Agentic AI is becoming principal way AI will be scaled across enterprises

Bain & Company's senior partner Pascal Gautheron says change management is super important in deploying AI at an enterprise level

Pascal Gautheron, senior partner and global leader of Bain & Company’s Enterprise Technology practice, believes that agentic AI will be the principal way for AI to scale in the enterprise ecosystem. In a video interview with Shivani Shinde, he talks about why AI scale adoption is still slow among enterprises, why RoI is still a hurdle, and why tech spends will only increase. Edited excerpts:

What are some of the trends you see in enterprise tech adoption of AI?

Essentially, what we saw in 2023 was the discovery of a new approach to AI. The year 2024 was all about businesses gaining confidence. This basically means businesses were going beyond testing. In 2025, we’ve started to see enterprises asking themselves how they can truly implement AI across their entire organisation. Towards the end of last year, we saw the emergence of packaging AI into agents. So, agentic AI is becoming the principal way AI will be scaled across enterprises.

Our conversations gained a lot of momentum in the first quarter of 2025 with business leaders who are now saying, “Okay, I’m confident that the technology can do what it claims.”

Where is India in the race to adopt AI in processes and core systems?

When it comes to AI and its adoption, the conversations we are having in India are mirror copies of the conversations we’re having in Southeast Asia, Japan, and Australia. They are at the same level of maturity — or emergence - depending on how you describe it. In other words, I don't see India being behind in any way, shape, or form.

 

Recently IBM’s Arvind Krishna said at IBM Think that only 25 per cent of enterprises have seen ROI on AI investments. Your thoughts?

 

That's true. Whether that number is 25 per cent, or slightly more or less, is debatable. We have seen return-on-investment proof points in production at scale, but primarily as single-use case successes. What we haven't seen yet is an enterprise rewiring itself to deploy AI across an entire supply chain or value chain. That hasn’t happened because the approach needed to connect different elements of AI to redefine a business around an AI-centric model isn’t there yet.

From a technical perspective, the capabilities needed for governance and control at scale are still lacking. On the other hand, for single-use cases — where you can control the variables — the technology is mature. The performance of large language models, small language models, and the ability to feed them the right data are all already available for isolated use cases.

 

Considering the current macroeconomic and geopolitical disruptions, do you think AI will now become even more important?

 

There is broad-based uncertainty around global supply chains. What we’re seeing with our clients is a greater focus on driving higher performance in specific parts of their supply chains. Right now, enterprises are taking more of a domain-based approach. 

So far, it’s less about responding directly to global uncertainty and more about optimizing within specific domains. It’s hard to move away from your ERP, your core banking system, or your core insurance platform. But what's really interesting is that as some of these systems come up for renewal, companies are starting to ask whether they should go ahead with traditional upgrades or consider the use of agentic AI.

 

We're also seeing a mindset shift around data. The traditional approach involved building massive data warehouses or lakes, which often became dumping grounds for all organisational data. Now, agentic AI enables the creation of virtual data products on the fly to support these agents. That fundamentally changes how organisations invest in core data: it's now about enabling virtual agents to operate, rather than building large, static data lakes. 

We're essentially expecting the proportion of an organisation’s spending on technology to increase, not decrease —even though the cost of performing certain tasks will come down due to technology-driven efficiencies.

 

When you talk to clients about an AI roadmap, how significant is the change management aspect? 

It’s massively significant, super important. The way to handle it is to invest almost as much in user testing during the design and deployment cycles as you do in building the actual solution. 

The second major change management challenge is organisational transformation across all levels. Reskilling and retooling the workforce are core parts of the transition. 

There will be a significant replacement cost. On one hand, adopting AI and GenAI skills will optimise parts of the supply chain, engineering functions, and more. But on the other hand, it will require investment in new areas and skills.