Pick one task, hasten slowly: How to avoid AI decision paralysis

Almost every day a new report or piece of analysis sets out the productivity and efficiency gains purportedly available to financial advice practices through the adoption of artificial intelligence.

For a practice, just starting out on the task of analysing where AI might make useful differences to its business and to its clients – let alone which is the best system to adopt to achieve them – is daunting, and can lead to decision paralysis.

Despite the hype, and even in the face of a seemingly endless stream of case studies, the jury remains very much out on whether AI yet brings the enterprise-wide benefits to advice practices that its champions claim.

One such report, report released last week by Iress and Signal Advisory, AI in advice: adoption, beliefs and barriers, says that practices with AI embedded across their advice processes are more likely than those still piloting or partially deploying it to report that AI is helping advisers see more clients, serve clients better and retire existing software or services. 

They sound like compelling benefits, but the report adds that “the finding does not establish a causal relationship, but it indicates that the depth of deployment matters”. 

“Using AI in isolated tasks may create pockets of efficiency, while embedding it across the workflow creates more opportunity for those gains to flow through to business outcomes,” it says.

“The next phase of adoption will be about helping practices move towards embedded use, supported by integration, measurement and appropriate human oversight.”

So, it’s understandable why there’s growing pressure to jump on board the AI train, but it’s equally understandable why some practice principals remain reluctant to commit to a technology whose benefits are unproven, and around which there remain significant implementation risks.

But it ultimately sets out three useful adoption and implementation guidelines:

• Apply an AI solution to where the work is, and where the technology can add value.
• Understanding what the AI has delivered requires a business to establish a baseline against which changes can be measured – positive and negative
• Embed governance – data control, human reviews and accountability – inside the workflow, don’t try to impose it from outside.

The Iress/Signal report says among practices yet to start, 34 per cent are unsure where to begin or which tools to trust. A practice principal must decide which task AI will improve, which of dozens of products will do that job best, how it connects to software already in use in the business, and what to do – including identifying who is responsible – if something goes wrong.

The report is based on a survey of 839 advice professionals in Australia and New Zealand, and a key finding is that complexity is generally the first issue an advice practice runs into.

Poor integration

Poor integration with existing systems is the largest barrier to wider adoption, cited by 50 per cent of respondents. Marloo co-founder Hardy Michel says connecting an AI assistant to a dozen systems does not repair fragmented data. Claude for Financial Advisors, for example, “doesn’t close any of the tabs, it just changes who’s doing the tabbing”.

Cost also is a barrier, albeit a slightly murkier one. While only 15 per cent of respondents named cost, specifically, as a barrier, that may say more about how little is actually being measured: just 28 per cent have measured the time or cost impact of the AI tools they already use, 63 per cent say AI was added without replacing anything, and only 5 per cent have retired any software. 

And the pace of change is contributing to adoption hesitancy. The report says 47 per cent of respondents are reluctant to commit to any one tool. But doing nothing brings costs of a different kind, and waiting today can make the task of catching up more troublesome tomorrow.

It’s also critical that when an AI tool is adopted, its use is properly documented and staff are adequately trained on how to use it, and how to respond if it does something unexpected. The Iress/Signal report shows that 22 per cent of respondents “Cannot point to any policy, or do not know whether one exists”. 

A practice that has no, or inadequate, policies in place faces the risk that untrained staff using powerful tools may inadvertently expose sensitive client data to bad actors.

Make a smaller decision

Before implementing an AI tool, experts recommend users first identify the task or tasks it will be applied to, rather than buying the tool and then looking for applications – buying a hammer, as it were, and then searching for nails, when the job might be better done with a screw in the first place.

Licensees should first answer what problem is being solved, what happens when the AI is wrong, who owns the outcome and what evidence would justify wider use. Assured Support managing director Sean Graham says to start where the benefit is meaningful, the cost of error is limited, and outputs can be independently checked. 

Rushing in is not necessarily safer. ASIC Report 798, Beware the gap: Governance arrangements in the face of AI innovation, said some licensees were adopting AI faster than their risk and governance arrangements were being updated.

 Of 14 licensees in the ASIC report that were planning to increase AI use, only one had improved governance first. Two others had governance that lagged use, and one credit licensee kept running a poorly understood model but planned to expand its use of AI for fear of being “left behind”.

The Iress/Signal research was commissioned by Iress and 97 per cent of the survey respondents run Xplan as their primary planning software. The report’s finding that 43 per cent would rather wait for AI inside existing software plays into Iress’s hands as a powerful incumbent – but it also means that 57 per cent would prefer not to wait.

In addition, respondents were generally at the smaller end of the scale, with 63 per cent in businesses of five or fewer advisers, where the headspace required to consider the permutations and combinations of AI technology and its applications across a business may be at a significant premium.

The hype around AI will only escalate, and the pressure on businesses of all sizes to adopt it, or be seen to be falling behind, will only intensify. And while fear of missing out on the benefits of AI is a real motivating factor, the consequences of poor implementation may be equally far-reaching. Hasten slowly.