Artificial advantage

Few consumer-facing industries illustrate the potential and pitfalls of artificial intelligence more than the financial services sector.

Generative AI — the technology powering chatbots that can seemingly think and provide reasonably correct, if not insightful, answers to our questions — is arguably a wonder of the modern world.

The potential for providing financial advice is indeed palpable. Then again, it’s not hard to imagine the technology running amok especially with people’s money.

A recent report on digital banking services provides insight on the risks and opportunities.

JD Power surveyed digital banking tools, ranking our nation’s financial institutions. It found companies offering AI-powered tools generally scored highest.

“(AI) can really lift consumer satisfaction,” says Jennifer White, managing director of financial services Intelligence at JD Power.

She notes well-executed, AI-powered chatbot can lift satisfaction scores by about 16 percentage points.

Yet the technology cuts both ways. When AI falls short of expectations, it sours consumers’ experience.

It’s a tightrope to walk for the industry, which can see the potential and grasp how without proper precautions these robot brains may careen off the rails.

White notes AI is already accomplished at performing basic banking tasks like retrieving the current balance on a credit card or telling you how much you spent at Tim Hortons over the last month.

“When it comes to doing very transaction-oriented tasks, AI can be very helpful.”

The study, however, found AI tools struggle with more complex tasks.

“The virtual assistants struggle to deliver,” White says. “That typically makes somebody have to walk into a branch or call a contact centre.”

So for the time being, the AI experience is mixed, she adds.

Yet the potential is there to do much more. A major obstacle to AI doing more is ensuring it doesn’t hallucinate, essentially providing answers that sound believable but are really wrong.

Ensuring that doesn’t happen is no small feat, says Rizwan Khalfan, executive vice-president and vice chair of strategic growth, innovation and partnerships at TD. “Our fundamental relationship with our client is based on trust, and the currency of trust in an AI world is exponentially higher.”

TD’s mobile banking platform, by the way, ranked highest among financial institutions in the survey, with a score of 690 out of 1,000 possible points. (TD isn’t crushing competition; the average score among banks is 678 points.)

Khalfan notes all financial institutions strive to continually up the ante for what AI can do to improve the customer experience.

“It has to be simple; it has to be increasingly intuitive; it has to be more helpful in what clients are trying to achieve,” he says.

“If you keep those as fundamental design principles, then an-AI-driven app will perform well.”

Still, many bank-offered AI tools today are pedestrian compared with Google’s Gemini, Anthropic’s Claude or OpenAI’s ChatGPT.

There’s a reason, Khalfan says. Banks mostly use earlier versions of AI, based on a “deterministic language model.”

With this older version of AI, if the model doesn’t understand the question beyond its scope, it will say so and decline to provide an answer.

In contrast, generative AI models will attempt to provide an answer — right or wrong — that sound convincingly true, but may be far from factual. One recent Stanford University study shows even the top generative AI models’ accuracy ranges from 20 to 94 per cent.

“We know that AI models can come across very confident in their answers, even when they’re wrong,” Khalfan says.

He adds banks must ensure the technology does one of two things before offering a new generative AI-powered tool.

One, it must provide the right answer nearly 100 per cent of the time. And two, if it cannot provide the correct answer, it will then connect customers with a human who does.

Despite the challenges, there’s good reason to be bullish on AI’s upside in helping us manage our finances, says chartered professional accountant and TEDx speaker Nelson Soh.

“AI is a powerful tool because it can help consumers save time, gain clarity and set them up for success to achieve their financial goals,” says the managing partner at FSQ Consulting in Vancouver.

“It can help you create a budget, monitor your cash flow and give you alerts when it suspects fraudulent activity.”

But relying solely on generative AI to understand investment, tax and other complex financial decisions “is still not advisable,” even if it may provide seemingly solid solutions, he adds.

“AI cannot fully account for your personal circumstances, including your risk tolerance, tax situation, family dynamics or long-term goals.”

At least, it can’t yet.

In the near future, we are likely to see autonomous financial managers that can. These would involve the latest trend in the technology called agentic AI. This essentially involves AI making decisions without human intervention. Agentic AI tools could, for example, move “excess cash from chequing into high-interest savings accounts,” Soh says.

It could also detect fraud, alerting people in real-time, before sending funds to scammers.

Khalfan says the industry — including TD — has been working on these tools for years. And very soon, we will see the next generation of chatbots able to do more. That might mean less waiting in line at the bank or on the phone for assistance.

“There’s this concept of ‘private banking’ that is available to only wealthy Canadians today,” he says. AI-powered tools may provide that concierge service to all clients, regardless of wealth, to help them improve their finances.

“We can uplift financial wellness across the board — that’s the potential of AI in banking.”

But the stakes are high in getting it right, Khalfan says.

“We must have the right guardrails to ensure the answers you’re getting we can stand behind.”

Otherwise, banks run the risk of AI eroding the most valuable currency they have: our trust.

Joel Schlesinger is a Winnipeg-based freelance journalist

joelschles@gmail.com