ORLANDO — While AI innovation continues to move at a rapid pace, enterprises are still slowly adopting the technology and incorporating it into their workflows and systems.
For many enterprises, the slow adoption rate is due to a need to reorient their data environments. Others are still looking at governance, and still others are thinking about what the future of work will look like.
In this Q&A from the Gartner Data & Analytics Summit last week, Beena Ammanath, global head of Deloitte’s AI Institute, dives deep into the challenges enterprises face when adopting AI technology.
Why does the interest in AI not match the current level of adoption?
Beena Ammanath: The pace of technology change moves at its own pace. Newer models and AI iterations are coming out, but the pace of adoption of that technology and enterprise moves at the pace of change management within the enterprise. That’s two different speeds. That’s the difference between AI and applied AI.
When you take AI as a core technology, yes, it’s moving fast. You see a lot of headlines about it, but when you try to apply it in a bank or healthcare environment, it moves at a different speed. Because when you apply it in the enterprise world, there are a number of challenges that come with scaling AI.
What are some of the challenges that come with applying AI in an enterprise environment?
Ammanath: One of the things about scaling AI is around having the right data foundation.
Most CEOs or chief AI officers are caught in that in-between phase where there’s pressure from leadership to see AI value, but the foundation isn’t right. Investing in the foundation is a bigger investment, not an immediate return. Laying the foundation will lead to more long-term, sustainable value from it.
The second one is in a risk-averse or high-risk environment, like financial services or healthcare, where you want to have the right governance model. If you don’t have the right governance model, you can’t build trust, and adoption naturally slows.
So, governance becomes a big factor when you want to scale.
What shift in data management has led to the emphasis on a strong data foundation being more important now than ever?
Ammanath: Most data management systems and data foundations were designed for some form of data processing. But now we’re looking at versions of AI that mean streaming data, unstructured data, and greater access to data. We’re looking at autonomous agents. These data points were not set up to support these three emerging areas. They were static data points built over time and not meant for streaming or unstructured data.
Now with AI, you need unstructured data to make your AI robust. With generative AI, you need more varieties of data.
How does having a strong data foundation influence the speed of the application of physical AI?
Ammanath: Physical AI is advancing very rapidly. What has happened now is that the software part of AI has grown. The software part has now evolved and can do better, smarter things in the physical world.
Typically, when you’ve talked about AI in the past, we’ve focused more on software than on infrastructure or hardware. Now you’re bringing in more of that intelligence into the physical world. It’s not just manufacturing. You hear about robots doing household work or being an aid. We’ve seen it in hospitals where you’re using it for patient lifting or patient transportation. You see it in banks from a vault management perspective.
Physical AI is definitely growing, but I think that’s also been in the making for a while.
The data foundation for that is actually stronger because of the IoT wave that we had 10 years ago. Manufacturing is more ready because the sensors already exist.
How much does sovereign AI inform data?
Ammanath: It will have a huge impact outside the U.S. Sovereign AI is where there is a dependency on certain organizations, certain geographies. It’s a huge topic outside of the U.S. on what happens in a world where you’re fully reliant on certain companies and geographies for what you might be doing in AI.
It started in the EU and the Middle East, and now, almost all countries are beginning to think about the need for sovereignty, like what needs to be owned, what data needs to be within the country, and what can be uploaded to the cloud.
What are some of the other obstacles organizations are facing in this age of AI?
Ammanath: One is the workforce impact. It’s understanding what the future of the workforce and talent looks like. What does the job of the future look like at the task level, and how do you start preparing? If people gain AI fluency, is it to do their current job faster or to help them prepare for a future job? For example, in a financial analyst’s job, 60% of their time usually goes into gathering and organizing the data and sourcing it, 20 to 30% into modeling, and 10% into actually synthesizing the insights and applying them to the business. Part one and part two can be done with AI. In the next few years, it will become more reliable, and you can have AI handle that part, and the financial analyst will have to focus mostly on part three. But if you train them on doing the existing jobs faster, you’re not preparing the financial analyst for what their job will look like two years from now, which is more focused on part three.
But can we really know what the role will look like in two or three years?
Ammanath: That’s the work that needs to happen at the leadership level.
The first step was to have the entire workforce trained on AI, and I see more and more leadership conversations on what we are training them for. It’s more about getting to the next level of what training should look like for future work. For that, you need to figure out what the jobs of the future are and what the tasks are.
You won’t believe how many times I get asked, “Is AI going to take away the job?” It’s not that it’s going to take it away, but it’s going to certainly change.
My job is changing right now. It’s not this one day, suddenly turning on the switch, and then the job is different. It’s slowly changing all of our jobs. But it’s not sudden.
It’s going to be incremental. Similar to the adoption of AI, job changes are not going to happen overnight. Over time, jobs will change and evolve.
It proves to me again that the adoption of AI moves at a different speed than the creation of core AI. That’s the fundamental difference.
Editor’s note: This interview has been edited for clarity and conciseness.
