State of the Contact Center 2026

Key findings at a glance

• Hunting time fell 54% year-on-year. The time customers spend in routing before reaching a queue dropped from 5.15 minutes in 2024 to 2.37 minutes in 2025, as organisations replaced static IVR menus with CRM-native and conversational AI routing.
• Connection rates climbed 8.1 percentage points, from 52.5% to 60.6%. More queued calls are reaching a human, not fewer.
• Average queue ringing time fell 12% (0.90 min → 0.81 min), confirming the routing gains aren’t being absorbed elsewhere in the journey.
• Outbound is the larger half of the contact centre. 60% of calls across the dataset are outbound. The “contact centre as inbound queue” model describes a minority of activity.
• Voice is growing, not retiring. Cross-vertical call volume rose 16.1% year-on-year between 2024 and 2025, and active agent headcount rose 17.6% — voice is being augmented by digital channels, not replaced by them.
• Agents are not (yet) underwater. Headcount has grown roughly in line with volume. The average contact centre agent handles around 346 calls per month.
• 76% of leaders have formally adopted a Human-in-the-Loop model. “Total automation” is, for now, a rejected position.
• Sentiment toward AI rises with seniority. Scepticism rises with proximity to the customer. Executives are optimistic about ROI; frontline agents are sceptical about reliability. Both views are correct about different parts of the picture.

Methodology

We combined two proprietary datasets:

• Operational call data. An analysis of 58.2 million calls handled across the Natterbox customer base in 2024 and 2025. All metrics — hunting time, ringing time, connection rate, talk time, call direction, missed calls, queue abandonment — are derived directly from our telephony records.
• Voice of the Contact Center research. A structured survey of 178 contact centre leaders conducted in late 2025, spanning C-suite executives, Directors, Managers, and Team Leads with operational or decision-making relationships to their telephony platform.

All figures are drawn from production telephony data and primary research. No third-party industry estimates have been used.

2025 benchmarks for contact centres vs. 2024

The headline of the year is that the routing bottleneck — the part of the customer journey that has been quietly broken for two decades — finally moved.

Reading the data. Hunting time is the metric to watch. A 54% fall in a single year is not a story about agents working harder; it’s a story about callers spending less time inside menus that were never built for them. The shift is being driven by two architectural moves: CRM-native routing (using identity and history to route on the first ring) and conversational AI (replacing keypad trees with natural-language intent capture).

The fact that connection rate improved by 8.1 points alongside falling hunting time matters. These are not zero-sum gains. Customers are reaching a human faster and more often.

The case studies in the dataset trace the same pattern. BuyaCar reduced call abandonment by 33% and cancellations by 25% by replacing menu-based routing with direct routing to dedicated account managers. The mechanism — match the caller to the right human before the conversation begins — is the same mechanism driving the cross-vertical numbers.

What was the role of the contact centre in 2025?

The contact centre was not the inbound complaints department most tooling assumes it to be.

Six in ten calls leaving the contact centre are initiated by the agent, not the customer. That ratio inverts the assumption baked into most contact centre architecture, in which IVR depth, hold music, and callback offers solve the central problem.

The inbound 40%. Inbound has become the highest-stakes surface a brand owns. By the time a customer chooses voice today, they have usually exhausted self-service or they’re carrying enough emotion that a chatbot won’t do. Inbound agents are inheriting harder problems by default — the easy ones never reach them. Routing improvements matter most here, because the cost of a poorly routed inbound call compounds: longer hunt, longer queue, lower connection rate, higher abandonment.

The outbound 60%. Outbound has narrowed in character even as it has held its share of volume. Cold-dialling has fallen across most segments; outbound contextual calling has risen — proactive service, scheduled callbacks, follow-up on a digital signal, retention motions, renewal conversations. When every minute of outbound is competing with email, SMS and WhatsApp for the same outcome, the calls that survive are the ones where voice is genuinely the best channel.

The structural implication. A telephony stack optimised exclusively for inbound queue management is solving the smaller half of the problem. The larger half — outbound dialler logic, click-to-call from CRM records, conversation logging, compliance trail on every agent-initiated call — sits in a different design conversation entirely.

How efficient was voice customer service in 2025?

By the standard benchmarks — speed to connect, time in queue, connection rate — voice was more efficient in 2025 than at any prior point in the dataset. The numbers in summary:

• Hunting time: 2.37 minutes (down from 5.15 in 2024).
• Average ringing time: 0.81 minutes (down from 0.90 in 2024).
• Connection rate: 60.6% (up from 52.5% in 2024).
• Average calls per agent per month: ~346 (cross-vertical).

Three structural factors are doing the heavy lifting:

• Routing has moved from menu-based to AI-driven. Hunting time fell 54% in twelve months. That is the single largest year-on-year movement in any operational metric in the dataset. Identifying the caller via existing CRM records and using voice AI to interpret intent in natural language has eliminated most of the IVR menu tree that historically defined the wait experience.
• More work is happening before the conversation, not during it. Authentication, context-loading, intent capture, and screen-pop now happen upstream of the call. The agent picks up a call already partially solved.
• Outbound calls are pre-qualified by definition. When the agent initiates the call from a CRM record, there is no routing problem to solve. The 60% outbound share pulls the cross-vertical efficiency numbers down by structure as well as by technology.

The efficiency story comes with a caveat. National Dental Care, after deploying AI to analyse 100% of interactions rather than the random 1–2% sample most QA programmes manage, recovered roughly 20 admin hours per month. The point is not that calls became faster; it’s that quality stopped being a sampling exercise. Speed and quality are not the same metric, and a healthy contact centre measures both.

Were contact centres betting on voice as a channel?

Yes. The data shows voice is growing, not declining.

Several common assumptions are worth re-examining in light of this:

• “Digital-first means voice-last.” The data does not support this. Voice volume grew 16.1% across the customer base in the same year that digital channels were also expanding. Voice is not being substituted; it’s being augmented.
• “Younger customers prefer self-service.” True for routine queries — but customers across age groups still escalate to voice for high-stakes interactions: fraud, refunds, complex servicing, complaints, and anything involving perceived risk.
• “Voice will plateau.” No sign of it yet. Headcount growing at 17.6% — slightly ahead of volume — suggests organisations are betting on continued expansion, not managing decline.

The pattern is consistent with what many contact centre leaders describe as the “trust premium” of voice. Customers reach for voice when the stakes are highest. Across the customer base, that’s a growing share of interactions, not a shrinking one.

Are contact centre agents overworked?

The honest answer is: not yet, but the curve is moving the wrong way. The structural numbers are reassuring on their face. Volume grew 16.1% year-on-year. Active headcount grew 17.6% over the same period. Agents are not absorbing the surge alone — organisations have hired, broadly, in line with demand, and the average agent handles around 346 calls per month. Two things should temper the reading:

• Hiring at this rate is a function of growth, not a long-term operating model. Adding agents at 17%+ a year is sustainable while volume is also climbing at that pace. It will not be sustainable when volume normalises and the marginal call still has to be handled. The only structural answer is automation of the work around the conversation — routing, transcription, CRM logging, post-call wrap-up — so that a flat headcount can absorb a rising volume.
• Per-agent volume is a misleading single metric. A 346-call-per-month agent dealing with high-stakes, emotionally heavy interactions is not the same workload as a 346-call-per-month agent in low-complexity retail support. As inbound increasingly inherits the hard calls (the easy ones get resolved in self-service), the average call is heavier than it was a year ago, even if the average count looks stable.

Two case studies in the dataset show the workload-relief mechanism working. George P. Johnson saved 10–20 hours per week by automating manual data capture, freeing leadership time toward risk identification rather than data entry. Seattle Sounders FC used AI-derived trend data to drive a 27% productivity increase by validating coaching priorities rather than manually sampling calls. In both cases, AI didn’t replace people. It replaced the work surrounding the work.

What are the attitudes of contact centre leaders and agents toward AI?

Sentiment is overwhelmingly positive in principle. Adoption is overwhelmingly cautious in practice.

The Human-in-the-Loop consensus

76% of contact centre leaders have formally adopted a Human-in-the-Loop model — using AI for high-volume, low-judgement work, and reserving human agents for high-stakes, emotional interactions. The split, by interaction type, is striking in how cleanly it falls along the contour of risk and emotional weight:

The pattern is consistent. Where the cost of being wrong is reputational or emotional, humans stay in charge. Where the cost of being wrong is a re-asked question, AI takes the load. The interesting territory is the middle — automated service requests sit almost exactly on the 50/50 line, which is where most of the operating-model debate inside contact centres is currently happening.

The Trust Gap by role

Sentiment toward AI is not uniform across the organisation chart. It varies — predictably, and consequentially — with how close the respondent sits to the customer.

Two observations worth sitting with:

• Optimism rises with seniority. Scepticism rises with proximity to the customer. Executives see AI primarily as a portfolio-level investment question. Frontline agents see it primarily as a reliability question — because they’re the ones who carry the cost of an AI hallucination into the next call. Both are correct. They’re answering different questions.
• The middle of the org chart carries the most weight. Directors are the gatekeepers of trust in the reporting layer; Managers are the gatekeepers of adoption on the floor. Programmes that win executive sign-off but lose the middle stall quietly. Programmes that win the middle but lack executive cover get under-funded. The pattern across successful AI deployments in the dataset is that they win all four layers, sequentially, on different arguments.

How AI is actually driving efficiency in contact centres today

Four use cases account for almost all of the productivity gain visible in the operational data:

• Pre-conversation routing. Identity-resolution and intent-capture at the start of the call. This is the mechanism behind the 54% drop in hunting time. It is, by some distance, the highest-leverage AI deployment in a contact centre today.
• Quality assurance at scale. Traditional QA samples 1–2% of calls. AI-driven QA reviews 100%. National Dental Care recovered ~20 admin hours per month by replacing manual sampling with automated review. Seattle Sounders FC reported a 27% productivity uplift driven by trend visibility across all calls, not a sampled subset.
• Post-call work. Transcription, summarisation, CRM record-writing, disposition coding, follow-up task creation. George P. Johnson saved 10–20 hours per week of leadership time by automating data capture that had previously been manual. Most agents spend a meaningful share of their shift writing about the call they just had; this is where the next round of per-agent efficiency will come from.
• Multilingual coverage. AI-driven real-time translation and transcription is starting to allow contact centres to operate across languages without staffing every language natively. This is the use case earliest in its adoption curve in the dataset, but it is the one that most directly addresses the executive priority of “global consistency”.

These four — routing, QA, post-call work, multilingual — are the operational pillars where AI is producing measurable contact centre gains today.

What this means in 2026

A few things follow from the data, not as predictions but as structural observations.

• The routing problem is largely solved. Contact centres that haven’t yet moved off menu-based IVR are now operating against a benchmark that is roughly half their current performance. The gap will become harder to defend internally.
• The next bottleneck is the work around the conversation, not the conversation itself. Talk time is not the constraint. Wrap-up, logging, QA, and follow-up are. This is where agent time is recovered, and business-impacting insights are uncovered.
• Voice volume is rising, not falling. Strategy decks that assume voice is a sunset channel are not describing the dataset. Plans for 2026 should assume voice grows.
• Trust in AI will be won or lost in the middle of the org chart. Executives have already approved the budget. Agents will adopt what works. Directors and Managers — the Pragmatists and the Implementers — are the layer where AI programmes succeed or quietly stall.

The story of 2025 was not that AI replaced anyone. It’s that AI removed the work between the customer and the agent, and the agent and the record. The story of 2026 is likely to be how much further that pattern can run.