EY leverages AI for audit

Ernst & Young has been ramping up its use of artificial intelligence technology to make its audits more thorough for clients worldwide, while turning auditors into AI developers.

Processing Content

Michel Porter, a 39 year-old partner at EY’s San Diego office and global assurance analytics leader who started as a traditional auditor. He taught himself data and AI and ultimately overhauled EY’s global audit data platform Helix, which has grown into one of the biggest business intelligence programs in the world, used in over 150 countries.

“I started in the audit practice in general, but then started to gravitate a lot toward data and the use of analytics, and then that eventually started to merge into cloud-based capabilities, automation, a lot of things at scale for the global firm,” said Porter. 

He has come to focus more on AI. “A lot of where I’m spending my time, both within EY and within the market, is really thinking through how AI and the intersection of data is going to impact our client services going forward, and how it is impacting the skills that we need to create internally,” said Porter. “How is it creating potentially new roles and capabilities within the firm, and then also helping a lot of our clients think through what does that change management look like for them? A lot of what we’re seeing, both internally and externally, in relation to demand for skill sets, is really the high expertise of a domain expert, in our case, accounting and finance, but then in engineering in general. A  lot of people are seeking out those skill sets in the market, and we are trying to create as many of them internally as possible, due to the natural acumen for problem solving and the speed at which we can ultimately develop things internally.”

Sandra Oliver, EY’s global assurance talent leader and an audit partner in Houston, works with Porter. “I’ve been in various roles, but in my global assurance talent role, I’m working very closely with our innovation transformation leadership teams, thinking about all things talent-related in AI,” she said. “We’re a people business. Obviously, there’s a lot of thinking going around about how AI is impacting our work, what that means from a learning and development perspective for our people as we bring AI into our daily routines, and then thinking about what that means to recruiting as well.”

EY has a Shapers of the Future campaign that is emphasizing non-linear career paths for employees and how they’re adapting to emerging technologies like AI. . 

“That perspective has really been about how we can have our people show how what they’re doing on a day-to-day basis really impacts our future,” said Oliver. “We’re really trying to get day-to-day stories of what our people are involved in, what they’re passionate about, and bringing that to life: how what they do personally also impacts the future and really impacts what they do every day. While we’re focused on AI and technology, there’s a lot of things around our younger people. It’s important that they bring their whole selves to the office, to be able to bring their passions and their skills outside of just the traditional accounting and auditing.”

EY has developed agentic AI and other forms of the technology at the firm. “We have AI automation built into our technology platform that actually touches 130,000 of our assurance professionals,” said Oliver. “We are very bullish and proactive about the use of AI in our professional daily work. That’s how we foresee the future. We see AI as being complementary, not replacing the need for accountants or auditors, but it’s really going to enhance their roles. It will allow them to really focus on the higher-risk areas, while AI is going to focus on the more objective, routine administrative tasks. We’re very focused on building that responsive use of AI learning into their daily work, into their upskilling as well as into broader core curriculums.”

EY’s audit and assurance people are able to leverage the technology. “Right now everyone is trying to [move] in the direction of getting as much done as possible in as little time as possible,” said Porter. “I think that AI is really at the intersection of that, given its capability to accelerate each individual’s workload.”

AI-powered assurance

EY is pursuing AI-powered assurance in two main ways. “One is embedding AI into our core platforms from an agentic standpoint to aid us in terms of highly repetitive tasks that happen at a global scale, where we need AI to do the exact same thing over and over with a high level of consistency and accuracy,” said Porter. “For those use cases, we’re really trying to embed it into our core audit platforms. We’ve had a lot of releases recently with that full agentic capability being added. The other one is utilizing it at the individual level to really accelerate the throughput of each individual as much as possible.”

He estimates the firm is able to accelerate its assurance work by about 125% to 150% through AI.

“That’s with the combination of the firmwide tech that we would embed into our platforms to be pushed down, but then also using AI as an assistant to really be able to help them on their individual tasks on a day-to-day basis,” said Porter. 

EY is able to audit AI on a day-to-day basis. “We have the usage of it with our own individual people, but then we have our own clients’ implementation as well, where we’re really looking at how we provide trusted services over AI in relation to the market,” said Porter. “We’re also looking at what’s the best and brightest way to audit AI once our clients are ultimately putting that in.”

Billing and cost savings

Despite those extra efficiencies, EY hasn’t seen clients demanding reductions in their audit fees as a result of the increased efficiency.

“There are efficiencies gained, but the client complexity is definitely increasing because we have companies building AI into their workflow, into their systems, into their processes, impacting their controls,” said Oliver. “There’s always two sides to the efficiency equation. There are certain things that are making life easier for hourly savings. There are also things that raise complexity, so that needs to be weighed in, as well as the cost of technology that everybody is seeing and talking about. It’s definitely an equation of all the different factors that go into our conversations with clients about these”

Nevertheless, clients are asking about the potential cost savings from AI usage. “Whether or not it’s boards, CFOs, CAOs, whatever the rank may be, there are a ton of questions either about our implementation of AI or clients’ implementations of AI,” said Porter. “What is the cost associated with this going forward? What is the cost to provide services? What’s the cost to provide products? Everyone is searching for that crystal ball. When I talk to a lot of CFOs and boards in the market, right now one of their questions is: ‘We’ve invested a lot, and now how do we measure ROI? How do we measure the cost of these implementations?’ To then think through was it worth making the investment or not, really thinking through that whole investment life cycle of the systems. I think that the audit profession and assurance services in general are going through very similar exercises, but there’s a lot of levers that are moving in different directions associated with them.”

Some companies are hoping to control costs by using more open source or “open weight” systems for more routine AI work, while reserving the use of the most advanced and expensive “frontier AI” systems for solving more difficult analytical problems.

“Essentially, it’s coming at it from an angle of how do we control the cost and how do we what is that we’re trying to do.” said Porter. “Which model is the right model for the task that you want to do? You don’t need a PhD to answer the phone.”

It isn’t necessary for firms to use the most advanced AI models for mundane tasks. “Pick the right model for the right task, which in reality isn’t a lot different than the way that we manage our human resources,” said Porter. “It’s what’s the right professional, what’s the right skill set necessary to fulfill which job, and then being able to determine which of those models is provided at which time, depending upon the question, is obviously a very core one.”

EY similarly is careful about safeguarding confidential client data from the public-facing AI models. 

“We take the security and isolation of our clients’ data very seriously in terms of what we bring in, and we would never put them in a situation whereby it could be exposed in relation to the market or exposed to any of these providers,” said Porter. “We have processes in place whereby we have certification, we have infosec approvals, we have that entire supply chain well thought through to make sure that we take it very seriously.”

“All of our people are required to go to e-learning on responsible use of AI and how they are supposed to be using tools, the restrictions on clients’ data, the very strict policy that we have, as well as the learning to go with it on when to use it, how to use it, the tools that they can use it on, just because of the risks,” said Oliver. “It’s certainly a new world for a lot of people.”

Avoiding hallucinations

EY is also careful to avoid the phenomenon of AI “hallucination” where the systems produce wrong information that looks like it might be correct.

“Particularly in relation to our assurance services, where our clients are really relying upon our expertise, we take the accuracy of the systems that are coming back very seriously, and we do very strenuous testing essentially for all of our internal platform builds,” said Porter. “We have agentic AI embedded within them to make sure that prior to release, we have the correct accuracy standards and human-based controls after that to be able to check what AI is doing with additional preparer and reviewer setup to make sure that the risk of hallucinations or incorrect answers are dramatically minimized from a risk standpoint. We’re consistently assessing everything from a model standpoint. Although AI is not always correct, we do very much value the professional skepticism and skill sets of our existing professionals to be able to make the final determination and conclusion associated with our client work.”

“Given that there are tons of models that are released very frequently that are all over the news all the time, I would rely upon our internal processes as a firm to onboard the correct models and manage those, go through the requisite security and infosec considerations, isolation from client data,” said Porter. “But I think that what we’re probably seeing is that there are so many different versions and so many different options in that. What we’re really trying to do is put in place processes by which to pick the right ones for the right task, and the operating cost coming off of that, but also performance, also quality of really what it is that we’re trying to do. You mentioned hallucinations a second ago. I mean, 

“I think that’s what, particularly in relation to our assurance services, where our clients are really relying upon our expertise, we take the accuracy of the systems that are coming back very seriously, and we do very strenuous testing essentially for all of our internal platform builds, whereby we have agentic AI embedded within them to make sure that prior to release, we have the correct accuracy standards, and the correct human-based controls after that to be able to check what AI is doing with additional preparer and reviewer type setup to make sure that the risk of hallucinations or the risk of incorrect answers and things like that are dramatically minimized from a risk standpoint. In all reality, we’re consistently assessing everything from a model standpoint, and then we’re also acknowledging that although AI is not always correct, and we do very much value the professional skepticism and skill sets of our existing professionals to be able to make the final determination and conclusion associated with our client work.”