How Canon's AI committee accelerates GenAI adoption

As generative AI (GenAI) reshapes business operations, successful adoption requires more than just technology -- it demands a strategic, cross-functional approach.

Digital imaging company Canon created a multidisciplinary AI committee to accelerate GenAI adoption, according to Michael Lebron, head of digital applications for the company's Americas division. Key to the company's success is a dual horizontal and vertical strategy -- democratizing AI knowledge across the workforce through education and tailoring use cases to specific verticals, such as customer service, sales and finance. Lebron's experience also highlights the importance of starting with education before diving into technology deployment -- a vital lesson for CIOs aiming to drive enterprise-wide GenAI adoption.

In the following interview, Lebron shares insights into how Canon's AI committee fosters collaboration across teams, manages data privacy and compliance, and overcomes challenges to operationalize GenAI at scale.

Editor's note: The following transcript was edited for length and clarity.

What is Canon's cross-functional AI committee?

Michael Lebron: It's a multiorganizational committee made up of business units, including legal, finance, marketing and service -- so we have representation from all the business units within Canon Americas. It generates a two-way street between the technologists within corporate IT and the business units.  

The committee focuses on mobilizing the organization and democratizing the understanding of GenAI. We discuss compliance, the art of the possible and even share ideas at an individual level. We meet monthly and have subcommittees that focus on GenAI subtopics.

What were the main problems that the committee was meant to solve?

Lebron: The primary goal was to accelerate our pace of GenAI adoption. The committee has helped us create channels of communication and roll out the technology vertically and horizontally.

Horizontally, we're training our general workforce on the risks of GenAI and empowering them to use it effectively. However, we also break our strategy down into the verticals themselves. For example, customer service and sales have different needs. Therefore, the technologies and the approaches that we seek to implement are also very different.

On the vertical side, we're entering high-risk areas, such as information security. We're getting a deeper understanding of how cybercriminals can use GenAI to potentially manipulate and penetrate our defenses.

What GenAI use cases have been most valuable internally?

Lebron: Most people focus on the technology itself, but our learning and development efforts have had the biggest impact. Education can help overcome individuals' fears of using AI.

We began with basic GenAI education, including its applications and potential risks. However, we've extended the education to include prompt engineering. That's the horizontal strategy -- democratizing the knowledge and use of GenAI and promoting its use at scale. This includes the typical chat interfaces for day-to-day private productivity, such as Microsoft Copilot or Google Workspace with Gemini.

On the vertical side, we see the most value coming from customer-facing apps in the service and support areas. For example, we are beginning to deploy chat interfaces to enhance call deflection and improve customer satisfaction.

Did you design any frameworks to manage data privacy compliance while still allowing innovation to thrive?

Lebron: We have developed a policy that balances the safety of our enterprise with the encouragement of exploration. We allow all our employees to use GenAI tools, provided they've gone through a compliance and information security approval process. A separate committee reviews all new GenAI technology.

With that said, we didn't want to stifle the use of what we considered publicly available GenAI tools, such as ChatGPT and Claude. We have a provision within our policy that allows for the use of that. However, it requires management approval in addition to taking certain learning and development classes. Education is at the center of our strategy.

How do you structure teams to best align GenAI efforts with business strategy?

Lebron: That gets back to the vertical and horizontal approaches. Vertically, we take it by business function, including commerce, sales automation and finance. Each one of those will have different projects and programs.

We also have horizontal applications. For example, we are looking at the quality of our data and using GenAI to identify inconsistencies. We're also looking at using agents to help us continually improve the quality of our data. These are horizontal applications because they go across the organization, regardless of the data itself.

What steps have you taken to upskill staff?

Lebron: We have an extensive learning and development program. First, we begin with an intranet site where we post surveys to gather information from end users. We also use it to publish information about our workplace GenAI policy -- that's an ongoing campaign.

Secondly, we offer regular training to end users, typically in the form of online training. For example, it could be general training on prompt engineering. However, we also offer tailored training for people in verticals, such as finance or support.

Lastly, the biggest success is what we call AI Unplugged. Every two weeks, one of our AI experts hosts a one-hour session. We have between 400 and 500 people attend every two weeks. Sometimes, we bring in outside speakers, and at other times, we just discuss use cases.

We take feedback from our intranet surveys, and those responses can become topics for AI Unplugged. It becomes almost like a show. I've attended it numerous times, and sometimes I've spoken. Sometimes I'm just in the background because we have a chat running, and I'm answering questions. Sometimes we'll bring in a third party, and at other times, it's a fireside chat.

What are the biggest challenges you've faced when operationalizing GenAI at scale?

Lebron: The biggest challenge is lack of dedication. I run all our customer-facing systems for the Americas, and GenAI is not as much of a focus as I would like it to be.

Secondly, our data is a challenge. For example, we were launching GenAI chat functions on our commerce site for sales, and it was coming back with the craziest responses. We went around in circles for probably four weeks, thinking maybe we need to tweak the large language model (LLM). However, when we really started to dig deep, we realized it was the quality of our data.

We weren't ready and ended up pulling the function because it recommended a Sony product on the Canon website. We have a product that interfaces with Sony, and the AI hallucinated due to the way our data was structured. So, the structure of the data is equally important as the data itself.

In response, we initiated a data quality project that identified a multitude of issues with our data, which were the root cause of the hallucinations. We're currently in the midst of cleaning up that data.

What's the biggest lesson that you've learned about moving from experimentation to enterprise-wide adoption?

Lebron: The biggest lesson is getting back to focus. An organization can't just say that GenAI adoption is important -- they need to provide dedicated resources if they want to scale it at speed. Scale at speed requires focus, and you really can't scale at speed when it is a secondary or tertiary priority.

Another key insight is that horizontal, company-wide education should come before vertical deployments. We should have started by building a fundamental understanding of what GenAI is and how it can be applied across the organization. Addressing this knowledge gap through education is the critical first step before looking at specific use cases or technologies.

Instead, what happened is we started by chasing specific technologies and LLMs to deploy, focusing too much on the tools themselves. That was a mistake. We took a vertical approach first -- asking, 'Where can I adopt this?' -- instead of focusing horizontally on broad education.

Tim Murphy is site editor for Informa TechTarget's IT Strategy group.