Deloitte's Nishita Henry: "Boards are scared" of generative AI

Managing modern cloud infrastructure is a challenge for a lot of companies that operate outside Silicon Valley in the traditional economy. Nishita Henry's job involves helping those companies find opportunities and avoid problems while operating on AWS.

Deloitte's point person when it comes to Amazon and AWS, Henry manages both the traditional consulting services that Deloitte provides to Amazon, such as tax planning, and also advises clients on how to work with AWS. That latter part includes acquiring cloud services from AWS as well as selling their own software on the AWS Marketplace.

After several years of rapid change for clients that finally realized they needed a cloud computing strategy, another disorienting time has arrived thanks to the rise of generative AI.

"I think boards are scared. I think boards see disruption written on the wall" when thinking about the impact that AI could have on their businesses, Henry said in an interview last month on the sidelines of AWS re:Invent 2023 in Las Vegas. In that interview, Henry also touched on cloud migration, cost optimization, and the rise of AI.

This interview has been edited and condensed for clarity.

What do most of these customers that you're talking about — Fortune 500 companies, they've done the migration, and they're into the transformation part of things — what do they need help with? 

A lot of times it is, how do they integrate across their organization? How do they integrate their back office and their front office for a more seamless experience? How do they create new products and solutions for their own customers? How do they transform their business model, so they are disrupting their markets before they get disrupted?

Of course, cost is always a big conversation. So, how are they creating cost optimization so that they can unlock funding for that future growth? How are they making sure that they're not only creating new things in the cloud, but they're retiring the old things? 

I want to come back to the cost thing because it's obviously been a big issue for customers over the last two years, a lot of AWS customers are trying to consolidate their spending as much as possible. Can you give me a sense of how those conversations are happening?

There were some that said, "let's just get everything migrated," and some of those clients actually saw their costs increase, because they didn't optimize, they didn't retire the old things, they didn't transform their underlying business. And that's where a lot of our customers are now, like, "help me, we have to actually get the cost savings we predicted from our business case, and we have to create opportunities for growth."

That's a lot of what we like to focus with our customers on. It's like, "look, yes, you should get cost optimization, no doubt. At the same time, your actual focus should be on developing new services and products in the market to create your future growth."

There are some clients that went in and said, "okay, we're not going to do a lift-and -shift migration, we're going to go application by application and rebuild." Now, that takes a long time, and so then those customers are on this balance of "oh, am I being too cautious, because it's going to take many, many years to fulfill this roadmap versus trying to get everything there." And they're having different parts of the organization do the actual migration, or just choosing to say, "I'm not going to migrate it at all, I'm going to just go buy a new third-party solution to put on the cloud," so that they can go and kind of create a net new process and a net new way to approach the market.

Traditionally, the cloud vendors have been a little hard to work with on cost optimizations. It feels like that has changed in the last few years, would you agree? How did AWS and its customers get there?

I think economies of scale lead you to cost reduction. And I think AWS can actually point to actually proactively providing cost reduction to their customers, because they have created that larger global economy of scale. I think there is more increased competition in the (cloud) marketplace, which is normal and natural, and actually helps all of us up our game in terms of quality, the types of products that are out there, and cost.

I think that's really kind of a healthy competition that's beneficial for the overall consumer, and I think that they are actively working to make sure they're providing a balance of cost for performance. So you've probably heard a lot this morning in Adam's conversation, the chips that they're building, the new storage solutions they have, it's all being optimized for price and performance, understanding that customers are sensitive to that; as well they should be.

You don't have to have the largest large language models to solve all the world's problems, but you need very specific functional ones to do the things you're doing. And that's where you have some of that trade off.

How many of your customers or clients are actually doing stuff with generative AI?

I'd say every one of our clients is absolutely having conversations from the boardroom down. Every one of our clients is doing some sort of kicking the tires, ranging from what I call proof of value to proof of concept.

Proof of value is really being focused on proving to their own internal stakeholders that there's there there; (generative AI) is beneficial to the overall business, it's going to be secure, it's going to be trustworthy, and it's going to truly help their overall strategic objectives. And that's not doing anything (in terms of building product), but actually probably using open-source public data to prove some things out.

Then there's ones at the proof of concept stage that are truly using their own data to help them do something better, faster, cheaper. And so we have a range of clients on that spectrum for sure.

Deloitte's on the absolute proof of concept end of the spectrum (internally), where we're trying to make sure that we understand how we can best use our data? What are the appropriate guardrails we need to have? How do we make sure we're protecting privacy? How do we make sure that we're using an ethical framework or our trustworthy AI framework in order to stand behind the generated information of our models? And what's it going to do to the future professional services?

We always said, bad data in, bad data out. Right now it's like, bad data in, bad data out squared, right?

Everyone's having those conversations, and we're helping with our clients with a range of them, A lot of our clients are also just kicking the tires on the various models that are out there, because no one really knows which is best for which, so there is a lot of trials going on with various models of the same use cases, just to see what you get and how that works. 

You mentioned somebody earlier about conversations happening from the board on down. This has felt like a very top-down kind of emerging technology in the enterprise. When AWS first started and re:Invent started, cloud was very bottoms up; there were developers who were using it because they thought it was cool, they could start a project, they didn't have to ask the CIO for a server. This feels different. What are you seeing?

I think boards are scared, I think boards see disruption written on the wall. They were disrupted by cloud, because like you said, it did happen bottom up, and people didn't necessarily see the future of it. And I think that the speed of tech, the prevalence of it now being the oil that runs all of our businesses, they're very much like, "we need to get ahead of it." So it's definitely been pushed top-down because of what's happened in the past and the lessons we've learned.

In addition, it takes an enormous amount of data to make this work. So this isn't something you can just put your credit card down and start working on, you have to have an understanding of what data you have, is it ready, what are the underlying structures? And is there bias built into your original data structures that will lead you to bad outcomes?

We always said, bad data in, bad data out. Right now it's like, bad data in, bad data out squared, right? It's just something that people have to worry about far more.

How is that going, though? In my experience with the people who work in technology, the actual developers, administrators, those kinds of people don't like top-down orders. Is there resistance internally?

I don't think there's resistance from using it. I think people are actually interested in it, because they can go out to ChatGPT and they can create their kid's birthday invitation with the coolest haikus they can find, right? What I think is the frustrating part is where policy and regulation meets what you can and cannot do, and how fast you can do it.

So in most organizations, people are like, "okay, well, you can use it, but you can't use any of our internal data." Okay, well, then it's not going to be as good to me, so now what do I do; people want to use that internal data. Our legal and policy and leadership organizations have to think really fast around how they're going to create the right guardrails to allow the rest of the organization to move forward. And that's a hard thing to keep in step, but I think that's where the challenge is really. I don't think anybody from the bottom is like, I don't want to use this.