As part of our AI For Growth executive education series, we interview top executives at leading global companies who have successfully applied AI to grow their enterprises. Today, we sit down with Keith Strier, the Global and Americas AI Leader at Ernst & Young.
Keith is responsible for EY’s global AI strategy and practice. He actively advises public and private enterprises on the new value drivers, risks, architectures, use cases, design and operationalization of robotic, intelligent and autonomous systems.
Watch the interview to learn:
- The challenges of implementing AI as a function of the priorities or resources that different countries may have.
- Why the development process for AI is not simply an extension of the software engineering model.
- Along with identifying the highest impact use cases and ROI opportunities, how to work with clients to set realistic expectations about enterprise implementations of AI.
Marlene Jia: Thank you for joining our AI for Growth executive series. In this interview series, we learned from executives at leading global companies who’ve successfully applied AI to their enterprise.
My name is Marlene Jia, you can call me MJ. We’ll be chatting with Keith Strier from Ernst & Young. I think you’re all familiar with Ernst and Young, E&Y, one of the largest professional services companies, a Big 4, and really in the thick of implementations with enterprises.
Keith, thanks for being here with us. Tell us about yourself, how you first became interested in artificial intelligence, and tell us about your first project.
Keith Strier: Like everybody else, my sort of introduction to AI came through the movies. If you do a Google word search and go back 10 or 15 years to look at words like AI and machine learning. You just see these spikes in the searches and they all map to movies.
But since 2015 or 2016, that’s changed. Now there’s steady interest. It’s much more mainstream now.
I’ve been focused on digital transformation and digital strategy, both on operations—the workplace experience as well as the customer—for the past 10 years. My first introduction came in that context. I was delivering digital transformations, both internally as well as externally focused, and it just became clear two or three years ago that artificial intelligence, and more broadly, automation and robotics, became a bigger and bigger theme as companies are pursuing digital transformation or digital activities.
First it was something interesting to talk about, like, “Hey, here’s something else you should consider.” Then it became more of, “Wow, if we’re going to really invest in digital, and you’re thinking a few years out, you really need to be thinking about this.” And then, suddenly it became, “This is the one thing you want to talk about.”
Very quickly, within a period of maybe 24 months, it went from a nice topic to one of the top three or four conversation points for boards and C-Suites and so forth. AI really emerged for me in that digital context.
Because of my role as the Global Head of EY, I have a very global perspective from literally across industries, sectors, geographies, and time zones. I can tell you conclusively that while, a year or two ago, it might have been hard to list five significant companies that prioritized AI—outside of the big tech companies that you think about—today, it would be difficult for me to identify five that have not prioritized the role of AI. This is across all industries, including the public sector, so it’s quite a fascinating time.
MJ: I bet! You mentioned you started with digital transformation. Can you talk about some of the first projects that you had where the digital transformation project really started to go into the artificial intelligence implementations?
KS: One of the first for me was up in Canada, with an organization. A really important mission-driven organization: the Canadian Blood Services, which is like the Red Cross and manages the nation’s blood supply up there. We were doing some really interesting work on the digital transformation of the customer relationship [such as] customer analytics, customer experience. They’re a really innovative organization. Again, very purpose-driven, trying to make an impact in a very real way for citizens.
One of the key business drivers for their plan was finding new donors. They need a fresh supply of blood on a regular basis, but blood does not last that long. So you need to constantly have donations.
It’s getting harder because a lot of the folks who donate were older, so they needed to connect with the younger generation of donors and to get them more focused on blood donation as a mission, and you have to go to where those folks are. Those folks aren’t necessarily camping out on web sites [because] they’re on their mobile phone, they’re in social networks, that kind of thing so.
As we started the project broadly around digital transformation of the customer, where we ended up (among other things) was actually building a chatbot on one of the more popular social networks as a way to reach donors. This was the first “blood bot” in the world. I know that sounds creepy, but it was the first time a national blood organization had tried to reach young people through a social media channel, through an automated bot, as a way to connect with them versus other ways. There’s a lot of creative ways the Red Cross have worked around the world, but this was the first of its kind, and it has been very successful as one of the tools in their kit to reach younger donors.
Again, it’s not that AI is the answer to all your problems, but the combination of digital and social and AI offer very effective strategies to solve business problems.
MJ: It sounds like it’s just part of a larger toolkit and system of services that you guys have and that companies can utilize.
At the end of the day, companies usually have a specific business problem and this brings me to my next question which is, why do companies come to you guys in the first place? Why did they explore artificial intelligence versus some of these other solutions that you’ve brought up, like bots?
KS: First of all, EY advisory is a managing consulting firm, a part of the broader EY family member firms and service offerings. As a management consulting firm, we tend to lead with business issues first. Clients tend to come to us because they have a specific problem they’re trying to solve or a specific corporate goal they’re trying to achieve.
It could be something as simple as improving corporate culture or training and retraining our employees, or more profound, like doubling their market share and expanding into new markets. It’s different for every client.
We always attack through the lens of the business issue and we try to define that first, then you engineer the solutions to that. Again, you have a toolkit, and that tool kit 10 years ago was mostly around mobile technology and cloud and social media. The kit keeps expanding. Today, the new digital transformation toolkit includes all those things, but now you have artificial intelligence, robotics, and autonomous systems. Those round out the new digital equation.
We try to bring a very business-led, design-led focus to those technologies, but ultimately it’s about realizing value, not hammers looking for nails. I think companies are getting smarter to that, and that’s how we approach it as well.
MJ: Some of the companies that we’ve spoken to in the past always knew that they have a business problem, but they didn’t know where to start. Or there may be preceding steps that needed to take place before they can even implement the solution. Talk us more about the framework that your company or team takes in helping people narrow down what they can do and where to start.
KS: This is where it gets both interested and complicated. When you talk about this toolkit, you can’t use the same framework for all of them.
Take, RPA, robotic process automation, versus AI, for example. RPA has some real advantages because it’s pretty well-defined. The technology is reasonably mature. You know what a bot is and what it isn’t, and you can create layers and layers of bots, and they can do lots and lots of different things. But for the most part, you literally develop this bot to execute a certain set of commands to automate a process. You know what you’re getting into and the technology is pretty robust. Frankly, the vendors have done a really good job.
Compare that to AI. AI is considered on that spectrum of automation, more advanced automation, but it’s a completely different creature. An artificial intelligence is not one thing; it’s a spectrum of methods and domains. You’ve got machine and deep learning and computer vision and so forth, and you’ve got a lot of different disciplines under that banner of AI.
They all have different maturity levels. Computer vision is pretty mature at this point, and that’s why we have things like facial recognition on our phone and it’s starting to scale. Whereas there’s other parts of the spectrum that are much less mature, much more experimental and scientifically-driven.
Implementing AI is very hard to do because you don’t implement it like RPA, you apply it. When you apply something versus implementing it, it’s more experimental, it’s more continuous innovation. You have a hypothesis, you try a method, then you have to fine-tune that method.
Whereas RPA is more predictable. When you move forward [and] when you ask about readiness, the issue is that you have to appreciate that these different tools and the kit have different requirements and that there’s a different level of readiness for each one of them.
One model, one framework to solve them all, doesn’t really exist, so you do need this multilayered approach. Sometimes I.T. refer to it as bimodal: there’s a fast lane and a slow lane, but it’s even worse than. It’s multi-layered, it’s not two lanes, they might be 10 lanes.
Part of the struggle [that] companies are having is that they’re very good at deploying enterprise technology and they’ve figured that out over the last decade. But this is not enterprise technology, this is a combination of science and methods and I.T., and it requires nuances that go beyond the traditional I.T. skill set. It’s a new generation of skills that need to sit on top of the existing skills, and that’s one of the challenges of doing it well.
MJ: Can you walk us through the projects you’ve had in applying artificial intelligence? I understand you can’t name names but it would be great to hear about one.
KS: There’s numerous examples. I’ll give two to make it diverse.
One is more of a strategic example and one more an operational event. As I mentioned earlier, this is not just a private sector thing, this is a public sector thing as well. Governments all around the world are adopting artificial intelligence and robotics just as quickly as private sectors.
One of our clients is the roads and transportation authority of a large country in the Middle East. Very advanced thinking, very sophisticated in their approach. What they asked us to do is to reimagine what might be the roads and transportation experience for their citizens in the age of AI. What would that look like, and how can we improve, not just the efficiency and the energy use, but even the satisfaction and happiness of the citizenship? This would cover everything from optimizing bus routes to flying cars. Everything you could possibly imagine.
That’s a really important exercise, because the last thing you want to do is just go and start implementing stuff and not knowing why you’re doing it. What’s the reason, what’s the value… In that exercise, it was about strategically scoping all the possible investments but linking them back to the strategy, the vision, the priorities of the country. So that’s one example on an infrastructural, the public sector side.
Being bit more tactical, we work with a number of heavy industry clients like oil and gas. One of our clients here in the U.S.—it’s a global company but based in the U.S.—they identified dozens and dozens of possible use cases, but one was very specifically focused on well management, which is one of the daily activities that they have to do across the world. There’s lots of wells and it’s the lifecycle of how to manage that well and all the operations. It’s not a very glamorous or sexy job, but it’s essential to the operations of that organization.
Frankly, the systems that were in place to manage wells are pretty dated. There’s software that was written 20 years ago, and there’s a lot of manual entry, and it’s written on COBOL. There’s aging engineers who are in their 50s and 60s, reaching the end of their career, and they’re the only ones that know the system. So you’ve got that risk as well.
We came in and created a relatively straightforward system that combined robotics and cognitive tools to capture some of the knowledge of the aging engineers, as well as to robotize and automate what became a very manual process operating these dated systems. Frankly, the ROI of this is just staggering. It’s something like a hundred to one.
Instead of needing five really experienced senior well engineers, you only need only one now. A lot of the manual data entry [that was needed] between the systems has all been robotized. Some of the judgment and knowledge around the systems has been captured in a cognitive system as well to start building sophisticated expertise that could help mitigate brain drain [and] overtime. That is a very practical solution to a daily problem, but it’s one of limitless possibility.
That’s the challenge. The possibility space is so broad. We do need to apply some discipline and balance it off with enthusiasm for the technology, because you could spend all day long piloting stuff and waste a lot of resources doing it.
MJ: Now that you’re talking about the oil and gas phase, I’m sure that it’s global. You mentioned earlier that it’s actually very different across locations and countries.
When you were working on this project, did you notice implementation differences across countries? Different needs? Or was this primarily through the headquarters?
KS: Any global implementation is going to have local nuances. I wouldn’t characterize this as a global implementation. It was developed locally. But, to your point, absolutely there are differences.
There’s two big issues that come up when you view this through a global lens. Number one, the laws are different. Look at the GDPR legislation that’s hit Europe, for example. That’s only a European policy, but complying with it is so intense and so difficult, a lot of companies are going to implement that globally anyway, because it’s almost impossible to maintain separate systems.
There are local regulations, some privacy laws and other kinds of things that are very unique. When you start getting into big data sets, as it relates to customers for example, it gets a lot trickier than well management and backend operations. That’s less regulatory.
The other part that comes up globally is the talent issue. I mention this country in the Middle East. There’s actually a lot of AI activity in the Middle East. There’s a lot of innovation in almost every country, but they’re small countries. Their ambitions are bigger than their ability to staff up locally. I was recently in the Baltic region, meeting various countries like Lithuania and Estonia. Very advanced in their thinking, very sophisticated in their plans, but there may only be a dozen people the whole country who have the sophistication or the training to do some of this stuff.
The talent issue varies country by country, and when you develop a strategy globally, your ability to execute it locally could be partly impeded by the inability to find the right people. You can always import some of that talent. As a global company, you have the benefit of having other global partners and leveraging resources from different countries, but I definitely think that’s one of the bigger challenges globally.
MJ: How are you seeing the different countries solve for that? Do you see collaboration? Initiatives?
KS: There’s a global arms race right now to gain prestige. Countries are investing in making this a national priority. Both U.K. and France in the past month announced their national strategies. The White House had a summit on AI just last week China and other countries have been very vocal about it…
MJ: Justin Timberlake had his music video at an AI conference in Asia.
KS: One of the main things the countries are doing is articulating a policy and having the framework. That’s a great place to start. Within that framework, a big part of that will be education retraining. It’s not just new people, it’s taking the existing workforce and an environment that’s attractive.
One country that stands out—maybe not so much for AI just yet—on block chain is the country of Malta. Malta is a very tiny country in southern Europe. This little island nation off the coast of Italy. They created an economic and a policy environment that’s very attractive to cryptocurrencies and to the blockchain companies. So a lot of cryptocurrency startups and a lot of talent is relocating to Malta, because it’s a great place to do that.
Ultimately, you’re going to need a generation of talent to be trained in the smaller countries. So it’s going to be uneven for a while. The big countries [will] dominate for the shorter term, but that will change.
MJ: We have also seen some corporations do the retraining. It’s really smart, where they offer a new machine learning workshops and classes for their engineers. That’s really smart, because hiring is quite crazy and difficult.
KS: I totally agree. When I do talk to companies, I tell them that the only path forward—the critical path—is not to rely on outside contractors and consultants forever. You need those now, and there’s a period of time where that’s going to be important, but the long term strategy is to build up capabilities. The only way to do that meaningfully is to train.
You can’t hire everybody, and you certainly can’t hire enough because the truth is there is a shortage. I don’t know the exact number, but I’ve seen different statistics about there being 25000 to 30000 people in the world that have a specific deep or machine learning capabilities necessary. That’s not that many people.
The answer for scaling and maturing this capability will be a combination of some hiring, some training, and relying on some outside partners. You’re going to need all three to execute AI at any scale.
MJ: Last question from the global point of view, have you seen specialization of artificial intelligence research in any of those countries? For example, in country X Y Z is really good at this. Have you seen any trends like that?
KS: A little bit. Canada has been the birthplace of a lot of our deep learning methods and capability. There’s a concentration of data scientists that have led the charge in developing a lot of the deep learning methodologies, like George Hinton. You see the big technology companies have set up shop in Toronto or, recently, Montreal. They’re buying companies, opening up offices, as they’re tapping into that.
Around the world, it’s a little hard to say. It’s less about the development side and more on the adoption and applications side. You definitely see AI being adopted at scale in the public sector in Asia faster than in the U.S. China has been surprisingly vocal and transparent about their adoption of AI in the running of government services and, in particular, public safety and policing. There’s quite a few headlines on that topic, whereas I don’t see Western countries talking about it as much.
A little bit of this could be [for] show, and I’m not saying that’s not happening…I believe the headlines. It’s hard to answer the question because not everyone’s being as open. There’s a lot less PR in some countries.
MJ: Bringing this back to enterprise implementations, a lot of people are very curious about how long these implementations should really take. How long these pilots be? How much should you be experimenting and then moving forward? Maybe you can give us some more real context there.
KS: One of the challenging parts of AI is that, again, people tend to want to implement a classic I.T. model. A software engineering model. When you do a classic I.T. model, or even a business case for making investments in I.T., you look at costs and benefits. There is a pretty straightforward template for that.
Templates don’t work as well here, and one of the reasons is that the art of the possible is so broad that it be in sort of a pilot mode forever. You can be POCing everything for years.
One of the filters that you have to put on this, which is a little bit less important than I.T., is this readiness filter. Let’s just say you identify 50 uses cases—and I’ve had plenty clients who have done that—and said look, we’ve identified all these great use cases games. And they’ve even identified the possible ROI. Typically you’d stop right there and say, all right, let’s pick the top three and let’s focus on those.
That’s a reasonable thing to do, but here’s where you would do it differently with AI. You need to dive much more deeply and do your homework on the readiness. Even if you have identified three of the highest value potential use cases, and you’ve scoped them out and know what they could be worth, you then have to go back and say, are we really ready to do this use case?
For example, do we have the data to power the model for this use case? Do we know which technology vendor or vendors we might need to support this, and are they ready to serve us? Do we have the internal or external access to capabilities and resources to support the ongoing refinement-curation-training of this model?
You might identify three great use cases, but then find that, for a variety of reasons, it’s not a good idea to proceed with those. Then you’ve got to look at the next 3. It’s not such a straight forward analysis. You might have a great value story and not a lot of readiness, you’ve got to triangulate those.
MJ: Where do you see people go wrong in asking those questions? A lot of organizations do try to ask those questions, but we see a lot of implementations go wrong. [For example], if something doesn’t quite fit,or maybe they aren’t ready [because] they don’t have the data.
KS: One is readiness and just not peeling the onion enough to make sure that everything is in place to pursue that use case. The other one is setting some unrealistic—or not setting realistic—expectations.
When you make the statement that I.T. is implemented but AI is applied, this isn’t a fancy word game. That word, “applied”, has some important implications. One of the implications is you’re not going to have as much predictability to the cost or the outcome, and that’s not always a fun conversation to have with an executive sponsor. They want know how long is it going to take to do this, what is my resource commitment, what should I expect on the talent and when can I expect it.
These are reasonable questions to ask, but it can be quite disparaging when you sit down and say “we have a hypothesis, we’re going to apply this model, [but] we don’t know if it will take three weeks or 13 weeks to refine that model. We think this is the outcome but we’re not really sure.” The conversation is really unsatisfying.
It’s a discovery model, [but] that is changing. There are some parts of AI that have more maturity, and it’s a lot easier to build a chatbot today than it was two years ago. There’s low code and DIY and non-technical approaches that could piece it together. Even IBM is offering deep learning as a service. You’re seeing these capabilities mature, but you still have to appreciate pragmatically that these are still emerging approaches and technologies.
The big mistake is not setting that expectation up front and not socializing or educating your executives sponsors to understand that this is not an implementation with a finite timeline. You still want discipline and accountability, but you want a very realistic assessment of what it might take to get to an outcome and that it may not be completely predictable. That is one of the major failures points.
MJ: I can completely understand. If someone told me that, I’d say, “Wow you came with no answer.”
KS: Well that’s right. Not every client is ready, and I’ve had tough conversations with clients who are really excited about learning and for computer vision and chatbots. They want to implement them and they want 25 billion savings in six months.
We would love to do this work, but that’s not a realistic expectation. We don’t want to set ourselves up for failure. If we sit down and have a more holistic conversation about what it’s going to take, that goal may be achievable but it may not be. We might need more time, the value may not be in savings but it might be realized in other ways. If we can’t get there then we don’t take the work.
The challenge is having that conversation with people and helping them understand the nuances of the different projects.
MJ: I wanted to end with one last question, which is, what are some of the most unexpected things you’ve learned through all these implementations and seeing all of these companies across different industries?
KS: First, the surprise for me was how quickly this has happened. I mean I’ve been consulting for 15-20 years. We’ve seen waves of innovation and we’ve been through the cycles.
The digitalization of the enterprise is still ongoing. As mature as that conversation is, there’s still plenty of companies that have a long way to go. But [AI] has gone from this niche academic conversation to a top-of-mind, across the board, urgent conversation in such a short period of time that we’re all struggling to keep up. The pace has been really surprising.
Again, it’s not focused on one area. It’s like every region of the world, every sector of the economy, all at once. My peers in the consulting industry and the vendors are all just doing their best to keep up.
On a bigger level, very few topics stimulate the mind and the imagination like this one does. It’s such an intriguing topic. The human pupil dilates when it’s excited, [like] when we see something like fireworks, to let in more light and to process that information. What I find is that when I’m able to really educate teams to truly understand what AI is and how it works, their imagination dilates. The possibility space expands. Suddenly it’s like, “Wow, this isn’t just about automating some processes and creating some efficiency. This is about fundamentally changing the way we can run our business. This is changing the way we serve clients and adding to the services we have.” The transformative potential is really striking, and it takes some time for people to truly grasp it.
This isn’t just about incremental human performance, this is about machine scale performance. It’s about potentially millions of percent improvement, not 15 percent. The order of magnitude of impact is pretty cool to see when someone finally gets it.
It doesn’t take that much conversation, but the surprise is these “aha!” moments, which is kind of cool. I think that a lot of companies are gearing up to have those aha moments, and then it has to get back to the grinding work of translating those aha moments to real value. This isn’t just rolling out software, this is a much bigger deal than that, and that’s cool to be part of.
MJ: For sure. I think we’re working in a very exciting space and it’s an exciting time, to say the least.
Thank you so much for your time, Keith. Hopefully we can have a different conversation at a different time, because there are several things that you said that I wish we could cover here, but we’ll have to bring you on another time.
KS: I look forward to it. Thanks for having me! Thank you!