AI for every business

As seen on BNN Bloomberg
 

It’s only been a year and a half since ChatGPT was introduced to the world, but it’s already impacting the way many Canadian companies do business.

Whether it’s automating time-consuming manual tasks, improving interactions with customers, streamlining data analysis, or helping marketers create images and content, generative AI is being used by many companies today. And we’re only getting started.

“Generative AI is starting to reshape business operations in Canada,” says François Gaudreau, a partner in KPMG’s Management Consulting and Intelligent Automation practice based in Montreal. “It’s only going to accelerate, so it’s essential for organizations to understand AI and how it impacts them and their workforce.”

While it’s still early days for generative AI, companies are prioritizing the technology. A KPMG Canada survey of Canadians conducted last November found that 22% of respondents said they are increasingly using it to help complete tasks at work, with 61% of users reporting they use generative AI multiple times a week, up from 52% in May.

KPMG also found that generative AI adoption in the workplace is growing at an annualized rate of 32%. At that pace, the firm estimates that half of all Canadian workers could be using it within three years. “These tools allow companies to turbocharge the talent inside their organizations,” says Marc Low, lead of KPMG Canada’s Ignition Vancouver, the firm’s innovation and advanced technology group.

While there are still many uncertainties and risks around AI, the technology is only going to get more powerful and useful for companies. Now’s the time for businesses to think about how best to invest in AI to get the most value out of the tech and find ways to scale.

From Low’s perspective, we’re at a moment for education, training, and experimentation. “Companies must understand what the AI models can and cannot do, how to identify and manage the risks, and measure whether it’s creating value,” he says. “We help our clients through that journey.”

New AI-powered tools seem to be multiplying as fast as the AI adoption rate, with the technology appearing in business software platforms offered by large, established players, as well as an explosion of smaller, niche providers for specialized functions. “The marketplace is very loud and evolving quickly,” notes Low.

Executives may be tempted to wait until the market settles on some winners, but Low and Gaudreau advise against that approach. Instead, they advocate for developing internal proof of concept trials with whichever of the leading Large Language Models (LLMs) is most compatible with an organization’s objectives and IT environment.

“There’s a lot of value to be proven with one of the two or three core generative AI platforms,” says Low. “As you get more mature in the process, you’ll get into more narrow, specialized tools, but that conversation is down the road for most companies.”

One of the first places to start is to appreciate the shift AI represents. “It changes the way you interact with technology,” says Low. “Up until now, software engineers would create a function-specific product and it was obvious what the application is built for.”

Generative AI tools work differently in that you have to articulate the output you need from software. “We’re not used to thinking about technology in that way,” he adds. “You have a new paradigm in terms of what software can do for you, which requires a mindset shift that can be challenging to adopt.”

Company leaders need to think about one or two potential pilots, prioritized by impact and effort from within existing processes. When thinking of where to deploy AI, consider: where you might find efficiencies; if there are ways to drive revenue with the technology; if there’s an area where you’re strategically or competitively at risk; and whether you have the right resources to successfully implement generative AI, says Gaudreau.

As promising as generative AI may be, if your staff doesn’t know how to use it properly, then you’re not going to get very far. KPMG found that more than 90% of employees are eager to get trained on these tools, including how to write prompts, and verify responses aren’t misleading or biased. That training is critical, given the same study found that 56% of users fail to verify AI’s output.

When it comes to assessing return on investment, value tends to get measured in time savings, says Low. He typically sees a 20% to 25% improvement in a variety of tasks, such as quickly producing the first draft of a document, data gathering, or doing financial analysis more efficiently.

This is evident in call centres, one of 600 potential use cases KPMG has identified for generative AI. The technology reduces time per call by transcribing a conversation in progress, identifying questions being asked, and then providing answers in near real-time to the agent from a prescribed information repository, all in context with the customer’s account information. Gaudreau conservatively estimates that by resolving issues in an initial call and reducing the average call length, about a third of a call centre’s workforce could work on more strategic, value-added tasks.

Of course, there is a cost to using AI models. If you’re paying to use ChatGPT, for instance, you’ll be billed depending on how much text your prompt generates. Those costs may be in the pennies, but fees could add up depending on how you use the service. For those hosting AI models internally, cloud computing costs and infrastructure maintenance must be factored in. “They’re not deal breakers,” says Low, “but it’s part of the ROI discussion.”

Once you get familiar with generative AI and can demonstrate its value, you’ll want to scale up your usage. KPMG has developed an AI lifecycle management process to help companies evaluate and monitor business value from the initial concept to decommissioning. It includes five steps:

1. Intake
   During this process, companies must consider the use case, whether it’s aligned to a business or strategic priority, the key stakeholders and sponsors, the expected benefits and any potential risks.
2. Prioritization
   Here, you’ll want to think about the data and metrics you can collect to understand the value of the AI program to the business.
3. Optimization
   During this phase, observe whether you achieved your expected business outcomes based on the development of your proof of concept. Think about what a path to scaling might look like.
4. Development
   When you’re ready to develop your AI program in a bigger way, think about how the technology might be integrated into the overall enterprise and if you might need any additional infrastructure to support greater adoption.
5. Scale an monitor
   Once the AI solution is rolled out, see if you are generating the ROI you had expected. Look carefully at whether users are interacting with the tool as you had expected.

If you’re still skeptical about the benefits of generative AI, consider your “competitive moat,” says Low. If you’re susceptible to disruption, you’ll want to act. “The urgency will vary dramatically depending on the competitive dynamics in your company’s market,” he notes.

While the way in which businesses implement generative AI may be different, every company must act now. In a country like Canada, which lags other nations in per capita GDP output, generative AI may give companies a chance to make big gains. “The opportunity for small to medium-sized enterprises is enormous,” says Low. “Generative AI is an opportunity for companies to accelerate in the marketplace and increase their competitive advantage.”