ID&V: Multi-Billion Pound Bottleneck In Contact Centre Automation

By Sally Hodgin, Principal AI Consultant at Connect

According to ContactBabel’s latest research, 74% of all inbound calls require ID&V and a whopping 91% of those are still handled by a human agent. That figure tends to get a sharp intake of breath when I share it with contact centre leaders, but the real implication runs deeper than the headline cost.

If a human agent is completing ID&V, the rest of that call almost always stays with them. Handing back to an AI Agent after a human has authenticated a customer is technically possible but rarely happens in practice and it’s hard to imagine it makes for a particularly good customer experience. In reality, a human doing ID&V eliminates the possibility of downstream automation and self-service for that interaction entirely.

74% of inbound calls need ID&V. 91% are still completed by a human agent. The downstream effect: no automation for 64% of total inbound call volume.

An underestimated barrier to scaling AI.

This is, in my view, one of the most underestimated barriers to scaling AI in the contact centre. Anything personalised or account-related requires a customer to first be identified and/or authenticated. Without solving for ID&V, the automation opportunity for self-service in these tasks is fundamentally capped.

So what is actually preventing organisations from using AI Agents to automate ID&V? A recurring theme emerges from conversations with customers who have attempted to roll these out: the moment a customer reads out an account number, a postcode, a date of birth, or a vehicle registration, things start to break down.

The root cause? Put simply, a jack of all trades is a master of none. Large language models are built to handle general open-domain conversation, but an ID&V process in a live voice channel doesn’t require your AI to discuss the weather or explain how to make a lasagne — it requires accurate capture of alphanumerics which comes with a very distinct set of challenges. ASR output is inherently messy, customers don’t read out identifiers the way they appear on a statement, they pause mid-string, group digits differently, speak over background noise, and self-correct. Coupled with there being no existing customer record to validate against (since the identifier itself is what retrieves the record) the tolerance for error is effectively zero – one wrong character and the process fails. Breadth and versatility are what make large language models so powerful, but the level of accuracy required for ID&V demands almost the opposite — specialised models built and trained specifically for the task.

Using models trained for high ID&V success rates.

What changes the equation is using specialised micro-models, trained specifically for alphanumeric entity recognition, a specialist ID&V micro-model understands that “double zero”, “oh oh”, and “zero zero” should resolve to “00”. Rather than asking a general-purpose language model to do something it was never optimised for, you deploy a model that exists for one job: extracting account numbers, postcodes, ID strings, dates, vehicle registrations etc. from a transcript with the precision a regulated verification process demands.

Deploying a specialist ID&V micro-model improves verification success rates and, critically, removes the single biggest blocker to self-service automation — because every personalised AI journey starts with a verified customer. That compounding effect is what makes this one of the strongest ROI cases in the contact centre AI space. You’re not just reducing the cost of authentication itself; you’re unlocking an entire category of calls that currently default to an agent the moment ID&V is involved. Shorter queues, lower abandonment, reduced AHT, and agents freed to focus on genuinely complex work. The data to build the business case tends to land quickly and the payback period reflects it.

If you haven’t yet automated ID&V in your contact centre, or you’ve tried and had to roll back due to poor results, get in touch. We’d be happy to walk through how to build and optimise a solution that gets you results.

Source: ContactBabel Inner Circle Guide to Chatbots, Voicebots & Conversational AI — UK Edition 2026