The Sabio Debt-Disclosure Study Is a Pitch for AI Routing, Not Proof It Works
On 28 August 2026, Call Centre Helper published a guest analysis from CX vendor Sabio arguing that contact centres route sensitive conversations — debt, hardship, missed payments — to human agents by default, when the evidence says that's often the wrong call. Citing Grant Thornton's work on FCA data, Sabio notes that 58% of vulnerable customers don't disclose their circumstances to their financial services provider, mostly out of embarrassment or fear of judgement, even when they're already talking to a person. The piece also cites unattributed "industry analysis" claiming 42% of companies abandoned most AI initiatives before reaching production, with 46% of proof-of-concept projects scrapped entirely, and references the FCA's 2026 review of AI in retail financial services warning about systems that influence decisions customers don't adequately understand.
Does AI actually get more honest answers from vulnerable customers than a human agent?
Possibly, on the psychology, but the 58% disclosure gap is evidence of a problem with human channels, not proof that AI closes it.
The social-desirability-bias argument — people under-report shame-laden facts to anyone they think is judging them — is well established, and a voice interface plausibly removes some of that anxiety. But nothing in this piece measures disclosure rates to an actual AI system against the 58% human baseline. It's a plausible hypothesis dressed as a finding, published by a vendor that sells the AI voice channel it's recommending.
Where do the 42% and 46% failure figures actually come from?
The source doesn't say — no named study, no methodology, no date range — which is exactly the kind of number you shouldn't take at face value.
"42% of companies had abandoned most of their AI initiatives before they reached production" is a striking claim to leave uncited in an article about trusting AI with your most vulnerable customers. Before you let a similar stat justify a purchase, ask your vendor: who ran this study, on what sample, and does "abandoned" mean cancelled outright or simply redesigned? Vague provenance on the failure-rate number undercuts the confidence of everything built on top of it.
What should you actually ask before routing sensitive calls to AI?
Ask how vulnerability is detected, what happens on a false negative, and who reviews transcripts for missed distress signals.
Sabio's own piece concedes the risk: automated systems "without proper vulnerability detection" have chased customers already in crisis and missed obvious distress signals. That's a design and oversight question, not a technology one. Under the UK's Consumer Duty regime, a firm still owns the outcome even if an AI system produced it — so procurement should treat vulnerability handling as a compliance control, not a feature checkbox.
Frequently asked questions
Is this specific to UK financial services firms?
The regulatory hook — FCA Consumer Duty and the FCA's 2026 AI review — is UK-specific, but the disclosure psychology behind it applies to any support operation handling debt, health or hardship conversations.
Should we stop routing hardship calls to human agents?
No — the piece itself argues for weighing emotional stakes per call, not replacing humans outright, and flags that poorly designed AI can worsen outcomes for customers already in crisis.
How does this connect to what this site has already flagged?
It's another example of the pattern covered in "When Every Vendor Sounds Excellent, Polish Stops Being Evidence" — a plausible-sounding stat used to sell a routing decision, not to test one.
Source: Call Centre Helper (Sabio guest blog, "Why People Tell AI About Their Debt – But Not You").
Sabio explores how contact centres can use AI and human interaction to better support vulnerable customers. There’s an assumption baked into contact centre design that rarely gets examined: when a conversation is sensitive, the instinct is to route it to…