Customer care is where telecom AI has moved fastest from pilot to production. It is high-volume, text- and voice-rich, and full of repeatable intents: bills, plan changes, device set-up, faults and moves. It is also where the downside is most visible. A network optimisation model that underperforms costs money quietly; a care agent that loops a frustrated customer through a bot costs money, loyalty and, increasingly, regulatory goodwill.
The industry's expectations are high. In McKinsey research, about 75% of telecom executives said they aim to use agentic AI in customer service1, and TM Forum found as early as 2023 that customer operations was the most widely embraced GenAI use case, with 92% of respondents using GenAI to improve chatbot experiences11. Gartner predicts that by 2029 agentic AI will autonomously resolve 80% of common customer service issues without human intervention, leading to a 30% reduction in operational costs9. The question for operators is what has actually been delivered so far, and what separates the programmes that work.
What has been reported
Credible, published results remain fewer than vendor marketing suggests, but they are accumulating. The strongest share three features: a narrow set of well-understood intents, integration into billing and CRM so the AI can act rather than only answer, and measurement against resolution rather than deflection.
Reported outcomes from AI in telecom care
Selected published results; operators anonymised
| Deployment | Reported outcome | Source |
|---|---|---|
| AI-driven help-desk bot at a telco | 35% lower cost per call; 60% higher customer resolution rate | McKinsey, 2025 |
| GenAI virtual assistant at a European operator | First-time resolution up from 15% to 60%; online NPS up 14 points to 64 | Operator newsroom, 2024 |
| Agentic analysis of all inbound calls at a Latin American operator | Inbound sales up 40% within ten weeks; no negative impact on satisfaction | McKinsey, 2026 |
| Proactive care for customers with open service tickets at a Western European telco | Cost to serve that group down 35%; related inbound calls halved | McKinsey, 2023 |
Note: Self-reported programme results, not independently audited. Sources: [1], [2], [3], [4].
Source: McKinsey & Company, “Scaling the AI-native telco” (2025)
The European operator's case is instructive for its design as much as its numbers. Its GenAI assistant lifted first-time resolution from 15% to 60% and online NPS by 14 points to 64, and it automatically transfers questions it cannot answer to a person who can4. The Latin American example shows the revenue side: before the programme, fewer than 1% of inbound calls were systematically analysed; agentic analysis of every call doubled sales attempts during service calls2. Proactive care, meanwhile, removes contacts altogether: a Western European telco that contacted customers with open tickets before they called halved related inbound calls3.
Looking ahead, most telcos in McKinsey's 2026 research expected at least 10% cost savings in customer support, network and IT within one to two years, and close to 30% by 20302. Those are programme-level expectations, not guarantees, and the evidence on average handle time in particular remains thin: several operators describe AHT reduction as a goal rather than a published result.
Where the results come from matters as much as their size. The published gains cluster in a handful of intents: billing queries, payments, plan and add-on changes, device set-up and ticket status. These are high-volume, rule-bound and verifiable against system data, so an AI agent can both answer and act. Fault diagnosis is harder because it depends on live network and home-network data; moves, cancellations and complaints are harder still because they carry emotional and commercial stakes. Operators that start with the first group, prove resolution rather than containment, and extend only when the evidence supports it tend to build credibility with both customers and front-line staff.
There is also an operating-model shift hiding in these numbers. When routine intents are handled by AI, the contacts left for human advisers are longer, more complex and more emotional. Handle time per human contact can rise even as total cost falls, which confuses dashboards built for the old mix. Advisers need different skills, better tools (AI-generated summaries, suggested actions, real-time policy guidance) and different targets. Quality assurance changes too: rather than sampling a few calls a month, AI can review every interaction, flagging vulnerability, mis-selling and compliance risks for human review.
The customer is not yet convinced
Operators should read the demand side carefully. In a Gartner survey of 5,728 customers, 64% said they would prefer companies did not use AI in customer service, and 53% would consider switching to a competitor; the top concern was that it would become harder to reach a person7. A 2025 cross-industry survey of 5,000 consumers found 88% were satisfied with interactions handled mostly or fully by people, against 60% for AI-driven interactions, and 47% named the inability to reach a live agent when needed as their single biggest frustration with automation8.
There is also a cautionary tale from the labour side. Gartner research published in 2026 found that only about a fifth of customer service leaders had actually reduced agent staffing due to AI, and predicted that half of firms that cut service staff because of AI will rehire by 202710. Care that is designed to remove people, rather than to redeploy them to harder cases, tends to reverse.
The regulator's lens
Telecom complaints are far lower than they were: Ofcom's data for the first quarter of 2026 show 6 complaints per 100,000 fixed broadband customers and 3 for landline, against 35 and 42 in late 20106. Operators have earned that improvement and should not spend it on poorly designed automation. Ofcom's September 2026 research found 8% of online adults already use an AI tool in relation to their telecoms services and 2% to make a complaint; while it found limited evidence that AI has been a systemic cause of harm, it said it would be concerned if providers' AI applications acted as 'gatekeepers' that create barriers to support5. In the EU, Article 50 of the AI Act has required since August 2026 that people are told when they are interacting with an AI system.
Customer use is already mainstream in adjacent channels: Ofcom found that over half (53%) of adult UK internet users had interacted with a GenAI tool or an AI customer service chatbot in the past year, and 6% had used an AI tool to resolve a telecoms issue5. Expectations will rise with familiarity, and so will scrutiny of the moments when AI gets it wrong.
Telecom complaints have fallen sharply; AI must not reverse the trend
Complaints to Ofcom per 100,000 customers, industry average
Note: UK only. Q1 2026 figures for pay-monthly mobile (2), pay TV (3) and pay-as-you-go mobile (1) are not shown.
Source: Ofcom, “Report: complaints about broadband, landline, mobile and pay-TV services” (2026)
What the better programmes do
- Measure resolution, not containment. A contained contact that returns tomorrow is a cost, not a saving. Track first-contact resolution, repeat contacts and complaints per intent.
- Design the handoff. Customers should reach a person in one step on request, with the full context passed across so they never repeat themselves.
- Give agents tools, not only knowledge. The largest gains come when AI can execute: adjust a bill, reschedule an engineer, apply a credit within limits.
- Keep humans on judgement calls. Complaints, vulnerability, disputes and retention offers above a threshold stay with people, supported by AI summaries.
- Analyse every contact. Use AI to read all interactions for quality, compliance and sales opportunity, rather than sampling a few per cent.