The Trust Gap
New data paints a sobering picture of where AI-driven customer service actually stands. Forrester found US CX quality fell again to a record low of 68.3 in 2025, while the CFPB reported roughly 6.6 million consumer financial complaints that same year, more than double the prior year. Separately, Salesforce research found that 60% of consumers believe AI makes trust even more critical when dealing with a brand, not less. The pattern across all three is clear: AI is not rescuing broken service; it is exposing it.
Why This Matters for BFSI and Collections
The clearest warning in this data comes from outside financial services but applies directly to it. A tribunal ruled Air Canada was responsible for its chatbot inventing a policy, establishing that a company remains legally accountable for what its AI tells a customer, regardless of whether a human wrote the words. For collections operations, where a voicebot might discuss payment terms, settlement options, or hardship arrangements, that precedent is not abstract. Misinformation from an automated system carries the same liability as misinformation from a human agent. Klarna’s own reversal reinforces the same point from the business side: after positioning its assistant as replacing hundreds of agents, CEO Sebastian Siemiatkowski had to publicly clarify that customers must know there will always be a human available if they want one.
What the Numbers Do Not Say Out Loud
The rising complaint volume is not necessarily evidence that AI performs worse than humans at individual tasks; it is evidence that automation raises the bar for what customers now expect. Gartner’s prediction makes this concrete: by 2027, half of companies that cut customer service staff because of AI will end up rehiring for similar work under different titles. That is an admission that many organizations automated the visible front end without addressing what actually makes AI trustworthy: whether identity, context, and case history survive the handoff between systems and people. Bank of America’s Erica is held up as a model specifically because it stays narrowly scoped and backed by human support.
The Practical Read
For collections and BFSI operations, the operative distinction is not automation versus human service; it is whether escalation is designed in from the start or bolted on as damage control. That means a voicebot handling a payment conversation needs a clear, honest path to a human the moment a conversation moves into complaint territory, emotional distress, or a request the system is not authorized to resolve. As Zendesk CEO Tom Eggemeier put it, AI should be in service to humans. Given the CFPB’s complaint volume and the Air Canada precedent, that is the difference between a deployment that reduces regulatory exposure and one that quietly manufactures it.