Deloitte Takes ‘Holistic Approach To AI’ To Modernize Business Environments

‘Everyone from a development standpoint has IDEs (integrated development environments), and you may be out there writing code. But the AI ecosystem to go from using large language models around requirements to defining and identifying requirements from existing client data all the way through the development process to automated testing, we bring this into a holistic platform we refer to as Deloitte Ascend. It’s a holistic approach to AI. The unique thing we're doing is diving into the individual sectors to find out what each client's opportunity is,’ says Marlin Metzger, Deloitte's application modernization and innovation offering leader.

AI has become an increasingly important tool for global IT advisor and solution provider Deloitte in its quest to help businesses migrate legacy systems to become part of their cutting-edge IT environments.

Marlin Metzger, Deloitte’s application modernization and innovation offering leader, told CRN that his company is taking what he called a “holistic” approach to AI to help customers modernize their IT environments, particularly when it comes to bringing legacy applications with millions of lines of code up to date with current business requirements.

“[We are] figuring out the difference between what tech debt needs to be addressed versus where is an organization's intellectual property buried in all those millions of legacy lines of code that often aren't documented or understood by anyone,” Metzger said. “And there’s this whole evolution where AI could create net-new business opportunities, but a lot of clients aren’t able to fully take advantage of what's there for them with AI because a lot of the necessary IP and their data is distributed across legacy applications. Having access to it to truly innovate with AI is something they're struggling with.”

For Deloitte, the opportunities to help businesses with AI has become a major focus, Metzger said.

“There's a tremendous opportunity around GenAI,” he said. “The buzz obviously has taken the market by storm. The opportunity around using GenAI to automate processes, testing, code development, is there. But clearly artificial intelligence and machine learning overall and the [related] consumption of data is also something that Deloitte has been focusing on for years.”

There is a lot going on at Deloitte that shows how far the company has gone in its focus on AI and GenAI. For a closer look, read CRN’s conversation with Metzger, which has been lightly edited for clarity.

What are some things Deloitte has been focusing on regarding AI in the last couple years?

We’re making a lot of large strategic investments in AI and really taking a business-first approach. Everything we’re doing is based around what our clients are going to be looking for. There are sector-by-sector go-to-market approaches around what we believe is going to be there. But then there are the pieces that allow us to deliver faster and innovate faster for our clients. Think about common development platforms. Look at what's happening in the industry. There's a lot of evolution and innovation. We're doing the same thing. I think we've been at it a little bit longer than some of the others in this space. Think about integrated development platforms. Everyone from a development standpoint has IDEs (integrated development environments), and you may be out there writing code. But the AI ecosystem to go from using large language models around requirements to defining and identifying requirements from existing client data all the way through the development process to automated testing, we bring this into a holistic platform we refer to as Deloitte Ascend. It’s a holistic approach to AI. The unique thing we’re doing is diving into the individual sectors to find out what each client's opportunity is.

What is an integrated development environment or IDE?

Think of something like an Eclipse platform, or maybe Copilot or another AI-like version you have, that is basically orchestrating across the entire life cycle. It’s not just for code. A lot of what we bring that’s unique to the market is our in-depth business knowledge and acumen, and really being able to combine it with all the net new opportunities around AI.

When Deloitte talks AI, are you talking about GenAI or AI in general?

Always a good call out. There’s a tremendous opportunity around GenAI. The buzz obviously has taken the market by storm. The opportunity around using GenAI to automate processes, testing, code development, is there. But clearly artificial intelligence and machine learning overall and the [related] consumption of data is also something that Deloitte has been focusing on for years.

Which AI platforms have Deloitte been focused on?

It's literally across the spectrum. There are different models, large language models and foundational models, that are better fits for whatever use cases we're trying to solve. In my space around tech modernization and legacy modernization, we have a very configurable platform. A client may have already made an investment with OpenAI or with Amazon Bedrock or Google Gemini. Our products are configurable to run off of whatever the model is. There may be a situation where the client is already going down that path: they may have some experience, or they’ve already contracted with one of the AI partners, or it’s just a better fit. So which one are we using? It depends. But we have experience and investment across the board in all the big ones. And there’s research that continues to go out and figure out the best application for each of those models.

Have you had a chance to play with other platforms like Grok AI from X?

I have not. My team has a couple hundred product engineers on one side. I’m actually a tech guy at heart, and getting close to and playing with the models is something that I always find exciting. We’re getting the updates from them at this point. But there's a lot of love for it if you're running things locally. With Meta Llama and Ollama, the feedback is positive. There’s a lot of success with those models running things locally. When it comes to everything around ChatGPT-type responses, clearly OpenAI is out there from a market standpoint with a lot of the buzz. And we’re working with other models as well. Anthropic Claude is another other model where positive feedback has come in. It’s an exciting time.

Think back to how we shifted from work around phones and voicemails and all the paper processing that happened 25 or 30 years ago with everything going to the internet and then to the cloud. I think we're really at such a time again. I think we're in the early phases of democratization of these models and people having access to them. We have been using these models at Deloitte for quite a while, AI more broadly. But the level of adoption happening now because of the popularity of ChatGPT when it hit the market, and the fact that all of us in some form or fashion can now access these models, starts the innovator mindset, the new entrepreneur mindset, the business evolution mindset that our clients as well the folks inside Deloitte have around net-new ideas.

What is Deloitte doing internally for AI?

Trainings, trainings, more trainings. We have a whole set of campaigns around AI—AI Academy, AI certifications—across different partners. There is as much there as people can handle, and there's absolutely a focus in getting people certified at that next level of not just adoption but of awareness. It’s what we do. Our clients expect us to come in and understand what the latest models are, but also what the applicability of those models are and how to use them for net-new business benefits in their space. Our AI Academy is heavily invested in, and heavily attended, whether it be in-person sessions or on line. It’s something that I obviously use myself and find very valuable. Then there are CIO sessions that we hold at Deloitte University, which is our facility in Dallas. It’s basically a pure learning center our own teams attend, and we can bring in clients and hold CIO forums and focus sessions where we can share some of the knowledge that we have with them.

How do security issues impact how Deloitte and your customers look at AI?

I won't go too deep in security. I'm on the tech evolution and tech modernization side, and security is a piece of what we do. But I would say we've always had cyber threats, and now with the evolution of AI, there is the ability for more bad actors or other manipulators to create additional threats. So we have an entire advisory practice that is on top of all the latest from a security standpoint. I don’t think AI has completely changed the game from a security or cyber threat standpoint. It’s just the next chapter of having to stay in front of all of the potential bad actors.

As you look forward to 2025, what are some new investments you see your team making to expand your AI capabilities for your customers?

My focus is everything between legacy modernization. Think mainframe, distributed environments, cloud. The mainframe market itself is growing at, depending on whose stats you want to reference, 15 [percent] to 17 percent CAGR (cumulative annual growth rate), and is expected to be a $100-billion-plus market in the next couple of years. We are really looking at that space and figuring out the difference between what tech debt needs to be addressed versus where is an organization's intellectual property buried in all those millions of legacy lines of code that often aren't documented or understood by anyone. And there’s this whole evolution where AI could create net-new business opportunities, but a lot of clients aren’t able to fully take advantage of what's there for them with AI because a lot of the necessary IP and their data is distributed across legacy applications. Having access to it to truly innovate with AI is something they're struggling with.

To directly answer your question, we continue to heavily invest in our modernization approaches to be out there and identify that IP. We already do this today in an automated way, but with AI, there are net-new opportunities to pull out those business rules or intellectual property as well as the data that may have been sitting in archaic databases for the last 30 or 40 years which have grown massively but are not easily operable in an AI environment. So tools, people, and education around modernization is something for 2025 we're going to continue to push. We're also doing that with some of our business partners.

At last month’s AWS re:Invent conference, we had a lot of great conversations, including with the AWS team pushing Amazon Q. The big buzzword in the last several months around AI is agentic. What agentic services are we creating? What agentic processes are we putting out there? There are additional opportunities to partner with the hyperscalers and other ISVs around making our products available as agents to fulfill these agentic life cycles around business evolution that we truly see as the next step forward in modernization.

Deloitte works with the three primary hyperscalers: AWS, Google, and Microsoft. How about others including Oracle, IBM, Baidu, and so on?

Oracle is another one. There’s a big push around moving workloads to OCI (Oracle Cloud Infrastructure). There was a lot of interest in IBM booths at re:Invent, and we have a obviously, a partnership with IBM as well. But from a hyperscaler standpoint, the three you mentioned are really where we drive a lot of our work.

You talked about ‘tech debt.’ What is tech debt and AI’s role in overcoming it?

Tech debt depends on the organization and how you want to define it. Oftentimes it’s just referred to as legacy applications, legacy code, legacy data, systems or databases that are no longer able to be maintained. A mathematical way of looking at it is, where am I spending more money on maintenance versus net-new business evolution. You see this in the public sector space, the government space where it's no secret how much of the federal budget is spent on maintaining these legacy systems versus actually innovating these systems. Also, think of legacy code, legacy solutions that are no longer driving new business innovation. How AI factors into it is where you get into different models and the abilities you have.

I can now go in and essentially identify the legacy code that I'm working with, especially in the early versions of Java or .Net, and do code migration upgrades. I can use AI to understand it. Tactics we use here are around prompt engineering, including scaffolding prompting where you prompt against the model with a RAG-based based process and get a response back. And you can do this dozens and dozens, if not 100, times to actually get a series of prompts where you know what the responses are going to be for those prompts acting on this legacy code. You the n scaffold those answers into essentially a prompt workflow and through that process generate net-new code. So something that would take, I don't know, months to do for a sizable application, you're now doing in hours with this type of process in a highly automated AI-based workflow way. And from a testing standpoint, AI can be applied as well. We're still at the point where a human is engaged to sign off on what's happening there.

There are other areas of tech debt. The biggest one, if you get into really old languages that run on mainframes like COBOL, is being able to actually just understand what they do. You asked earlier about investments. One of the first pieces in that new functionality that we invested in, even before ‘AI’ became a household term, was being able to understand legacy applications with millions of lines of code, which would be absolutely impossible for a human to do in a normal time frame. AI has been fantastic at describing the code in these legacy languages and turning it into simple English. We have a chat version of this as well, so it's almost like you're interacting with the developers that wrote the code. You can ask questions on the code such as, where might an interest calculation be? You’ll be able to see that and all of the interdependencies to trace the data variables throughout the rest of the application. It's a truly fantastic way to make things that were previously very complicated and very technical available to be consumed by business users and others who didn’t necessarily come from a tech engineering computer science background.