TORONTO — Canadian businesses and governments have been slow to take up artificial intelligence. But Deloitte sees a huge opportunity to help the country’s major enterprises deploy the fast-evolving tools, the CEO of the consulting giant’s Canadian branch says. 

Deloitte Canada had 14,481 staff and $3.97 billion in revenue last year, from selling consulting, audit, tax and other professional services. Its largest line of business involves modernizing and running clients’ technology infrastructure. Generative AI will be “core” to a lot of that “transformation” work, CEO Anthony Viel said.

The firm is betting that the technology will become an everyday part of its customers’ and its own operations—an opportunity to save money and increase revenue. 

With boards and executives pushing companies to try generative AI, Deloitte Canada worked with customers on a number of pilot projects last year. But fewer have employed those tools in their everyday business, Viel acknowledged. In part, he blames Canada’s larger challenges with adopting new technologies. 

In an extended interview with The Logic earlier this month, Viel discussed how Deloitte Canada is bringing AI—both generative and traditional—to its clients and its own workforce, and why adoption hasn’t yet matched the hype.

This transcript has been edited for clarity and brevity. 

Deloitte Canada had an AI practice before a lot of the other big professional services firms. What was it doing before ChatGPT?

Dare I say it, it was doing traditional AI. Deloitte really started to [apply AI] in the early 2000s. The processing power and data availability was not as good as it is today. 

We were looking at things like customer segmentation and using AI [for] anomaly detection for financial crime, anti-money laundering and fraud. 

How much of that work are you still doing in 2024?

It’s a majority of the work, because it’s able to generate immediate outcomes on critical priorities around expenses and operational efficiency [or] productivity. There’s a straighter line between the work and the impact on the P&L [profit and loss]. 

Generative AI is still at that stage of: “Looks great, has a ton of opportunity. I want to talk about some proofs of concept or use cases.” But there’s not as many of those as, “We’ve got an adjudication model in financial services or claims model in insurance that’s been in operation for 10 years. I think it can save you X or Y and we’ve done it 15 times before.” 

Any company with engineers today can access a large language model and build on top of it. Why is anyone coming to you at all?

Access to talent, and to talent that’s done it before. It’s not [a question of] what you can legally do with generative AI, but what should you do with AI? We’re a solid brand to say, “That looks onside. That doesn’t look onside.” We’re the largest of our kind. What we haven’t done here in Canada, we can plug in [from] any of our experiences from around the world.

I’m hearing from a lot of enterprises and consulting firms that last year was really one of experimentation and proofs of concept. Was that also your experience? 

A lot of trying it out. But we augmented that with pushing our folks through literacy and adoption courses. We even set up our own version of ChatGPT that was appropriate to some of the protections and guardrails we wanted to put in place. We set up an AI institute, which shares principles of trustworthy and responsible AI and use of data. 

Your internal version of ChatGPT—

GenD, we call it. 

What are your employees actually using it for?

Over 50 per cent of our people are using it, day in and day out. We never use it with sensitive information, but for all the other work that’s on the periphery. We’ve got a [feature] that enables us to synthesize the information that we have, and then construct it into summarizations for working groups and the like. What used to be “listen, write, type,” is now condensed—a machine’s doing all of that—and presented in an output that’s consistent irrespective of whether you do it or I do it.

On the client side, were you doing a lot of pilots and proofs of concept using generative AI last year?

Yeah. Typically you’d say, “You can apply it in all these different areas.” Then you go through a prioritization and say, “Here’s the two or three things that can actually make a difference. Let’s experiment with those.”

One thing I can say to you, hand on heart: The math always works. So then our conversations are, “When the math works, what else has to be true around the processes, the people, to actually support that?” The experiment is: Is this gonna hit the P&L? Or save the balance sheet? Or enhance cash-flow? Or keep me out of jail? 

Is the answer to any of the questions you just asked “Yes”? Because the other thing I’m hearing is that the conversion rate from proofs of concept to full-scale use is fairly low.

We might have an issue, culturally, at the country level. We’re number one in talent, and nowhere from an adoption perspective.

You’re right. But it’s hard to answer the question as to why. The adoption could be not happening because the experiment is not delivering on the P&L, balance sheet, cash-flow statement, or regulatory context. If it doesn’t translate to that, you’re not going to adopt it and scale. You [may have] picked the wrong thing to start with. 

Or you’ve got an organization that is limited because of its willingness to change processes. Or its technology infrastructure is not modernized to take the speed or processing power required for the generative AI solution. Or there’s not a need necessarily in the market, because you’re already ahead of your competition.

How much of what people have tried was generative AI for generative AI’s sake, because of a board saying, “I keep reading about ChatGPT. What are you doing with it?”

There’s definitely a push from boards and other stakeholders that you’ve got to be doing something with this, [but] not necessarily direction on where. 

Around the world, 85 per cent of all of these investments-slash-experiments have been in the area of efficiency. Efficiency means you need fewer people to do the work that’s currently being done. 

What’s alarming is there’s only 15 per cent of organizations out there that are investing in growth. We need growth in a country like Canada as much as we need efficiency and productivity improvement. Generative AI should be allowing you to address more needs of your customers, and different needs that don’t exist today. 

Are you expecting that working on generative AI with clients will be a significant revenue driver for the firm in 2024 or 2025?

It’s going to hit all of our revenue. We have dedicated investments to particular industries and issues within industries for AI applications, some of which are generative AI. That will be core to a lot of the transformational work. We take a function of our client, modernize their technology, and run that for them. That’d be half of about $3.5-billion worth of work here in Canada, or $33 billion globally. It’s a big part of our business. 

There are no transformations today that don’t have AI in the middle of them.

How easy or difficult are you finding it to attract talent with AI skills, given how much demand there is right now?

Very, very difficult. 

Our talent tells us that meaning in [their] work is first and foremost. What’s attractive [about] Deloitte is you’re going to work on things that make a difference. Playing a role in a better climate outcome means more to an AI practitioner than, “Come and modernize the hell out of something less purposeful.” You get variety as well. You might be saving the planet one day and improving health care the next. 

And then being part of a global network. You can tell from my accent, I’m not from around here. You’ve got that mobility to work for Deloitte in other jurisdictions.

When it comes to pay, we try to be competitive. But the world is coming here to get our talent. 

At last year’s All In conference in Montreal, you predicted that Canada was headed into a “trough of disillusionment” [one of the stages of Gartner’s technology hype cycle] with generative AI. Are we there?

I think we’re still there. I won’t say we’ve come out until we see a bigger uptick in adoption. Why we’re disillusioned is because all of this technology hasn’t been applied to prove the efficiency and growth opportunities that come with a tech-enabled future.

I’d want to see the types of investments that you see from the technology providers building large language models—billions of dollars. It doesn’t have to be at that scale here locally. But are we courageous enough to make that investment for the transformation that we all know is going to happen? Once we do that, I think we’re on the other side of that trough.