Every operator knows it loses money between the network and the invoice. A call that is carried but never rated, a bundle discount that outlives its campaign, an interconnect invoice that nobody reconciles against traffic, a roaming partner that bills for sessions the home network never saw. Individually these are rounding errors. Collectively they are a line item that most boards would find uncomfortable if it were ever reported with the same rigour as capital expenditure.
Revenue assurance (RA) exists to find those losses. Yet after three decades as a discipline, the evidence suggests the industry is still leaking at roughly the same rate, while the products it sells have become far harder to assure: convergent bundles, partner-billed content, enterprise solutions with bespoke pricing, 5G network slices and, increasingly, API-exposed network capabilities sold to developers.
How big is the leak?
Estimates vary by method, but they cluster. TM Forum's Revenue Assurance Survey Report 2019–2020 put global telecom revenue leakage at about 1.5% of overall revenue3. An RAFM survey run by the Risk & Assurance Group (RAG) estimated that the industry lost USD 94bn to revenue-assurance-related leakage in the prior year4, and a later RAG survey, as cited by Subex, put consolidated worldwide losses from operating leakages at USD 149bn6. Fraud is counted separately: the CFCA's 2023 global survey estimated USD 38.95bn lost to telecom fraud, equivalent to 2.5% of industry revenue, up 12% on 20215.
These figures are self-reported and methodologies differ, so they should not be added together. But the direction is consistent with older evidence. In a KPMG survey of 137 telecom executives in 62 countries, more than a third reported leakage above 1% of revenue, and 41% said they failed to identify more than half of their total leakage7. Prepaid, roaming and postpaid were named as the most vulnerable revenue streams7.
Different lenses, the same conclusion: leakage is material and persistent
Selected industry estimates of telecom revenue leakage and fraud loss
| Source and period | What it measures | Estimate |
|---|---|---|
| TM Forum RA survey, 2019–20 | Revenue leakage, share of revenue | About 1.5% |
| RAG RAFM survey, 2020 | Revenue-assurance-related leakage, one year | USD 94bn |
| RAG survey, 2021 (as cited by Subex) | Operating leakage, worldwide | USD 149bn |
| CFCA fraud loss survey, 2023 | Fraud loss, share of revenue | USD 38.95bn (2.5%) |
| KPMG survey of 137 executives, 2012 | Operators reporting leakage above 1% of revenue | More than one-third |
Note: Figures come from separate surveys with different methods and scopes and should not be summed. Sources: [3], [4], [5], [6], [7].
Source: Ericsson, “An AI use case for reducing revenue leakages” (2020)
Where the gaps are
The most useful evidence is not the headline loss but the coverage map. In a RAG survey of 51 businesses that rated themselves against 25 categories in the RAG Leakage Catalog, 20% of telcos had not implemented fundamental RA controls; around 40% reported no gaps in core areas such as billing accuracy, usage reconciliation, tariff management and rating, while roughly 20% rated that coverage as poor, very poor or non-existent2. Fewer than 20% rated their asset assurance coverage as good or very good, and most had limited or zero coverage of leaks in promotions, loyalty benefits, fair-usage policies and contract compliance2.
That pattern matches how leakage has moved. Classic usage leakage between switch, mediation and billing is well understood and often controlled. The newer leaks sit in commercial complexity: interconnect and roaming settlements where partner data never quite matches internal records; product catalogue and order errors where what was sold is not what is provisioned or billed; partner and wholesale revenue shares; and enterprise contracts whose discounts, credits and service-level penalties live in spreadsheets. Bad debt and errors in billing and collections, rating and tariffs, usage and subscription fees were the top five causes in the RAG survey cited by TEOCO4.
The AI gap in assurance
KPMG's 2025 RAFM survey, focused on telecom players across regions and service models, captures the gap between intent and delivery. 76% of respondents believe AI will drive transformation in RAFM operations, and 30% have apportioned budgets for AI-, ML- and RPA-driven business assurance1. But only 14% can currently derive full-scale value from their existing AI, ML and RPA tools, just 30% have ensured data-lake readiness for ML deployments, and 53% cite partial control testing as a major challenge1. Only 12% rate their overall risk coverage as excellent, and 40% of controls are only partially reported1.
Strong belief in AI, weak foundations to use it
Share of RAFM survey respondents, % (2025) (%)
Note: KPMG in India survey of telecom and adjacent organisations; sample size not disclosed.
Source: KPMG in India, “Revenue Assurance and Fraud Management (RAFM) Survey Report” (2025)
The earlier RAG survey told a similar story: fewer than 20% of RA respondents said their teams already used machine learning, 42% had no plans to, and 64% reported a lack of executive sponsorship for RA initiatives4. Where AI has been applied well, it tends to be on narrow, data-rich reconciliation problems. Ericsson describes a Middle Eastern tier-one operator with 12 million active subscribers that used machine-learning anomaly detection on API response times and infrastructure metrics to anticipate CDR leakage between mediation and billing, reaching close to 99.999% billing accuracy on reconciled records3.
The lesson is that AI in assurance is only as good as the reconciled data beneath it. Anomaly detection on a single system finds symptoms; tracing a leak to its cause requires the order, the catalogue entry, the provisioning record, the usage records and the invoice to be joined reliably. Operators that have built that joined view find that AI agents can do much of the investigation work: grouping thousands of breaks into a handful of root causes, drafting correction scripts and estimating the value at stake, leaving analysts to confirm and approve.
What a modern RA function looks like
- Full-population, not sampled. Every usage record, order, partner invoice and settlement is reconciled end to end, continuously, rather than sampled monthly.
- Controls mapped to the product catalogue. Each new tariff, bundle or partner offer launches with its assurance rules, not months later.
- Business assurance scope. Promotions, loyalty, credits, wholesale and enterprise contracts are brought into the same control framework as usage2.
- AI on the exceptions. Models and agents cluster breaks, identify root causes and draft corrections; people approve financial adjustments.
- Reported like any other loss. Detected, recovered and prevented leakage is tracked monthly by the CFO, with named owners per control.
There is also a regulatory and reputational dimension. Over-billing is a leak too, just in the customer's direction, and it is the one most likely to reach a regulator or the press. A control framework that treats billing accuracy symmetrically, catching both under- and over-charging, protects margin and trust at the same time.
Finally, the economics are unusually clear. Unlike most transformation spend, recovered leakage falls almost directly to EBITDA, and prevention compounds. Cross-industry research from BCG found that almost three-quarters of companies lacked automated revenue assurance processes, and that standalone assurance programmes typically yield revenue increases of 3% to 5%8. Telecom's data volumes and product complexity make the case stronger, not weaker.