Mohamad Ali, Head of IBM Consulting, joins Jeremy and Henrik to share how IBM used AI to automate 94% of HR transactions, and cut HR operating costs by 40%, all while creating a model other organizations could learn from. This episode offers a rare inside look at AI transformation at scale, from infrastructure and incentives to the people and playbooks that made it real.
As Head of IBM Consulting, Mohamad Ali led one of the most ambitious enterprise AI transformations to date. By making IBM its own “Client Zero,” his team tested every AI solution internally before bringing it to market. The effort began with massive hackathons involving 150,000 employees, turning curiosity into capability and belief at scale.
Mohamad shares how leadership alignment, process redesign, and broad employee engagement drove $3.5 billion in cost savings and renewed growth. Jeremy and Henrik reflect on why IBM’s model may signal the next evolution of consulting — where organizations act as their own laboratories for change
Key Takeaways:
LinkedIn: Mohamad Ali - IBM | LinkedIn
IBM: IBM
00:00 Intro: HR Automation
00:41 Introduction of Mohamed Ali and IBM's Transformation
01:14 IBM's Enterprise Transformation
01:41 The Role of AI in IBM's Success
03:25 Rejoining IBM: A Strategic Decision
04:33 Key Components of AI Implementation
07:21 Employee Engagement and Hackathons
08:59 Technical Leadership and AI
10:37 Global Tax Optimization with AI
11:17 Scaling AI Solutions for Clients
22:00 Monetizing Digital Labor
26:50 Digital Labor and Procurement Projects
27:29 Unbundling and Economic Implications
28:44 Technological Shifts and Market Expansion
30:04 AI-Powered Business Transformations
32:22 Case Study: L'Oreal's AI Integration
39:13 HR Automation and Cost Reduction
42:09 Creative Innovations in AI Applications
43:59 Advice for Leaders on AI Integration
45:43 Final thoughts
📜 Read the transcript for this episode: Transcript of How IBM Used AI to Cut 40% of HR Operating Costs and Reinvest in the Company |
Mohamad Ali: HR is one where we, we apply this like at massive scale and um, we were able to lower our HR budget by 40% because we basically automated all what's called transactional hr. So if you needed to transfer an employee, you can't figure out how to do it in your HR system. So you call somebody. Right now we have this thing called Ask hr.
You go and you chat with it and it helps you out. And so 94% of all transactional HR has really been automated, and we've taken like some of those people and we've repurposed them. We've taken them out of HR and put them in consulting so that those consultants can now go into companies and help them with their kind of HR optimization.
I'm Muhammad Ali. I'm the head of IBM Consulting. We're a small team of 150,000 colleagues helping to solve hard problems with human. Digital labor, which we call the science of consulting. And before we could help our clients, we had to transform ourselves with ai.
[00:01:02] Henrik Werdelin: Again, thank you so much for coming on.
[00:01:05] Jeremy Utley: We actually thought when we originally got in touch with you, we thought we were gonna have a conversation about the future of consulting, which maybe we will. That's great. Congrats on being on the Forbes list. But when your team sent the information about the enterprise transformation at IBM, I mean, our eyes popped out of our heads.
These numbers are staggering. And so before we get to the future of consulting, we gotta talk about the wins that you've led at IBM. We'd love to hear how that transformation up to this point has unfolded. If you think about the strategy and tactics that you've employed, we'd love to start there.
[00:01:40] Mohamad Ali: Yeah. So, um, so Jeremy, even before I talk about the, you know, uh, productivity gains we've seen at IBM using ai, like, I, I think I need to provide a little bit of context, a little backdrop for please.
Right, please. So, you know, many years ago when I graduated from Stanford, I helped start a, a company called Neural Applications Corporation. And it's exactly what it sounds like. It's a company that a bunch of guys from Stanford decided. We're gonna build neural networks, right? And this was like in the nineties, and we built a bunch of software to do neural networks and we put 'em on these discs.
We tried to sell 'em for $2,500 and a few people bought 'em and didn't really know what to do with it. Then somebody came along and said, Hey, why don't you try this on a real problem? And somehow this guy was connected to a steel mill. We didn't know anything about steel mills, but a steel mill is a giant pot.
And you throw all kinds of metal into it and you pump it with electricity and it just melts. And so the guy said, Hey, listen, if you can come up with a better model for what happens in this giant pot, which is like the size of the house, then maybe we can control the electricity better and save some power.
And so we built a neural network that modeled what was happening inside his pocket. And at the end of it, we were able to save like a million dollars a year in power, right? At which point we turned it into an appliance. We started charging $250,000 for one of these things, and we sold 10 of them at, you know, at that point, just with the small amount of data compute and algorithms that we had, we could solve these like incredible problems.
And so that made me a believer, right? And so. But two years ago, I rejoined I-B-M-I-I, I worked for IBM in the beginning, in my early career, and then I left and I was CEO of a couple companies. And then I, I came back and part of that is because, you know, as gen AI sort of hit scale, I realized that you could go solve, like a lot of really interesting business problems with this technology.
You know, not only do you have to believe, but you also have to like know where the pitfalls are and then how to apply it. That's sort of what brought us to, okay, well let's take IBM as that Petri dish and see what we could do with this company itself.
[00:04:01] Jeremy Utley: So that's actually a really fantastic starting point and I thank you for rolling back the clock a little bit because one thing that'd be fascinating to learn is what did you see as, as an outsider with a prospect of rejoining the company?
What did you see about the company that in terms of leadership, in terms of capabilities, in terms of readiness, so to speak, that led you to believe you could affect a meaningful change at this company? 'cause I'm sure you actually could have. Yeah. Undertake a transformation effort in many different places.
Why did you decide to rejoin IBM?
[00:04:32] Mohamad Ali: Right. So as we started doing this, we realized that there were sort of three big components of applying a technology like AI to yourself. One is yet have leadership that actually believes in this stuff and are technical enough to go pull it off. Second is you have to be willing to go redesign all these processes, which is not an easy thing.
And then third is you have to actually have buy-in from, you know, the 270,000 employees we have that this is actually a good thing for all of us. And putting those three things together is really hard for most companies. Now, I knew we could get the leadership because the current CEO, you know, Arvin and I.
Peers for four years. Like we helped build the software business back in the early days. Right. And we're both very technical. He was head of research. You know, I also had a very technical background and I knew he got it. And you know, after we sold the last company and I was sort of just without a job and trying to figure out what Index Arvin called me and he said, Hey listen, I think there's a huge opportunity here.
At IBM, um, and particularly in the consulting part of the business where, you know, we have 150,000 people delivering a hundred percent with human labor, what if we could deliver with human plus digital labor, we should be able to clean up the market? And I said, Hey, you know, I, I think it's even bigger than that, right?
And, and, and he had already started this process of trying to apply almost a pre. AI technologies to transform the company. So there was, there was a real buy-in at the very top of the company that we could use technology to transform the business. And then, you know, the processes decomposing those, uh, you could, you can actually do that when you have that kind of sponsorship.
Then the last piece is, you know, we work for a technology company. Wait,
[00:06:30] Jeremy Utley: well actually, so, so before you go to the last piece. Yeah. You said the second piece is a willingness to redesign, and you said you can do that if you have that kinda leadership. I would respectfully beg to differ. I've actually seen a lot of times where there's leadership in place, but the willingness to redesign among, you know, as some refer to middle management as the mushy middle.
Yeah. The willingness to redesign is actually where a ton of, uh, friction. Encountered. Right. So I, I don't think we can just take that for granted. Can you talk for a second about how did you assess the middle's willingness to redesign work processes?
[00:07:06] Mohamad Ali: Yeah, actually, so maybe it's a combination of the leadership, right?
And the last part of it, which is like, how do you get 270,000 people excited about redesigning what they do? Right? And so if we actually started with these. We're not gonna, you know, it's not about redesigning your work yet. Right. But we just wanted to get people educated. And so we ran these massive hackathons with like 150,000 people participating, using Gen AI to do cool stuff for themselves and their teams.
And they got really, really invested in it. And there were contests. And so as people got invested in it, we said, Hey, you know, how can you apply this to make your area better? And in some ways we started harvesting that those like there was a bottoms up, a tops down. We have people in the tax department, so in the tax department who've applied Gen ai, we have reduced a hundred thousand hours.
Now you figure the people in the tax department would say, Hey, this is, this is bad for us. In fact, they were so proud of what they did. 'cause they were competing against the, you know, the fp and a people, you know, out to that
[00:08:17] Jeremy Utley: supposedly creative individuals, right? Yeah, exactly.
[00:08:20] Mohamad Ali: Yeah. And then it started building on itself.
Like we were able to start reinvesting some of that money into more engineering, more products, more sales. And, you know, our, our business has. Growth has flipped by eight points from minus three to plus five, right? It's eight points of growth and you know, the stock price has doubled, right? So I think people are seeing the benefit of it and they're actually like loving being part of it, which is really hard.
I think a lot of companies, it's hard for people to, you know, really participate. I mean, I think a lot of companies, it's like, here's some AI and we're gonna just give it to you and tell you to use it. And this was both tops down and bottom built.
[00:08:59] Henrik Werdelin: Could you talk a little bit on the, you said you gave the background of you and your colleague, the CEO, being technical, um, a lot of CEOs that we talked to are excited about ai.
Where does it make a difference that the leadership is technical and if they're not, what would you advise kind of companies that do not have technical CEOs to kind of do in order to understand it at a level that is technical? I
[00:09:27] Mohamad Ali: mean, you guys know this better than anybody. You're, you're in the industry.
You help drive startups, et cetera. There's a lot of hype, and I think a big part of being technical is being able to figure out what's hype and what's real. And I think we were able to quickly narrow down, like after we did all that hackathon stuff, we ended up with 200 processes in the company that we wanted to redesign.
And then we narrow that down to 70 and we picked 70 where we knew the data we could get to the data. Right? You know, you guys know, like in ai, it's really about the data, the compute and the algorithms, right? And so we needed to make sure we could get quality data for these problems. These problems could be decomposed.
These problems like there's like low hanging fruit. It's like very much document oriented, which is, you know, really a good use case for certain things. And so we were able to pick out the applications very quickly that were actually the workflows where we're gonna see benefits, like the tax processing, for example.
Right. Could
[00:10:37] Henrik Werdelin: you tell me, just to the, to the, the texting, just curious like what's the actual use case there? You mentioned a few times.
[00:10:43] Mohamad Ali: Oh sure. You know, we're a global company. We operate in a hundred plus countries, we have to file taxes everywhere. There's a ton, a ton, a ton of paperwork. Right? And so parsing through all of that, being able to prepare these documents, being able to like optimize a variety of things, you could actually like decompose that workflow and say, you know, at this stage.
It would be really great if I had something that parsed through these documents and highlighted the things that mattered to me so I can actually do this thing faster, right? So that's a great example. And we're not just applying it to ourselves, right? So we now have built this playbook and we're applying it to clients.
So there's a government client, and you guys have probably used this service. It's one of these things where you go and you submit your documents. Then you Wait, wait, wait, wait, wait. Right. I'm sure you've had that experience with some government service. Right. And so this is the first full year in production where we've augmented the process with ai.
It's very paper-based, well, I shouldn't say paper-based, document based, right? So it's a good application for this. And so far we've processed 2.5 million of these things, and it is 12% faster than last year. The backlog is finally coming down after years of the backlog growing, right? So, you know, we're doing it to ourselves, but we're also figuring out how to do it.
In client scenarios and in the client scenarios. Henrik, I, I see you're looking at me. You know, quizzically, a big part of it is, I don't think it's technical anymore. Like we're trying, we're figuring out the technical part. It's all those people things around how do you get, you know, people. That's funny.
[00:12:25] Henrik Werdelin: That was exactly actually the question, uh, is because where you obviously has a very unique insight is you guys are technical by design, I'm sure, and culture, but obviously also a consultancy. And so by now. If people are trying to get benefits out out of ai, how much is that actually new technology and how much is this is just atomizing workflows, understanding where the technology is at this state, and then applying in many ways, pretty off the shelf technology, uh, in many cases.
[00:12:59] Mohamad Ali: Yeah. I think you're hitting on something really important. It's not just. Hey, listen, I've got this giant philanthropic model and I'm gonna throw it at these documents and magic's gonna happen. Like that's actually a small part of the problem actually being able to understand the workflows and decompose the workflows and yeah, I mean that's something we've been doing for years for all kinds of clients, right?
And so that's helpful. Sometimes we use AI models that aren't like a giant philanthropic, like there's a time series model that we use. It's actually an LLM. If we didn't have that, we'd use a traditional time series model to solve particular. And so we're actually one of the things that we built in the consulting organization.
So we have 150,000 people, right? And so everybody's gonna go sort of do their own thing, but we needed to organize that somehow. So we built this thin layer of software and you know, as a software designer myself, like I help architect it, which you could think of it as an AI virtualization layer. And so then on top of that, we built like these things called digital workers, right?
They use specific tasks like the tax thing and so forth. And then below that, like all the models you could possibly want, the AI models. Also non-AI models, right? So now we can sort of pick and choose from Llama and from Granite and from open AI and from philanthropic and from, you know, some non non-AI models, and then build these things that are smart and those smart things than we could bolt it into the workflow.
So it's, it's way more than just a, a big model. So you're right about that. It does take all these pieces to make it work.
[00:14:36] Jeremy Utley: One thing that I'm thinking about is you have to think systemically, right? To break down to atomize, understand and decompose a workflow. You actually have to take a step back in order to be able to do that.
Take a step back from the work, right? To think about the work. Yes. And one challenge that we've observed in organizations is nobody's got bandwidth to actually think about their work. They're too busy working in the business to work on the business, so to speak. So it's kind of uniquely well suited to a consultancy because you actually, you already had the objectivity when you're working with a client.
Right. How did you get the teams inside of the organization space to have that kind of objectivity on their own work? Because it would strike me that it was actually probably much more difficult to do it inside of IBM than for a client.
[00:15:24] Mohamad Ali: And in some ways, like we treated IBM as a client, we actually call it client zero, right?
That's the term we use at IVF. And there are really two organizations that supported this and organization. Number one is the internal CIO organization and organizational number two was IBM consulting, which is about 150,000 people of the 270,000. So now you have these two services organization. That are guiding this process, but inside the core of the company, you have, you know, the, the leadership, all the leadership structure, the CEO actually met once a week with the leadership team, um, all the business unit owner, the functional owners, et cetera, to drive this.
There was a team that specifically there were four teams. The second team was, okay, here are all the workflows that we need to go decompose. And then the fourth team actually was how you engage the employee population, right? So these hackathons and that type of thing. Um, but um, but we sort of treated IBM as a client and we called it client zero because part of what I love that.
Yeah. Part of what we're trying to do is, is look, you know, I, ibm, M's been in the AI space since like 1957 or whenever that conference was in New Hampshire, right? IBM was one of the four, you know, like, uh, scientists who were there. And so we've been around the space a lot and I mean, it's some ways, it's why I came back and, and, you know, when Gen AI came into existence, it, it was a bit of shock to like the whole, you know, AI world, right?
Mm-hmm. But it was like so good and so pervasive and so quickly. So, you know, from, from our CEO's perspective, who's very much a technologist, he's like personally like leading our quantum charge. Right. And, you know, he could tell you exactly how the qubits behave, for example. Right. So I think his, can
[00:17:17] Henrik Werdelin: he, can he explain to entanglement to somebody like me?
'cause I'm still trying to get.
[00:17:22] Mohamad Ali: That's, uh, but, but look, if we're not gonna be the ones that bring, you know, gen ai, like in, in that burst that happened two years ago, then we should be the one that shows how gen AI could be used productively at scale, beyond anything else. And I think with the three point wow billion of costs that we've removed.
From, you know, our $20 billion spend, which is only a part of the company, which is 16.5%, which is actually pretty amazing. Like this is probably the largest scale application of this technology in existence. And so I, I think this was really important to him, was really important to the leadership. So, you're right.
I mean, this is, this is a little bit different than what we see at clients, but we have to create a playbook out of it.
[00:18:06] Jeremy Utley: I love, I love, love, love the humility and ambition. Interestingly, entangled there to use henrik's word of, of that statement. If we, if we can't lead to the, call it technology development, let's be the first case study of organizational transformation.
That's a wonderful kind of balance of humility and ambition.
[00:18:28] Mohamad Ali: I do wanna qualify that, right? Because there are certain aspects of this technology where we actually think we're leading, like governance, for example. Like we have a product for governance that is probably the best in the planet. We actually use that in our organization.
You can go into a government and say, I'm just gonna put AI in here. You have to have a whole governance framework on it, right? Small models like the granite models, the time series models, those are things that we develop on our product side. The time series model's actually the best on the planet, most downloaded, et cetera, right?
And the small models. When we started this. Internally we were, you know, using a small number of tokens. Now it's like billions of tokens that we're consuming all the time. And if you run that in a giant model, it's super expensive. So 38% of all of our calls, as of right now, I can watch it and see, goes to a small granite model.
Right? And that costs like way less. And so still pieces is technology that we're, that we're building we're really good at. But I think the big thing here is. When we show that it actually works.
[00:19:34] Jeremy Utley: Yeah. And just, I think we, we had a overlap there. Did you say you've removed $3.5 billion of cost? Did I get that right?
Yes. 16% of the total cost. How do you Yeah. Start to quantify that?
[00:19:48] Mohamad Ali: Right. So, and I think this is part of why this stock price has done so well, because we actually quantified it not just to ourselves, but to our investors, and every quarter. We actually report on it and we would actually show how it adds up.
Right? So, so if you take in 2024, the number is three point, uh, uh, $5 billion out of a $20 billion spend that we targeted, right? And what we would show is the improvement in the profitability. And then we would also show we're investing it. So for example, our r and. As a percentage of revenues used to be 9%, now it's 12%.
So if you take that three points and the additional profitability and, and you know, you, you can actually add it up and get to the 3.5 billion. And so once the analysts can see this, then they know it's real, right? Because you, you guys know what people say they've saved a lot, but you can't actually see it, it dropped to the bottom line or you can't figure it out on the, on the incremental investments.
And once they're able to see it. Um, they give you credit for it. How do we do that? One of the people on that weekly call that the CEO had was the IBM controller himself, right? A guy named Nick. And Nick's job was to get all these numbers so we could, you know, file it in our 10 Ks and 10 Qs, and Nick was not going to get this wrong.
Right? Or like, or you couldn't come in and say, you saved this money and it's not really true. So, so Nick had to keep us on this.
[00:21:18] Henrik Werdelin: I think that's, that's incredibly important and I think not what we, you see a lot of places. Um, I had an interesting, um, conversation a few months back with this guy who basically pitched that the future of HR would not be human resources, but chief resource Officer.
And then, um, and since you have this now framework where you think about people and agents, models, whatever the terminology is in York. How do you increasingly think about, um, both kind of like, how do you sell it to customers? Like do you charge a, you know, like a hour spent on an agent in the same way, or like, how, how do you, what's your kind of mental model for all of that?
[00:21:59] Mohamad Ali: Yeah, so this is a very, very interesting question. Um, because I actually think that in the future, I mean, just like you have. Millions of iPhone apps. You guys remember in the old days, we used to build giant apps and they still exist. I mean, some really good ones like Oracle and SAP and so forth, right?
Heavy duty stuff. Then the iPhone, you know, the mobile came along and we started building instead of hundreds of apps, like millions of small, tiny apps. And then tomorrow we're gonna be building trillions of. What, what I'm calling digital workers, for lack of better term. These are like little bits of software that call an LLM, right?
And they do something very specific. And so these things are gonna exist out there. And how, how do you monetize it as a consulting company? How do you monetize? We know how to monetize human labor. What We don't know how to monitor digital labor. Right. And I think, I think where you're going with this is what we're, we're trying to figure out.
So today when a client comes to us, like for example, we have this chemical company that, that came to us and said, we have, you know, like this is our procurement organization today. Procurement is very paper-based, right? Great application for ai. If, if you were to apply ai, like what could you do with this?
And in this particular case, we've signed a deal where we're gonna reduce. The labor costs by one half, right? It is actually remarkable what we're committing to doing here. Wow. Offer can be done, you know, and like us, they'll be able to reinvest that money in r and d and, and sales, whatever they wanna reinvest it in.
So how do we charge for that, right? So today the charging model is very simple. It's a fixed price project. Uh, we can now do it much more cost efficiently, so we'll charge you less to do it. Actually, um, your margins go up by doing that, right? Because a bunch of what you're putting in there is software, digital labor, software, and that has higher margins.
So that's good for the client. It's good for us. But I do actually envision a world, you know, maybe five years from now will, there'll be trillions of digital workers out there and Germany, you need to do something and you'll just go buy a vodka home and there'll be an Amazon-like place where you're gonna get 'em.
And there'll be a different monetization scheme for that, right? So I think over the next five years, like how this stuff gets monetized will be fascinating.
[00:24:23] Henrik Werdelin: One thing, I don't know if it's at all interesting, but, um, one of the projects I'm involved in is basically this kind of thing of it as, uh, Shopify for the agent web, like where people can come in and they can do something.
And I think when we started it, everybody just assumed that the business model of yesterday would kind of just, you know, continue, right? So for consultancy it's our, you know, rendered and, and for software, it's SaaS, right? It's like X dollars per month. And what we see is that that is increasingly difficult, both because people have subscription fatigue, but also because that obviously these models are basically human wrapped on top of an LL model.
And so the business model that is emerging are, you know, aging is free, but you pay for human in the loop or, uh, purely success based or the specific output have a prize and new interesting models. You know, I saw the other day where somebody, they will pay you money if you complete the, the, the course that they've done, and then they will take much more money if you do not.
And so there's kinda like, you know, it's, and so. Alignment. Yeah, that's really fascinating. Kind of
[00:25:29] Mohamad Ali: you, Jim.
[00:25:31] Henrik Werdelin: Exactly right. Totally. So it's just really fascinating, you know, how what are the business models that we have not quite predicted yet, will kind of emerge on the back of all these digital workers.
[00:25:42] Jeremy Utley: Yeah. Are there, just on that point, are there experiments that are in flight now in terms of business model? Like how do you think about. Experimenting your way towards the future when it comes to something like business model pricing, et cetera?
[00:25:55] Mohamad Ali: Yeah, actually we have, we have one client where we're trying something creative like this.
So, um, it's called a pod model. And so this pod is a, um, a pod can do a task, right? So let's say it can migrate, um, a cluster of applications to the cloud. This particular product people. We do a lot of that. We do a lot of cloud migration or VMware migration and that sort of thing, right? And, and these pods historically has been, you know, human pods.
There are 10 people. These are the skills of the people. You know, you can containerize, uh, so many applications that one particular pod, and that's what we sell. Now those pods are a combination of human plus digital labor. And so we've actually sat down with this client and we say, okay, well you know, you can have this pod and they'll do that, or you could have this pod and it's gonna do this.
And so now they're buying these human plus digital labor pods to do these particular things. So Henrik, it's, it's basically sort of your outcome space thing, but at a more granular level than a, you know, a large project, right? Like this procurement project I mentioned. And so I do think that, um, the large projects are gonna get atomized a little bit.
And I think to the extent that, that these like units, I'm not quite sure how small the units could be, but the units have to be able to, to, to combine human plus digital, right? You can't have one unit human and one unit digital. Like it doesn't work for this pod concept, but I think there's something there, right?
[00:27:29] Henrik Werdelin: Do you ever worry that one of the kind of observations I've had with just my own use of. Systems is that I now code a lot of things that I would normally buy. But you know, I, I would normally pay for a, um, a contact management system, but I only use like 10% of it if five, 2% of it probably. Right. And so now I could just basically vibe code myself into that specific feature, right?
And then I let go of the subscription of, of the system that could do the, the a hundred percent. And so, quote unquote, a worry. Slash like, uh, kind of an observation that I'm not sure that I know what the consequence will be as is if all this stuff is a little bit like the unbundling of the album. Like when suddenly we didn't buy albums, music albums anymore.
We, we bought like all the single thing. If I now buy the features, it seems that a lot of the money that came outta the music industry didn't go anywhere. Like it just kind of evaporated and so. Where are you on that we suddenly are creating all this technology and it's actually not kind of ACC created to any specific business.
It's kind of like, we'll just kind of evaporate out of the overall spend of software.
[00:28:45] Mohamad Ali: Yeah. You know, you know, another really good question because in some ways it's sort of like the burning question of what's the economics of all of this in the end, and you know, for me, I think they're, they're kind of two things that I look at.
One is that. Historically, like if you go back hundreds of years, like every time there has been a technological shift that lowers the cost of doing something, the GDP has expanded, right? Now the, you know, it's been difficult at times because certain jobs go away and certain new jobs are created and, um, and with this, there is a possibility that that's the case here too.
And let, let me kind of explain why I think that might be the case. So, in the services business, uh, many decades ago when offshoring became a thing, and so you could basically build the same application with cheaper labor, like the market didn't shrink, the market actually expanded. When cloud came into existence, and you guys know this well, you're a technologist, right?
It's like with cloud, you could build apps way faster because the hyperscaler had like all these prebuilt things and you know, you could load get stuck, right? And it turned out that the market didn't shrink. They were just like poor demand for more apps because like, we're always solving more and more problems.
And I think what's going, what's likely going to happen is that there's going to be more demand for like. Lots more of these sort of gen AI powered thing. Having said that, you were right. Like and so, so, so from that you could say, well, like, like a services company like us, there could be a lot more work, but if you don't lead in it right, you're gonna fall behind and you're getting a lot less work.
So this is, this is a part of why Yeah. I'm the, a paradox part in this. Right. Um. Now, now for the software side, right? It's like what happens to SaaS companies? Um, and you're starting to see a little bit of the multiple come down because people are kind of wondering is it gonna get disaggregated? 'cause you're right, you could vibe code a lot of good stuff now.
Right? And as you put these gen AI. Kind of user experiences on top of applications. Like you almost don't even know the applications are there anymore. Like you're not using the native user experience. And so they become, you know, sort of glorified databases. Now, companies who provide this, the software are not just gonna let that happen, right?
So I, I don't actually know Henrik chance here, but I see those dynamics.
[00:31:17] Henrik Werdelin: Yeah. Another, uh, just anecdotal thing that I looked up the other day when the ATMs kind of came into place, everybody was very worried about the people, right. Who worked in banks. Turned out more of them got hired over time. Same thing apparently happens with the self-serve kiosks in McDonald's.
[00:31:33] Mohamad Ali: Yeah.
[00:31:33] Henrik Werdelin: Like also I have now front and so. It is interesting that just if you look at all the data points of things that happened in the past that kind of like was similar-ish. Yeah. But I mean like obviously it's tough not to kind of go like, oh, I wonder how this gonna play out specifically with kind of like maybe even junior staff in a consultancy where you are like, learning how to do something in advanced model might be able to do a lot of the same work.
So now like a project manager will have to sit and go like, ah, you know. Do I save the X amount of thousand a year for this specific one and throw a digital work at it instead? Or, or do I just kinda roll with it with the benefit of having a human of the loop?
[00:32:09] Mohamad Ali: Yeah. And, and you know, those are like, a lot of what we've applied is due, is what I call low hanging fruit, like these documents centric things, but there's like a whole other class of work that is coming at us and coming at us like hard.
Right. So we're doing this project at L'Oreal, right? The cosmetics company. And this one's public. I could talk about it. What it is is L'Oreal makes, uh, 11,000 products, right? So if you think of a lipstick, and a lipstick is a little bit like that, steel fat that I mentioned earlier, right? You basically pour a bunch of ingredients and outcomes, something with physical characteristics, texture, taste, whatever.
And L'Oreal has, you know, decades of data of, you know, stuff you put into the pot and characteristics that come out in the product. So we're working with them to actually build an LLM from scratch. There aren't too many projects where we build an LM from scratch, but you can imagine like the engineers that, especially the junior engineers who get to work on a project like this, because at the end of it, you know, if you could sort of predict what goes into the pot.
And, and what comes out then you don't have to go build these things in the lab and test them. Right. Or at least you could routine. It's almost
[00:33:18] Jeremy Utley: like a al it's an alpha fold for cosmetics.
[00:33:21] Mohamad Ali: Yes. Right. And so your product cycle comes out and L'Oreal has a goal of moving to 95% bio source. Right? So, so they get two goals.
They get like shorter r and d, so they can outperform their competitors. And they have a model now that can help 'em get to more sustainable, uh, outcome, which is like still a big deal in Europe and you know, many parts of the world, right? So those are like, for sure, those are not projects that we would've naturally gotten before.
But now with this technology, I think there's a whole class of projects like that coming
[00:33:51] Jeremy Utley: that actually gets us something that I was hoping you touch on. So I appreciate you kinda opening the door. Broadly speaking, one way to think about the, the implications of gene AI are kind of efficiency gains, or you call it value capture and then new things to do or value creation, right?
And the, you know, the 3.5 figure you shared with us earlier in terms of reduced cost, et cetera, clearly a great value capture play. Tell us a little bit more about value creation, and this is a great case, right? Building an LLM is probably something you've never done for a client before, right? So that's a great, what are other ways you're thinking about creating new value and perhaps how are you organizing the organization not just to do the same set of activities with less human labor, but to imagine fundamentally new set of activities, maybe with more, even more labor because it's worthwhile.
How do you think about organizing the organization to do that?
[00:34:45] Mohamad Ali: Yeah. Uh, I think you're right. I mean, part of this is productivity benefits and there's actually I think, a ton of opportunity for productivity benefits. And as I think about this, I actually think by the end of the decade, almost every large company will actually have to run AD bp 10 and 20%.
More lower cost at the same revenue to be competitive because their competitors are gonna do that and we can see that that is possible today. Right? So, um, so that's one thing. The other part of this is that, um, I think some of that productivity gives you flexibility to invest in new things. Yes. And yes.
And, and, and this a great example right there, right? Which is faster r and d cycles. And you know, I mean, for. A lot of companies like RD cycles make or break the company, right? And if you could do something in 12 months that your competitor's gonna take 24 months, like game over. Right. Um, y you know, this, this government scenario that I talked about.
Yeah, sure. It's, you know, I guess you could say that that's productivity and that's lower cost. But the key metric there is actually the customer satisfaction. Customer satisfaction has gone way up, right? Because. You're not waiting for this thing forever and when you get it back, it's actually higher quality in this particular case.
Right? And so what does that level of customer satisfaction mean? So if you apply that to a company, that company is going to win more business. Right? And you're gonna have a growth. Just last night I was, um, in Chicago with a client and the client said, Hey look, you know, we do this, they call trade marketing.
You know, we have all this like docs. Whatever. And then when we put our product in in Walmart, you know, there's one set of trade marketing and we put it in, you know, Costco, it's another set of trade marketing and like, how do I go from this to that quicker? Because right now it takes me a long time and if I can get stuff on the shelf quicker.
Then, you know, my revenues are gonna go up. So I think there's gonna be a whole bunch of like revenue opportunity as well for companies. Yeah.
[00:36:50] Jeremy Utley: Yeah. Are there mechanisms like a hackathon for value creation strikes me that the hackathon's a really elegant structure for value capture. Do you do the same kind of thing to imagine new possibilities?
Or is it more just organic?
[00:37:04] Mohamad Ali: Yeah, so we use it in um, two places. At least. Two places, like in our product, uh, you know. Like, we have a whole product business, right? Like half the company's product, half the services. And in our product business, we are also using it to, to say, okay, well what kind of products should we create?
And I think that, you know, that's for a little bit of L'Oreal example, it's less about building an LM from scratch and saying, Hey, listen, if there's this big security problem, we all know, like cybersecurity is just getting worse and worse and worse, right? Like the, the bad guys are, you know, getting better and better, better.
And so there aren't enough. Human beings to actually keep up. Right? So you're gonna need a bunch of technologies to come to market that's going to be AI like inspired, right? And so that's a category of new products, right? And so, yeah, like we, we have been applying the hackathons to also like, how do we make our products better?
[00:38:00] Henrik Werdelin: I was wondering a little bit on the kind of like mixed wave of kind of like innovations you're gonna do in this space. So you, you've done the low hanging fruits. Yeah. When you look kind of 12 months out, what's the, the thing that, and, and you guys obviously are, are more advanced than most people, so like it is impressive just to hear kind of like how much impact the low hanging foods have had, but what's kinda like the, the thing that you get excited about 12 month out?
[00:38:29] Mohamad Ali: So, you know, if you guys, um, not that could do this, but if you listen to earnings call from the last quarter, our CFOI do
[00:38:37] Henrik Werdelin: that. I do that religiously on calls like IBM earnings calls on the calendar, never miss it. Just on his runs. He's
[00:38:43] Jeremy Utley: always listening
[00:38:43] Henrik Werdelin: to
[00:38:44] Mohamad Ali: em on your runs. There you go. Podcast. So he actually committed to raise that 3.5 billion to 4.5 billion by the end of next year.
So we, we actually see like we, the runway's not over. Wow. And as I mentioned, we had taken 200 processes of which only 70 we sort of apply this to. Um, as of the 70 that we're applying it to, there's sort of incremental gains that we can have and then we can apply it to new processes. You know, Jeremy, I think you mentioned hr.
HR is one where we, we apply this like at massive scale and, um, you, you might have caught in the document that we were able to. Our HR budget by 40% because we basically automated all what's called transactional hr. So if you needed to transfer an employee to Henrik, you can't figure out how to do it in your HR system.
So you call somebody right? Now we have this thing called Ask hr. You go and you chat with it and it helps you out. And if you wanna find your W2 form, right? You think about how hard that is, right? We all had to do it and you just go up this thing and it, you know, chats with you, make sure you are who you are, whatever, and you eventually get it.
And so 94% of all transactional HR has really been automated. And we've taken like some of those people and we've repurposed them. We've taken them out of HR and put them in consulting so that those consultants can now. Go into companies and help them with their kind of HR optimization. Right. So we've been able to,
[00:40:13] Jeremy Utley: wow.
Wow. Wait, so, okay. I've gotta, I've gotta read that back to you because that's such an important point. I think when most people hear, we've reduced the cost of HR by 40%. What they think is we've fired 40% of our HR staff, right. And I think what you just said is we've, and I don't mean to put words in your mouth, that's why I want to kind of put a fine point on it.
We've redeployed 40% of our HR staff from what was a cost center to revenue drivers of HR related consulting for our clients. Is that what you just said?
[00:40:45] Mohamad Ali: So at a macro level, yes. That is what we are doing. Right? So we are taking out costs in some areas, and we are investing heavily in other areas. I mentioned r and d and sales, right?
And in the HR example. We were able to redeploy some of those people into categories where they're now effectively going out and selling that playbook they created for hr. Right? And so there's a large supermarket chain that you might have gone to. It's got 400,000 employees where this is one of the projects we put 'em on, right?
Because they know how to do this right? And so that's incredible. Our company, we have that flexibility. Not everybody's gonna have that flexibility.
[00:41:24] Henrik Werdelin: I have one last question for you, at least for me. Um, we talk a lot about. Cost savings and optimizing. Yeah. Um, one of the areas that is always interesting is kind of like resourcefulness is when you don't just do something faster or you get somebody to kind of, uh, a digital worker to do it for you, but you actually become much better of what you do.
Do you have examples of that?
[00:41:47] Mohamad Ali: Um, you get much better.
[00:41:49] Henrik Werdelin: Do you know, like what I mean? Like it's not like, it's almost like you can produce the work in higher fidelity. Um, not just kinda like have somebody else kind of do it. It becomes a better thinking partner. It allows somebody who were not able to design in the past to then suddenly make graphics for the deck and stuff like that.
Like you can suddenly do more than, than just doing faster.
[00:42:09] Mohamad Ali: Yeah. Yeah. I think that is true in like a lot of our creative stuff. I mean, and, and you know, you, you think. Uh, started an ad agency, if I'm not mistaken, like one of you guys like invested in one of 'em, right?
[00:42:21] Henrik Werdelin: Yeah. So,
[00:42:22] Mohamad Ali: um, so I think the lot, like a lot of the creative, I mean we do a lot of creative stuff, right?
So there's a new airline being formed, it's called Riyadh Air, and it's not a startup that gets $5 million. It has $50 billion of funding. They've already placed enough. Orders for aircraft to be the size of united. So this is Saudi Arabia competing with Emirates and so forth, right? And so we were hired to build a whole IT stack and they didn't actually want us to start with traditional airline software.
'cause airline software is like train ticketing software from 50 years ago. It might be on mobile device, but it's like rigid and whatever. They wanted to start with e-commerce software, like an Amazon shopping cart, right? So, so we built that and that's really cool. The creatives are just incredibly impressive because they also, you know, this is like, this is the Middle East.
Everything has to be like incredibly like compelling, right? And so we've used like the best Adobe flash and all of that. And what we've been able to do there is just iterate through things so fast that they can effectively come up with stuff that is better than anything else, right? Mm-hmm. You know, your mobile app, for example, when you.
Go to it, you know, buy a ticket. You get to pick a seat, and it's like a pretty static thing with this mobile app. It'll be like a full immersive experience. You, you're gonna be able to see stuff and it's gonna be fast. Right. And we wouldn't have been able to iterate so many options on that and get it so well done if it wasn't for ai.
[00:43:54] Jeremy Utley: Okay. Mohammed, last question and then I know, uh, we should wrap. Two years ago you started this journey. Imagine another leader like you is seeking to undertake the journey, starting today. What are your kind of two pieces of advice to help them succeed?
[00:44:15] Mohamad Ali: This is not one of 'em. This is like the zero, right?
And the zero is you've gotta get all the technical stuff straight. Right? And you know, like we, we had to build that, that sort of AI virtualization platform. We had to build the, we have 3000 digital workers in different domains and all that, so you gotta get all the technical stuff. But I would say probably the most important is actually around the people.
Like you have got to figure out how to get people excited about this journey. And for us, you know, we encouraged everyone to learn how to use Gene ai, put it to work in their own, you know, areas. The approach here is it's like Excel, right? It's like, you know, today, like we all use Excel. If you don't know how to use Excel, like you're not just not valuing the company, but you're not valuable in the market.
Learn how to use this, be the best at it. Like you'll be valued with the company, you'll be valuable externally. And so, you know, people are in invested in it. And I would say that if you can't get. You. You know, I mean, I'm mostly focusing on businesses here, right? Because that's what we do to focus on business.
If you can get your employees excited that this is gonna make their lives better, they're gonna make the company a winning company. They're gonna be working for a winning company. You know? They're gonna be valuable out of the marketplace if they don't believe that they're just not gonna come along.
Right. Even if you have technical pieces. Right.
[00:45:40] Henrik Werdelin: That's incredible. Good place to end. Jeremy Avi, that was a fascinating conversation.
[00:45:47] Jeremy Utley: Uh, thoroughly enjoyable. What a winsome human being. I, I really liked getting to know Mohammed. You know, one thing just right off the top we talked about at the end, but it's so fascinating to think about a, perhaps a unique advantage of a consultancy, which I had never thought about before, is if they have internal experts, they can turn them from a cost center into a revenue driver, right?
Hmm. So, unlike say, uh, you know, a football club. If their HR professionals become world class at leveraging gene ai, it's unlikely that the football club is going to start advising other organizations and turn those HR professionals into revenue drivers.
[00:46:29] Henrik Werdelin: You know what? I've actually done a bit of research on this lately.
It's really cool. And I think it happens more than people realize. So if you look at like Delta Airline, what do they make a lot of their money on their credit card then that's, that's not about travel. Like obviously AWS is really a big revenue driver from, from Amazon. And so if you start to actually dig into the numbers for a lot of these companies, a lot of them have really managed to find a way of taking a part of the system, a capability that they were incredibly good at, and then monetizing that in order to serve their customers even better.
And it's obviously awkward because like when you create the narrative for the company, like do you go out and say, you know, we are Delta, we are pro people of travel and credit cards, right? Like you can't really, and so right there, I haven't really seen a good kind of catch all kind of descriptor of this.
Um, and I think you might see more of it with ai, right? To the point about hr, if you become so good of using something internally. There should be no reason why you don't kind of spin that out and, and kinda sell it to other people.
[00:47:31] Jeremy Utley: I, I, I couldn't agree with you more
[00:47:34] Henrik Werdelin: my friend.
[00:47:35] Jeremy Utley: I know you, um, the other thing that jumped out at me too was having the controller involved from the very beginning.
[00:47:43] Henrik Werdelin: Yeah.
[00:47:43] Jeremy Utley: Um, to big one documents and catalog the wins and put them on the board and to be rigorous about what they are advertising on earnings calls and things like that. Reporting their gains on a quarterly basis, I think is a really smart way to keep yourselves honest and also signal the transformation to the market.
Why would the controller not be involved in the weekly updates when you're driving that kind of a cost reduction, for example?
[00:48:14] Henrik Werdelin: I think the, the only thing I would add to the points that you made was. The importance and probably difficulties of seeing it as your senior manager's job to make the organization excited about ai.
I think a lot of people are very worried about AI and you know, obviously understandable why they might be that, but without creating the environment of where people come to work and going like, I can now learn a skill that really will make me not, not much better in my current job, but basically much better at any job.
Um, is is not an easy task and they really seem to mention, obviously Mohamed seemed to be such a charismatic person. You could see how we'd be able to kind like get like the hoo ha going, but uh, yeah. But that is probably something that we haven't really explored on these podcasts on. How do you become better of basically promoting that internally?
[00:49:09] Jeremy Utley: Feels like that could be a whole other conversation with Mohamed is just how do you get people excited?
[00:49:14] Henrik Werdelin: Yeah.
[00:49:14] Jeremy Utley: Um, I love the frame of IBM as the first customer, so to speak, or the first client is a client zero, I think is what they called it. Yeah. Um, and really propping up. I, he didn't put it this way, but the way I took his comments was they created an internal AI consultancy.
They treated themselves as the first client and they treated it like a consulting project, which is actually how they get bandwidth. And I couldn't help but wonder in a lot of organizations, if. Having a consulting engagement, even as a framework would be a really useful way to think about deploying experiments, because right now I think a lot of people are just moonlighting or it's bolted onto their existing job.
What does it look like to be an internal AI consultant to actually look for opportunities and to create proof points? I thought there again, it's maybe a unique competitive advantage of a consultancy that they can just deploy a consulting team. Towards the inside of the organization, but it feels like a model that anybody could leverage.
[00:50:17] Henrik Werdelin: Hmm. Yeah. And then potentially sell. Tell this.
[00:50:20] Jeremy Utley: We recently had Adam Brockman and uh, Andy Sacks on to talk about their book, AI First. Right. I was reading that on the flight this past week, and in the conclusion they talked to Ethan Mooch, former guest of our podcast as well. And one thing Ethan says to them in the end is, organizations need to build.
AI r and d labs, and it strikes me that that is what IBM did without calling it that they'd be the internal team that was working on behalf of IBM as client Zero was effectively an AI r and d lab.
[00:50:55] Henrik Werdelin: Yeah, and I, I mean like maybe just to, uh, end on that, when he talks about, and obviously he runs the consultant side, so.
Obviously, and, and, and with reason is very proud of what they've done there. And so I think kinda like corrected us when we said, you know, like that it was humble and ambitious to say we didn't get there first with the foundational model. So at least we should be there first with the applied ai. Um, and then obviously correct us 'cause like they do technology too.
And so, uh, you know, worth measuring. But it is very interesting. Two things. One is. Being able to realize what game you cannot win, and then kinda like find another game you can. Um, I obviously very, very inspiring. And the second thing is to see applied AI as a discipline. I think a lot of us, when we hear about AI technology, we think about the core models and like the, the bits and bumps that goes into that.
Really increasingly like the, the difference between what is technology and what is systems design is kind of like blending, right? Because increasingly, as you were saying, you need to be able to understand the system design and how you re architecture that in order to really use AI in a very effective manner.
And so these disciplines are kind of like also kind of, kind of merging to the point where, you know, applied AI is kind of a technology. Itself.
[00:52:24] Jeremy Utley: Mm. Skillset in itself. For sure. Yeah. It's been another great episode of Beyond the Prompt. Thanks for listening with us today. If you enjoyed this episode, if may, if we may save ourself.
Yeah. I mean, I loved it. Did you? I hope so. I, I, I really enjoyed hashtag hashtag, hashtag I loved it too. That's the code word for today. I loved it too. Just like Jeremy and Henrick did. Perfect. Love it. Hit like, hit, subscribe, share with a friend. Share with a colleague. And, uh, assure somebody who needs to get excited.
Hopefully this might be a point of leverage to get them excited. Until next time,
[00:53:00] Henrik Werdelin: take
[00:53:00] Jeremy Utley: care. No, no. Go Rik. You always say, bye-bye. Bye-bye. Okay.