AI for Every Role Across the Enterprise

Christine Calhoun:

As AI becomes more agentic, how are different types of workers — across roles, skill levels, and industries — using it in different ways? Shishank?

Shishank Gupta:

That's interesting, right? I think the question really is about what will AI do? I do believe AI is for everyone, every day. Now, how you do it, where you use it is up to you. And we've seen examples within our own organization. I'm sure Jared is seeing it within his, and we're doing it for our clients. Now, I'll pick different personas. For example, there are highly skilled, very few resources available, whether it's in the legal and contract side of things. They're using agents, they're using AI to go through plethora of documentation, make sure they don't miss any clauses, they're able to call out any risks in any of the contract commitments and so on and so forth. Then there is the other side of the spectrum and I spoke about the operations team. They're using it to drive efficiency, automation and outcomes in a consistent fashion, create more knowledge assets and so on. There's leadership who's using this for dashboarding, for decision-making, augmenting their decisions in what they do. And interestingly enough we are also seeing, apart from the marketing and other advertising in those use cases, we are also seeing use cases for our frontline users. That's an interesting one. You think it's more technology driven, how would frontline users use it? We see that the frontline users are also leveraging AI in the way they interact with internal systems, whether it's about getting help with policies or anything that they need. Now, obviously the way we access and use technology is different. Developers will use it from their laptops in the way they're so comfortable using. Whereas maybe a frontline user is more comfortable using an audio-based interaction multimodal model, wherein they can interact with an AI system by just voice prompts. They don't know how to write technology prompts. They can't engage with models and so on. So as I said, AI is for everyone every day. We just have to define the use cases, having the right mindset towards AI and saying, why not AI will help us uncover the use cases for every persona, whether it's a knowledge worker, a frontline worker or anybody in between. I think it's just for everyone.

Jared Spataro:

Well, I think it's worthwhile, it's useful to kind of take a moment and look at the technical underpinnings. When we're trying to make AI useful for everyone, we start first by giving the models access to data that's relevant to answer the questions or perform the tasks that people have. That's called grounding. It's something that we've done as an industry now for two and a half years or so. It has its limitations, but it's very valuable. On top of grounding, we're now finding that we can teach the models how to use specific tools. So for instance, at Microsoft, we've been hard at work teaching our Copilot product how to use the Excel application as a tool to do financial modeling. Finally, we're finding that you can get down to the model layer and you can actually train the models on tasks associated with jobs. And when you train the models at this level, they can become incredibly good. In fact, we're finding nothing can beat a model that's trained for a particular domain, at least in our experience. So then when we look at individual jobs, as Shishank was talking about, you have a tool that's really purpose-built at some point. You're grounding in the right data. It has access to other tools, really specific specialized tools like calculators, computing devices, all sorts of different types of tools. And finally, it itself as the model might be trained. So when we see that, we are in the process of trying to produce products that a finance professional could use to really accomplish the work in a much better way, automate that work. And the same applies for just about every domain that you find in a corporate setting, whether that's engineering, R&D, whether you're looking at marketing sales or any other. So that's where we think the state of the art is taking us is into these really specialized types of tools that have some technical underpinnings so that we really can realize what Shishank is talking about. AI for everyone, but not in vanilla type of way, in a very specialized role-specific type of way.