Adviser trusting AI

target readers-cv

By Gemma Livermore

Financial institutions are pouring money into AI, but its value depends on accurate data, clear ownership and knowing when a human should take over.  

There’s no lack of AI ambition within financial institutions. Trust, however, is harder to come by. Across banking, insurance and wealth management, the same problem keeps creeping in. Advisers double-check AI-generated answers before they’ll put them in front of a client, which quietly cancels out the time AI was meant to save.  

The fix is simple but useful. Firms need their AI drawing on accurate, approved information, and they need someone answerable for what it produces. 

Why don’t people trust AI’s answers yet? 

Many advisers have a diplomatic way of describing AI-generated material. It’s helpful, they say, until it’s wrong, and confidently so.  

Senior buyers tend to feel the same way. Recent research among 200 C-suite and senior leaders found that, as buyers, 75% trust human guidance over AI-generated content when the two don’t match. Nearly half (46%) say AI has made them more reliant on people for high-stakes decisions. The most common reason they give? Accountability.  

Financial services is built on it. A client who moves their pension wants someone who’ll stand behind the recommendation. You can’t take a chatbot out for coffee and ask it tough questions about the small print. 

AI still has a real role in client-facing work, but clients need to be able to trust its workings, trust in whatever’s underneath. And right now, what’s underneath can be confusing. 

What happens when AI draws on the wrong information? 

AI carries the quality of its sources into everything it produces. A tool working from an outdated rate table will present it with the same polish as approved copy.  

The same research found 42% of organisations run into at least one AI content failure every month, whether that’s outdated information, inconsistent messaging or unapproved material. Forty-one percent say unapproved AI-generated content goes to customers a few times a year or more. 

That’s embarrassing. And in financial services, it’s a conduct issue. A wrong fee or an unapproved claim can turn into a complaint or, worse, a conversation with the regulator. 

It’s also where the expected efficiency disappears. Among leaders who’ve had problems with AI outputs, 57% say teams now spend time correcting errors, 51% have added review or approval steps, and 45% report lower employee confidence in AI tools. An adviser who rechecks every AI-drafted suitability letter is doing the job twice. 

Why is fragmented data such a problem for banks and insurers? 

The temptation is to blame the model or the prompt. One financial services COO quoted in the research put the cause elsewhere: “Maybe 90% of the time it’s a data-related issue, not a functionality or prompting problem.” 

The leading breakdowns are structural. Almost half of leaders (46%) say different teams use different systems, 43% have no single source of truth and 42% say content becomes outdated faster than anybody updates it. 

Financial institutions feel this more than most. Decades of mergers and product launches leave behind a vast sprawl of disparate systems and document stores that don’t agree. A head of technology at another financial services firm noted how hard it is to fix “when you’re not a startup starting from a blank slate.”  

AI tends to smooth over that disagreement. A relationship manager gets an answer, with no hint that three systems held three versions of the truth. It’s little wonder only 1% of leaders feel confident their AI-enabled go-to-market operations meet an enterprise standard across every readiness area tested. Governance of AI outputs is the weakest area of all. 

 Source: The New Go-to-Market Standard for B2B Selling, Seismic and NewtonX, 2026 

Who should be accountable when AI gets it wrong?  

People have to decide which source wins when two systems disagree. But ownership is scattered. IT is the function most often named as responsible for the accuracy of customer-facing AI output, at just 29%, while 16% of organisations say nobody owns it at all. 

That’s a tricky position for an industry that’s spent the past decade giving senior managers specific responsibilities. Most banks have a clear answer for who signs off a new product, but ask who signs off what their AI says about it and the room often quiets.  

Accountability needs a name and the authority to approve, correct or stop an output before it reaches an adviser. It also needs a human checkpoint before a client sees anything. Some 89% of leaders agree clear content ownership and connected systems help organisations reach purchase decisions faster. 

Where should humans stay in the loop? 

Getting the foundations right is what lets firms decide, deliberately, where people add the most value. 

The research suggests the answer depends on the task. Buyers still lean on a human expert to negotiate pricing and terms (55%), to judge whether the business case fits (49%) and to understand risk (45%). One CFO in the study said they’d trust AI over a person on pricing, but would still want someone with 30 years in the industry for a specialised technical question.  

Financial services maps neatly onto that split. AI can assemble a portfolio review or summarise a client’s history in minutes. Explaining why a recommendation fits this client, in this market, with this appetite for risk, is still a human job. So is owning the consequences. 

Key takeaways 

  • AI is only as trustworthy as the information it draws on, and most errors start with the data rather than the model itself. 
  • Every firm needs a documented rule for which source wins when systems disagree. 
  • Accuracy needs an owner and a human approval step before content reaches a client. 
  • People still want a human for judgement and accountability, and that’s where AI should free up their time. 

Conclusion  

For all the money going into AI, the deciding factor in financial services may turn out to be the quality of the data underneath it. Accurate, approved information, a named owner for what AI produces and a clear sense of where people add value will do more for adoption than the next model. The real test comes when an adviser reads an AI-drafted client pack and sends it without double-checking every figure. That moment is closer than it looks, provided firms do the boring work first. 

About the Author

GemmaGemma Livermore is Head of Financial Services Marketing for EMEA & APAC at Seismic and founder of the award-winning Women of FinTech community. With more than 20 years’ experience in financial technology, she is a recognised advocate for trusted AI, diversity and innovation across banking financial services, helping marketers and sales teams leverage the right content, insights and technology to sell more effectively.   

LEAVE A REPLY

Please enter your comment!
Please enter your name here