In 2017, at the FTTx Council, we put a deliberately awkward question to a room full of operators and investors: “You laid the fibre. How do you operate and monetise it?”
At the time it landed as a provocation. The industry was in a build frenzy, and operations were something you worried about later. Nine years on, the fibre got laid. The awkward question didn’t go away. It moved into the P&L.
Look at what happened next in South Africa — the most developed fibre market on the continent, and therefore the most flattering case available. Vumatel has passed 2,040,231 homes and connected 864,208 of them: a take-up rate of 42%. Openserve has passed 1,454,000 and connected 756,409 — 52%. Herotel: 585,981 passed, 293,036 connected — 50% (operator disclosures, 2025).
Read that again. In the best market on the continent, the best operators are earning revenue on roughly half the footprint they paid to build. The other half is capex in the ground, fully spent, returning nothing.
Widen the lens and it gets starker. South Africa has just 5.35 fixed broadband subscriptions per 100 people. Kenya has 3.04. Côte d’Ivoire 1.60. Ethiopia 0.59. Nigeria — a market of over 200 million people — has 0.08 (World Bank, 2025). This is not a civil-works problem. You are looking at an operations problem.
The gap between a home passed and a home connected, billed, working and retained is where fibre business cases go to die. That gap is not closed by trenching. It is closed by the stack — and the cost of the gap is far larger than most boards realise.
The missing middle
Most of what is written about OSS/BSS addresses one of two audiences: the tier-1 group running a multi-year transformation, or the start-up ISP shopping for billing software. The operators who actually have a problem sit between them.
Call it the missing middle: the WISPs and fibre operators who have scaled past roughly 100,000 subscribers and are heading towards half a million. These are not small businesses. They are serious infrastructure platforms — usually private-equity or DFI-backed, often the consolidation vehicle in their market, carrying an aggressive build plan and a board expecting a return. Almost all of them are running on systems designed for a company one-fifth their size.
That is the pattern we see repeatedly in diligence. The stack was chosen when the business had 20,000 customers and a founder who knew every site personally. It was pragmatic then. At 250,000 customers it is the single largest constraint on the business — and, critically, nobody notices the moment it stops working, because it degrades rather than fails. Provisioning takes a little longer each quarter. The billing exceptions queue grows. Nobody can explain why churn is up in one region. Field costs creep. The board asks for a capex ranking and receives an opinion.
The tier-1s have this problem too, but they have nine-figure programmes and armies. Vodacom spent ZAR 20.3 billion of capex in FY2025 against ZAR 152.2 billion of revenue; Airtel Africa $471 million against $4.96 billion (company annual reports, FY2025). The missing middle has neither the budget for that nor the option of ignoring it.
What it is actually costing you
Let’s put numbers on it, because this is where the conversation usually fails. Take an operator with 250,000 subscribers at roughly the South African 5GB basket price of ZAR 309 a month (ICASA, 2025) — an annual revenue base of about ZAR 930 million. The figures below are illustrative, but every input is sourced.
Churn. A single percentage point of churn is 2,500 customers — around ZAR 9.3 million of annualised revenue walking out of the door. Three points is ZAR 28 million. Proactive assurance — knowing a service is degrading before the customer calls — is the difference between a renewal and a cancellation.
Revenue leakage. The quiet one, and the biggest. PwC research indicates many operators lose 3–8% of total revenue to billing errors, fraud and reconciliation gaps; broader industry estimates range from 0.5% to 10% depending on methodology. On a ZAR 930 million base, that is ZAR 28–74 million a year — services delivered but never billed, discounts that outlived their promotion, provisioned circuits with no contract behind them, partner settlements nobody reconciles. Leakage does not announce itself. The revenue line is simply smaller than the network you built should produce.
And note the margin arithmetic: recovered leakage is revenue you have already earned and already incurred the cost to deliver. It drops to EBITDA at close to 100% incremental margin — unlike new subscribers, who cost you acquisition and delivery.
Field economics. Field-service benchmarks suggest roughly 14% of truck rolls are avoidable, and about 33% of service queries could be resolved without dispatching anyone. Across dispersed African geographies — where a truck roll is not a 20-minute drive — this is one of the largest controllable lines in the business.
Capex misallocation. The one that matters most to investors and that almost nobody does well. “Where should the next rand of capex go?” is answerable only if you can join network data to revenue data — utilisation and degradation on one side, ARPU, churn risk and demand on the other. Most operators cannot join those datasets at all, because the OSS and the BSS were never designed to talk. So capex gets allocated by intuition, by whoever argues loudest, or by wherever the last complaint came from. With SSA fixed and fibre capex running at roughly $2.4 billion in 2025 (GSMA), the misallocation cost across the region is not small — and inside a single operator it is the difference between a build that compounds and one that dilutes.
Now convert it to enterprise value. Emerging-market telecom services trade at around 6.7x EV/EBITDA (Damodaran, 142 emerging-market firms, January 2026). Recover even half of a conservative 3% leak — roughly ZAR 14 million, at near-100% incremental margin — and you have created something like ZAR 94 million of enterprise value from work that involves no trenching, no spectrum and no new customers. Retain one point of churn, at a sector EBITDA margin around 36.5% (Vodacom, FY2025), and that is another ZAR 23 million of value.
This is the reframe. At 8,000 subscribers, a systems upgrade is a cost you defer. At 250,000, the stack is a revenue instrument and a valuation lever — and the annual cost of a serious platform is routinely smaller than a single point of avoidable churn.
Why the arithmetic is harsher in Africa
There is a reason this bites harder here than in Europe or North America: affordability makes service quality existential.
A 5GB fixed broadband basket consumes 3.36% of monthly GNI per capita in South Africa and 3.65% in Ghana — but 12.29% in Kenya, 15.47% in Nigeria, 21.11% in Malawi and 44.21% in Mozambique (ITU, 2025). Where a customer hands over a tenth to a sixth of their monthly income for connectivity, tolerance for a service that degrades, drops or bills incorrectly is close to zero. There is no goodwill buffer, and no second chance. Churn assumptions imported from European fibre models do not survive contact with these markets.
Where the stack actually is — and where the industry pretends it is
The journey has four steps. Fault management — the network breaks, an alarm fires, someone reacts. Service assurance — shifting from “is the element up?” to “is the customer’s service actually working?”, which requires correlating network state with service state with customer identity. Proactive and predictive operations — anticipating degradation before the customer notices. And the stated destination: autonomous networks, self-healing and closed-loop.
TM Forum frames this as six levels, from Level 0 (fully manual) to Level 5 (full autonomy). Level 4 — the one everyone talks about — means proactive, predictive, multi-domain, self-healing.
Here the narrative and the evidence diverge sharply. A Bain–TM Forum survey reports around 20% of operators reaching Level 4 or 5 in selected domains. A TM Forum regional benchmark places only about 4% at Level 4 overall. An Accenture study found 79% of telcos still at Level 0 or Level 1, with just 22% expecting Level 4 by 2030.
The honest conclusion: autonomous operations are real in narrow domains and rare in the round. The marketing has run well ahead of the deployment — which is exactly why the ground is open to operators willing to be pragmatic rather than fashionable. TM Forum associates autonomous maturity with up to a 55% reduction in operations and maintenance costs and a 71% improvement in customer satisfaction. Note “up to”: a ceiling observed in favourable conditions, not a promise.
The new generation: agentic AI, and why it changes the maths for the middle
2026 is the year AI stopped being a feature bolted onto OSS/BSS and started becoming the operating layer. Amdocs launched aOS, an agentic operating system designed to sit on top of any existing BSS/OSS stack. Ericsson made AI agents core to its OSS/BSS. NVIDIA is powering long-running agents piloted for self-healing operations. The emerging pattern is the multi-agent NOC: specialised agents that detect, diagnose, orchestrate, execute and verify, in parallel and continuously.
Strip away the launch language and the useful capabilities are narrower and more concrete: assisted troubleshooting, predictive maintenance, anomaly detection, ticket correlation, optimisation recommendations. Early reported results cluster around 30–40% reductions in mean time to repair, 20–25% fewer duplicate and false-positive tickets, and 15–20% lower field dispatch costs. Those are vendor-reported and directional rather than bankable — but they map precisely onto the lines quantified above.
The strategic point for the missing middle: AI is the first operational technology that does not require tier-1 scale to benefit from. A 24/7 NOC roster costs roughly the same whether it watches 100,000 subscribers or 5 million. Sophisticated operations used to demand a headcount base only the giants could justify. A multi-agent NOC does not care about your subscriber count. The scaling operator’s disadvantage was never technology. It was payroll.
And 100,000–500,000 subscribers is the one band where this genuinely works: enough data volume for the models to be useful, enough cost base for the savings to matter, and none of the legacy estate that makes tier-1 transformation take four years.
Does it make commercial sense? The honest answer
The economics have shifted. When we argued in 2017 that growing operators should consume managed or as-a-service OSS/BSS as opex rather than buy it as capex, that was contrarian. Today an entire category exists to serve this buyer, priced to scale with growth, and cloud-native challengers claim to run at a fraction of on-premise cost. The old trap — stay on home-grown systems, or buy a scaled-down tier-1 stack and inherit its rigidity, two-year cycles and integration tax — is no longer the only option.
But pricing is opaque, and that is worth naming. Almost no vendor publishes per-subscriber pricing. Buyers cannot benchmark what they cannot see, and the missing middle — real budgets, but no tier-1 procurement muscle or comparable deal data — routinely overpays, over-buys, or both. When credible analysts cannot agree on the size of the market (2026 estimates range from roughly $28 billion to $58 billion depending on definition), the buyer’s first task is interrogating definitions, not brochures.
And AI does not let you skip rungs. An agentic layer on top of a broken inventory produces confident, automated, wrong decisions — faster.
So the test is not “can we afford this?” At 100,000-plus subscribers you can. The test is: does the annual platform cost come back within a year through retained revenue, recovered leakage and avoided truck rolls? At this scale it usually does, often several times over. The mistake is buying the platform hoping it will impose the discipline. It won’t.
How to go about it at this scale
Start with the inventory, not the software. Know what you have, where it is, and which customer sits on it. At 250,000 subscribers most operators’ inventory is materially worse than they believe. Unglamorous, and almost always the real first project.
Fix fulfilment, assurance and billing as processes before automating them. If the process is broken, automation makes it broken at speed.
Buy the outcome, not the architecture. Consume as a service, pay per subscriber, keep the exit open. A scaling operator’s greatest asset is optionality; a five-year on-premise licence is the fastest way to lose it.
Instrument one revenue-linked use case and prove it. Pick the line that hurts most — usually leakage or churn in a specific cohort. Measure the baseline properly. Deploy narrowly. Prove payback in a quarter. Then extend. Big-bang transformation at this scale is how operators burn two years and a CTO.
And do not chase Level 4. Get assurance genuinely working, billing and inventory clean, then automate selectively where the revenue case is clear. As the old line goes: for every rand of revenue I find, I also find a rand of cost.
Where Digitalthings fits
We have been having this conversation since 2017 — and testing it in diligence ever since.
Digitalthings is a boutique strategy and investment advisory firm specialising in digital infrastructure, broadband networks and services. OSS/BSS is not a footnote in our work; it is a standing workstream.
In due diligence, we review network management and business support systems and, critically, the risks they carry for growth — because a stack that works at 50,000 subscribers and breaks at 300,000 is a valuation issue, not an IT issue. We validate whether the core processes — fulfilment, assurance, billing — actually exist and suit the plan.
In operational assessment and strategy, our workstreams cover deployment, the NOC, the field force, customer services and outsourced arrangements — and whether organisation, staffing and tooling match the growth case. We hold benchmark data on how comparable FTTx operations are genuinely staffed and equipped, which is usually the fastest way to establish whether a business plan is credible.
And in stack strategy and vendor selection, we aim for fit-for-purpose rather than tier-1 bloat. In an opaque, over-marketed category, we help operators specify what they actually need, interrogate vendor claims, benchmark commercial terms, and sequence the roadmap so each phase funds the next.
We do this for the investors underwriting these businesses — private-equity and infrastructure funds, development-finance institutions, banks and operators — across South Africa, Kenya, the DRC, Namibia, Malawi, Tanzania and beyond.
The question we asked in 2017 has simply changed tense. It is no longer “how will you operate and monetise it?” It is: you are operating it now, at scale — do you know what it’s costing you?
Most operators in this band cannot answer that question with a number. The ones who can are quietly outperforming the ones who can’t. If you suspect your stack has become the constraint rather than the enabler, that is a conversation worth having before the next capex cycle rather than after it.
Figures are drawn from Digitalthings’ market-intelligence platform and its underlying sources, including the World Bank, ITU, GSMA, ICASA, Africa Bandwidth Maps, Damodaran and company annual reports, together with published research from TM Forum, Bain, Accenture, PwC, Amdocs, Ericsson and market analysts. The 250,000-subscriber illustration is indicative and based on published basket pricing and sector multiples; it is not a forecast. Vendor-reported performance figures are directional and unaudited. Data is current as of July 2026.