Author: Gary Bhomer | Director, Tel-Consult

I’m moderating a panel at the MVNO Nation Conference in Alicante in November: “AI Full 360°: opening your eyes to the full potential of AI for MVNOs”. The panellists are already adopting AI, and I expect their perspectives to differ sharply. This post is the context I’m bringing to the session: five practical lessons on where an MVNO should start, and what to get right before the tools arrive.

The appetite is clearly there. A recent HCLTech and Mobile World Live survey of nearly 200 senior telecom executives, MVNOs included, found 60% see AI as a key driver of future revenue — yet only 25% believe their organisation is ready to operationalise it at scale. For MVNOs that gap matters more than for most: we compete on speed and focus, and AI that delivers smarter, faster decisions is moving quickly from advantage to table stakes.

1. Question the work before you automate it

The common starting point is to buy the technology, roll out a copilot and run workshops to find use cases. Understandable — but it produces small, disconnected gains, because it automates today’s processes rather than questioning them. There’s little point in speeding up a process that was broken to begin with. The better question: if we were building this business from scratch today, would we design the work this way?

Telstra shows where this is heading. Using Salesforce’s Agentforce platform, it now runs an AI agent through the interim service refund process for customers waiting on a fixed-line repair — validating eligibility, calculating the refund, monitoring for restoration, then applying the refund and updating the customer record. Telstra says Agentforce already supports more than 1,000 contact centre representatives and covers millions of interactions a year. Note that Telstra describes this as a guided workflow in which its people remain accountable, not an unsupervised bot.

The lesson is about scope. Agentic AI has moved from answering questions to executing a whole multi-step process, trigger to closed record. That’s the shape of opportunity for MVNOs: end-to-end workflows that today bounce between systems and people — outage credits, billing disputes, porting and SIM-swap checks, plan changes. Work that once consumed heavy IT and operations effort is becoming economically viable, and MVNOs often have less legacy process to unpick than incumbents.

2. Move from reporting to acting early

Most operators have BI tools, but BI is retrospective: it tells you what happened last month. Someone then interprets it, agrees a response and pushes it through systems and teams that don’t talk to each other. By the time action lands — if it lands — the moment has passed.

The shift worth making is from reactive reporting to predictive action: acting on the signals that come before an outcome rather than analysing the cohort that has already left. Churn is the obvious case — falling usage, a failed payment, a spike in care contacts, a port-out enquiry or a poor network experience can each flag a customer at risk well before they go. Vendors like SourseAI are focused on exactly this.

3. Fix the data foundation

None of this works without accessible, trustworthy data — and this is where most telcos and MVNOs struggle. Customer, billing, network and marketing data sit across separate platforms, so simple questions take days. For an MVNO the problem is compounded: the data is split across your own BSS and CRM, your MVNE or platform provider, and your host MNO.

Lycamobile shows what’s possible. Its partner Cloudwick built a central data lake and data mart to create a single source of truth, and reports that average query time — gathering, querying and dashboarding — fell from 4.5 hours to five minutes. It’s a vendor case study, so treat the number as indicative, but the direction of travel is the point.

Two practical points. First, you rarely need to replace your BSS: you can adopt AI by adding a data and intelligence layer on top. Second — and this is the one I’d underline — look hard at the data access your wholesale agreement gives you. If usage and network data from your host MNO arrive late, aggregated or not at all, that will cap what any AI can do for you. Make data access a commercial term, not an afterthought.

4. Match the tool to the decision

Not all AI is the same, and using the wrong kind in the wrong place is an expensive mistake. Large language models are excellent for language tasks — drafting content, summarising conversations, helping agents find answers — but they can be inconsistent and can confidently invent answers. Where real money or compliance is at stake — pricing, credit, fraud, eligibility, refund calculations — use deterministic, explainable models that give the same answer every time and can show their working.

A simple test: if you would have to explain the decision to a customer, a regulator or your CFO, you need to be able to trace how it was made. Accenture’s Dan Rice describes the settling operating model as the agent proposing while the human stays in the lead, and regulation is heading the same way — the EU AI Act sets transparency and human oversight requirements, most strictly for high-risk uses. It is far easier to build explainability in now than retrofit it later.

5. Put the data to work across the whole base

An ex-Vodafone colleague of mine framed the customer lifecycle as IBRO: Inflow, Base management, Retention, Outflow. With clean data and sound processes you can make better decisions at every stage — which customers to acquire and through which channel, how to grow value in the base, who to save, how to handle those who leave. The aim is to optimise acquisition, retention and ARPU together, not in silos.

The relatively easy start is marketing mix modelling (MMM) on the acquisition side: one of the fastest, most measurable wins, and it doesn’t need the breadth of data retention modelling does. Australian MVNO More used MMM to spot channels with diminishing returns and reported a 24% reduction in digital cost per acquisition over a single quarter. That’s one operator over one quarter, and I’m close to it as More works with SourseAI, but it shows how quickly the discipline can pay back. From there, extend the same foundation into retention: act on the early signals from lesson 2 before a customer ever shows up as outflow.

Questions for the panel

These are the questions I’ll be putting to the panel — and the ones I’d suggest any MVNO asks itself:

• If you were designing your MVNO from scratch today, which single workflow would you hand to an AI agent first — and why that one?
• Where does the data your AI depends on actually live — your BSS, your MVNE’s platform or the host MNO — and who owns the output?
• What has AI cost you that the business case never predicted?
• Where would you refuse to let an AI decide, even if it could?

If you’re attending

Day 1, Tuesday 3 November: I chair the “AI Full 360°” panel at 12:40, then host the “Show me the value” Changemakers Roundtable at 16:30. Come and say G’day — I’m looking forward to meeting industry friends and colleagues.

Sources

• Survey: HCLTech and Mobile World Live, Telecom Pulse Survey (August 2026)
• Telstra: Salesforce, “Telstra uses Agentforce to improve customer experience”
• Lycamobile: Cloudwick case study
• More: TelcoNews, “More deepens SourseAI deal to cut digital CPA costs”