Front page
NewsContact Center1h · 12:01 GMT · 5 min read
Why Contact Center AI Now Needs an Evidence Layer
New research from IPI and Talkdesk, alongside a subtle AWS workflow update, points to a practical contact center issue: automation is moving faster than the evidence layer, handoff model, and operational controls many enterprises need to prove outcomes.
I’m less interested in whether contact centers are using more AI (spoiler: they are), and more interested in whether they can prove what happened when a customer needed help.
That distinction feels important this week. IPI’s research into vulnerable customer evidence, Talkdesk’s EMEA data on customer experience automation, and a new Amazon Connect email feature all point to the same operational pressure.
Contact centers are not short of automation ambition, but can they capture decisions, manage handoffs, and produce evidence when the stakes are regulatory, commercial, or customer-critical.
IPI Research Exposes the Vulnerable Customer Evidence Gap
IPI’s latest research, conducted with Cavell among 400 compliance, operations, and customer operations leaders at UK retail and financial services organizations, highlights a uncomfortable gap between confidence and retrievable evidence.
According to IPI, 98% of surveyed businesses say they are prepared for the Financial Conduct Authority’s increasing scrutiny of vulnerable customers. Board-level attention is clearly there, with 94% saying vulnerable customer scrutiny has been discussed at board level in the past six months.
The operational picture is less reassuring. Only 25% of businesses have vulnerable customer interactions automatically logged and searchable. The remainder rely on manual notes, records spread across multiple systems, or inconsistent capture.
That matters because Consumer Duty is not just a policy exercise. The FCA expects firms to demonstrate good outcomes, including how vulnerable customers were identified, supported, and treated across the interaction.
IPI’s findings suggest many organizations would struggle to do that quickly. Nearly two-thirds of respondents said they would need three working days or more to retrieve a single vulnerable customer interaction record for a regulator. Just four percent said they could do it on the same day. Annabelle Goymer, Head of Customer Success at IPI, framed the operational issue clearly:
“What regulators are looking for is evidence of good outcomes: that the customer was understood, supported appropriately, and left better off for the interaction.”
For contact center leaders, the lesson here is to design service workflows where evidence is created as part of the interaction, not reconstructed after the fact.
That becomes even more important as buy now pay later regulation expands the number of journeys that may require a clear affordability, vulnerability, complaints, or support record.
Talkdesk Data Shows AI Adoption Is Outrunning Coordination
Talkdesk’s State of Customer Experience Automation in EMEA adds a second piece to the same puzzle. AI adoption is moving, but not evenly, and not always with the operational foundations needed to support it.
The report says 30% of EMEA organizations use AI to resolve more than 40% of customer issues, while 17% have yet to use AI to resolve any. That split is interesting, but the more useful signal sits underneath it.
Talkdesk found that 59% of EMEA organizations using AI agents say human agents now handle more complex cases. Yet only 13% coordinate AI and people dynamically with orchestrated handoffs between them.
That is the contact center problem in miniature. AI can remove simpler contacts from the queue, but the remaining work often becomes more judgment-heavy, more sensitive, and more dependent on context. Pedro Andrade, Global Head of AI at Talkdesk, argued that leaders need to look beyond initial deployment:
