From P&G to Microsoft, How Companies Are Speeding AI Adoption

AI’s rapid evolution in the workplace—from tentative experimentation to agents capable of handling complex tasks—has created a moving target for businesses trying to keep pace with the best ways to use the technology effectively.

That rapid clip—and the maxim that the only constant is change—is reflected in the case studies by Harvard Business School Associate Professor Iavor Bojinov. Between January 2025 and June 2026, Bojinov has written eight cases about different companies’ approaches to AI. As the landscape has evolved significantly in just 17 months, he has made multiple revisions to his research.

From “Building an AI Factory at Procter & Gamble” to “Whoz: Building an Agentic Operating Model,” Bojinov’s cases trace both big and small companies on the different stages of their technology journey, with some taking a slower, old-school approach and others diving right in. For example, P&G worked methodically to build trust in AI among employees, partly by letting them know that if something went wrong, it was on the company. Meanwhile, Whoz, which provides AI talent solutions, sped forward, informing employees that AI would underpin every aspect of their work.

“We have a couple of startups that show the frontier, and we have a bunch of legacy companies that are trying to get there, and we talk about strategies for them to bend that adoption curve,” explains Bojinov, the James Dinan and Elizabeth Miller Associate Professor of Business Administration.

In an interview edited for length and clarity, Bojinov provides five pieces of advice for business leaders about how to encourage employees to incorporate AI into their work, as well as how to make the most of the technology.

Create three layers of trust to nurture buy-in

"I wrote a case on Microsoft, 'Microsoft Customer and Partner Solutions: The Deployment of Copilot,' which looks at the sales organization at Microsoft. These are the people whose job it is to sell you Copilot, and among them, adoption of AI goes up, hits about 22% to 25%, and then goes down to 5%.

You essentially need people to trust the tool. There are three layers to that trust:

1. You need to trust the algorithm and the AI. Trust that it’s accurate, it’s going to give you consistent answers that are of high quality, it’s free from bias, and respects your privacy.

2. You need to trust the developers of the tool. You need to know this tool wasn’t built to try to replace you. If you’re training the tool to replace you, you’re not going to adopt it. You need to know the developers understand your needs, and that this tool is going to satisfy your needs.

3. If things go wrong, what do you do? Are you empowered to take control and fix it? And if you don’t, whose fault is it? Are you still responsible for the work, or is that the responsibility of AI? And the flip side is: If your work gets better, do you get the benefit, or is it the folks who built the AI system who capture that value?"

Figure out the AI-related challenges, then address them

"How to get people to adopt AI depends on which of the failure points you have. If you don’t believe the AI is of high enough quality, it might be that you just need to wait six months and then come back to it.

If you look at the P&G case study, there are concrete examples of how they try to get trust in all three of these pillars. For example, to get trust in the developers, they worked really closely with the folks who were supposed to start using the AI factory to show them that it would work for them. To make sure that their needs were met, they ran these hackathon days where [employees] could go and try it.

To build trust in the processes, they took ownership of specific steps, and they said if something goes wrong, it’s on us and we’re going to fix it."

Encourage employees to scrutinize AI’s output

"Humans will say, ‘this is what I want to achieve,’ and then agents will produce initial drafts, and people will work off of those drafts to improve and finalize [the work].

That’s already how many software engineers work. They’ll have five to six different agents, maybe more, simultaneously working and writing code. They go back and they check it, and they tweak it, and then they send it out again. And I think this is the direction that most organizations, especially in knowledge work and information work, are going to shift toward."

Learn from companies taking bold steps—and reaping the rewards

"Whoz has basically said: This is how you work now. It's through this platform, and this platform is our operating engine, and it’s used in every function, including marketing, product management, legal, and document creation.

They're getting things like 50% improvements in their productivity when it comes to writing code, and they have similar numbers across the board, in terms of how long it takes them to do work, the quality, and the personalization.

Many of the startups operate like Whoz, and a lot of the large organizations are starting to move in that direction. For example, we wrote a case on JPMorgan and their generative AI tool called Connect Coach, 'JPMorgan Chase: Leadership in the Age of GenAI,' which has the same architecture as Whoz, but the tool sits on the sidelines. It's like a little widget that has tremendous capabilities, but they haven’t pushed employees to really pivot away from how they get their work done.

If you’re a legacy company, you need to see what your organization could look like and figure out how you can bend your adoption curve."

Ask employees to set aside time to play with AI

"What most organizations get wrong is they don’t create space for people to experiment. You need space for experimentation, and it needs to be both structured and unstructured.

JPMorgan basically had a challenge, and Microsoft did as well: ‘Save 10 minutes every week for AI.’ At JPMorgan, there are folks who start their leadership meetings by going around the table and having every single person share a key AI use case from the last week. You start to create this environment where people are curious and experimenting.

Try to identify people on the team who want to do this type of experimentation and are good at it. Maybe you can't get everyone to focus 20% of their time on this, but maybe you can afford to have one person [take that time] and have everyone else, in the short term, pick up the slack for that person.

This is where you need leadership support. For instance, at Microsoft, it’s hard for the sales associate who has a target to meet to take the time to experiment, because their job is at stake if they miss their sales target."

Photo credit: Russ Campbell