I recently spoke with Johnson & Johnson Executive Vice President and CIO Jim Swanson about the ways his company is deploying AI tools to help boost productivity and enhance core functions. Jim also spoke about the actions his team is taking to help cascade AI know-how across the organization and embed AI tools directly into staff workflows.
My conversation with Jim was the fourth installment in a series of interviews I’ve been conducting with innovative CIOs at some of the world’s top companies.
Asheem Chandna: How are you using AI to drive improvement at J&J?
Jim Swanson: Today, technology is fundamental to how we deliver breakthroughs in our two sectors – Innovative Medicine and MedTech. AI tools are an important part of that technology solution set.
We have three main layers of AI deployment. The first layer is about improving productivity around our enterprise capabilities. The second layer is centered around enabling our end-to-end processes, whether it be drug discovery, clinical operations, getting a product to market, or engaging with customers. The third layer is about working to embed AI into the core of our products and services.
Can you provide specific examples of how J&J is deploying AI tools to improve productivity?
One of the ways we’re deploying AI is to help us to assess supply risks by product and detect supply risks before they can cause disruptions. Let’s say there’s a major fire that’s creating issues with material suppliers. We need to know if that event will have any impact on our supplier manufacturing locations, materials, or products, so we can put contingency plans in place. We’ve developed an AI-enabled platform that helps us predictively sense, monitor, and mitigate these types of risks. It gives us insights to quickly make decisions, leading to more resilient operations.
Another example is in our Global Services function. We now have AI-powered, automated agents that can answer employees’ questions more directly. In the system that powers it, we’ve ingested all the information employees might ask about – from benefits and time off to switching managers. When an employee calls or chats with our service desk, they get immediate responses to their questions and are sent links to relevant documents referenced in the call.
Going back to the second layer you’d mentioned – how is J&J using AI to improve ‘end-to-end capabilities?
We’ve used AI tools for expediting and enhancing clinical trials, which are important for demonstrating safety and efficacy to regulatory and health authorities. If we are using sites that are slow to enroll patients, it takes much longer for us to get to the insights we need. So, we have started to use AI tools to improve our enrollment times. We’ve seen an up to 2.6x increase in enrollment at the better performing, higher enrolling sites recommended through our AI models, which in turn is accelerating our time to insight.
As we operate all around the world and treat patients of every background, our clinical trial populations must also adequately reflect this diversity. But sometimes, we face difficulties in ensuring our trial pools are sufficiently sourced from the patient populations affected by the diseases we’re aiming to treat. We are now applying AI models that allow us to better identify sites or patient segments to improve representation.
Through our biosignature platform, we are using AI-powered image analytics to better determine if a particular compound might have the potential to be a safe and effective medicine. AI allows us to make these determinations much faster than if we were to just wait for biological results. Our ability to use this treasure trove of information has helped us deliver novel molecular entities – or potential new drug candidates – faster than we would have been able to before embedding AI into our drug discovery processes.
We’ve also taken all our curated, medically validated, legally reviewed content around our products and created an asset called AskJIA, which we are deploying to train our sales reps to improve their understanding of our products. Further, we’ve consolidated all our provider interactions into a tool called Engagement.AI, which we are using to tailor insights to help inform healthcare providers as they consider treatment strategies for their patients in over 14 countries.
Can you tell us more about how J&J is embedding AI into its core of products and services?
We are focused on use cases that matter and that allow us to continue to build our set of capabilities over time.
For example, with the issue of arrhythmia (irregular heartbeat), we are using AI to help electrophysiologists determine which part of the heart should be targeted for ablation therapy, which is a procedure that can improve patient outcomes. In orthopedics, we’re using AI to automate the analysis of patients’ anatomy to help treat bony deformities of the foot and ankle. The software outputs a recommendation for the right instrument needed for the surgery and suggests a plan for correcting the deformity, simplifying the process for the surgeon.
On a practical level, how are you making sure AI tools are adopted and integrated throughout the organization?
First, we have created two councils, an AI Council and a Data Management Council, which allow us to properly mature the technology stack and apply these tools to use cases that matter. These bodies help ensure we adhere to ethical standards and have a clear understanding of how to use these technologies in ways that are scalable.
Second, one of my core strategic pillars is to build the digital acumen of the entire company – in more than 130,000 employees. My basic thesis is that to fully benefit from these AI offerings, our people need to understand what they are, how they work and how they can be used to our advantage.
We’ve created dedicated training curricula and coursework which provide employees an opportunity to experiment with AI tools. Across J&J, we’ve seen more than 30,000 completions of courses in our “digital boot camp” training, and 47,000 employees have taken a Generative AI course required anyone looking to use these tools.
Third, we’re placing a heavy emphasis on embedding AI tools directly into employees’ workflows. For our sales representative, these capabilities are integrated directly into their CRM platform. When a representative prepares for a visit with a provider, they can get information on the kinds of patients the provider usually sees, the interactions we’ve had with them previously, the types of advances or insights they might want to know about, the channel the provider prefers to be engaged on, etc. Similarly, with our drug safety teams, we have embedded these tools into their workflows so they can better identify and resolve cases.
To what extent are Gen AI tools cascaded across J&J?
Over the past two years, much of our Gen AI work was in proofs of concept and pilots; we were more focused on learning. But now we are at the point where we are deploying Gen AI to our benefit.
We created a group to review the nearly 900 Gen AI use cases across our organization and flag any ethical, regulatory or compliance risks. Through this process, we’ve been able to focus our efforts onto the highest impact, most scalable initiatives. In fact, we just launched a Gen AI-powered tool in our quality control space.
When service representatives get a call from a customer about a particular product, they must very quickly determine if the product is ours, where it is located, whether it has been implemented in a specific patient area, what regulations apply to it and how we as an organization should be responding.
Because we do business in 150 countries around the world and serve millions of patients, we have hundreds of thousands of SKUs, making it difficult for representatives to rapidly make these decisions. But now, reps can quickly access all our product information and history instantly. Our team members can go from weeks of researching these questions to answering them in just minutes.
Are you leading all these AI initiatives yourself, or do you also collaborate with startups?
We do not try to do everything ourselves. We understand there are more players outside our organization than within it, which is why we work with both large and small partners within the broader ecosystem.
Smaller AI startups can sometimes have disruptive approaches to the types of problems we are trying to solve, but they do not always have the same volume of data that we have. If a startup is operating in a truly novel area, we invite them in, run through proofs of concept, and determine scalability. If we decide the project is a good fit, we’ll pursue the partnership further; if not, we both walk away having learned something in the process.
Data is an area in which we sometimes partner with larger organizations, especially if we share a common goal – like finding new biomarkers or targets, strengthening real world evidence or improving drug safety and surveillance.
Our approach is centered on identifying value and partnering with whoever can help us achieve our objectives. When things don’t work out in a partnership, we try our best to apply lessons learned to the next set of collaborators.
With all these AI programs underway, you must have security concerns. What protocols is J&J putting in place to combat the growing risks associated with these types of threats?
As an organization, we watch out for all types of risks, which is why we spend so much time on governance. We don’t view this as an afterthought – it is part of how we think and embedded up front in our approach.
To defend against threats, we’ve created secure enclaves within the company for our large and small language models, and use vetted, curated data. Additionally, we created policies that educate our employees on the risks and restrictions around use of public Gen AI and direct them towards use of our internal generative AI application. We also review every proposed Gen AI use case through a process that includes security and architecture experts.
How does quantum computing fit into your future?
We are taking this emerging technology and trying to apply it to our specific business objectives, like enhancing and expediting our protein folding capabilities.
Today, it is really difficult to understand how proteins fold and find molecules that will match to them. There is a whole set of chemical types we cannot even go after yet because we do not yet understand these underlying relationships.
If we can use quantum computing to think about folding in a different way, we could potentially unlock more chemical structures worth pursuing. To do this, we will need to work with third parties in the quantum computing space who understand both the sciences and our goals and objectives.
Thinking broadly, how else would you like to see AI or Gen AI technologies deployed in the coming years?
As the use of AI models increases, organizations will face greater challenges in the areas of data movement, costs, copyright and intellectual property. Therefore, within our company, I would most like to see improved observability to help mitigate threats in these areas.
There are also evolutions coming surrounding the types of data being used within these models. Today, we don’t create our own large language models, but maybe down the road, if it gets really cost effective, we could. We do use small language models as well, and for both types, there is a heavy focus on prompt engineering and data curation to improve the efficacy of these models. I expect these capabilities to also evolve over time.
For us, it all comes down to outcomes. How can this technology benefit the patients and healthcare providers we serve around the world? When we pair amazing science with amazing technology, we can truly reimagine healthcare.
