don’t want abstract predictions. They want actionable, localised recommendations they can immediately execute at site level.

African operators are working with tighter margins than most, plus rural coverage gaps and infrastructure that’s expensive to maintain. How much of what you built for MENA carries over, and what would you have to rethink for, say, a market like Kenya, Nigeria, or the DRC?

The challenges we see in MENA, cost pressure, uneven grid reliability, remote sites, and the need to do more with limited resources, are very similar to what many African operators face. In fact, those environments are precisely why we built these systems the way we did.

The core capabilities, network planning optimisation, forecasting, energy intelligence, and operational decision support, are market-agnostic. They don’t depend on a specific country; they depend on understanding how networks behave and how infrastructure constraints impact decisions.

What changes from market to market are the local parameters: traffic patterns, energy economics, regulatory requirements, and operational priorities. Those are configurations, not a redesign of the platform.

So for markets like Kenya, Nigeria, or the DRC, we’re not starting from scratch. We’re bringing proven capabilities and adapting them to local realities. The underlying problems are remarkably similar, and the value proposition, making better investment decisions and running infrastructure more efficiently, is just as compelling, if not more so.

Moving from a system that recommends to one that actually decides and acts is a big trust step anywhere, but maybe more so where teams have been burned by tools that overpromised. How do you approach that, and where do you keep a human in the loop?

We don’t ask operators to hand over decision-making to AI on day one. The system starts by making recommendations and explaining why it made them. Humans remain in the loop by validating actions, confirming recommendations, and defining which categories of decisions can be pre-approved.

What’s important is that the system doesn’t stop at making a recommendation. After an action is executed, it remeasures the relevant KPIs and verifies whether the intended objectives were actually achieved.

Over time, it learns from these outcomes and from the operator’s own way of working: their data, their operational mechanisms, their expert decisions, and their business priorities.

So autonomy is progressive. The system continuously adapts to the specific operator and builds a track record of successful decisions over weeks and months. As confidence grows and performance is proven, the level of autonomy can gradually increase.

“In our view, the highest levels of autonomy don’t come from removing humans from the process. They come from creating a continuous feedback loop where human expertise and machine learning reinforce each other until the system has earned the right to act more independently.”

–Jawad Maaloum, CEO and Co-founder, MLnetworks

Looking at the next two or three years, what has to happen for agentic systems to become normal in African telecom rather than something only the big groups can afford?

I actually think Africa is in a position to adopt agentic systems faster than many people expect.

The economics are simply too compelling. Operators are being asked to expand coverage, improve customer experience, manage energy challenges, and control CAPEX and OPEX at the same time. Doing that with more dashboards and more manual processes is becoming unsustainable.

What needs to happen over the next two or three years is not a technological breakthrough. The technology already exists. What needs to happen is a shift in mindset, from using AI as a reporting tool to using AI as a decision-making partner.

The adoption path will be gradual. Operators will start with assisted decisions, prove measurable outcomes, build trust, and progressively increase autonomy. Once they see faster decisions, better investment efficiency, and lower operational costs, the business case becomes very difficult to ignore.

I don’t think agentic systems will remain something only large operator groups can afford. In fact, markets under the greatest cost pressure often have the strongest incentive to adopt them, because every avoided dollar of unnecessary CAPEX and every improvement in operational efficiency has a much bigger impact on the business.

MLnetworks will be at DTW Ignite 2026 in Copenhagen from 23 to 25 June. If you are attending and want to connect with the team, find them there.