AI Adoption: inside-out perspectives with Microsoft

AI alone won't revolutionise a business just because the tools are available. We explore how Microsoft transitioned from trial phases of AI to achieving tangible results—and why governance, training, and clear KPIs are crucial in moving from novelty to tangible outcomes. With insights from Marijke Schroos, General Manager, Microsoft Belgium and Luxembourg, and Michiel De Keyzer, a Director at PwC, we delve into daily adoption mechanics, from leader-driven prompt sharing to community building and weekly reports that push usage beyond 90%. 

Our discussion goes beyond buzzwords to reveal how teams truly capture value. We explore how to align AI with specific processes and challenges, establish baselines, and assess outcomes like cycle time, quality, and risk reduction. You'll discover how engineering advancements can be transformed into industry-specific solutions across finance, HR, legal, and procurement, and why AI is now a business-driven agenda with IT ensuring security, architecture, and data readiness. You can also take a glimpse into how local market feedback informs product development, while public sector constraints inspire practical innovations like fraud detection and service productivity. 

Trust and responsibility are at the core of every decision. This episode covers data classification, access controls, content policies, and review boards that mitigate bias and misuse—alongside training that empowers individuals to verify sources and question outputs. Education emerges as a key driver, with teachers and students using AI to enhance critical thinking rather than bypass it. Looking forward, agentic workflows will reshape work structures as teams learn to manage fleets of agents. With the addition of quantum computing, breakthroughs in fields like Pharma could accelerate significantly, displaying results rather than interfaces. 

If you're committed to AI adoption, this is your guide: establish boundaries, aim for impact, measure what counts, and combine human insight with machine efficiency.

Join us and listen to all episodes on www.pwc.be/aiunscripted

Transcript

Setting The Stage: AI At Microsoft

Patrick Boone 0:00

Ever thought how AI impacts a technology platform company? Well, today we're gonna address that in our AI unscripted series. And today I am here with the general manager of Microsoft, Marek Schroes, and I'm also here with my colleague Michael Michael de Keyser. Thanks for being here. I think uh the starting point. In previous sessions, we've talked about AI from more a user perspective or people that we're interested in. But of course, for Microsoft it's a completely different ballgame. It's your core or one of your core uh businesses. Yes. How is it impacting Microsoft?

Marijke Schroos 0:43

Well, of course, we're also users of the technology, um, but we're also bringing the technology to market. And um, it was actually um I became general manager after uh 18 years at Microsoft, about three years ago, in October, when actually the open AI can analysis were going. So I think it's uh it's the ideal time to be a country manager uh because so much is uh is impacted on the economy uh of also society. So it's quite exciting to be in this role and also helping helping companies and institutions to uh to get more uh out of technology.

Patrick Boone 1:18

Okay, and Michel, you're working closely with uh Microsoft because we're one of the partners.

Michiel De Keyzer 1:24

I think it's one of our main main partners, global alliance, but also on the Belgian level, we are working very closely together in bringing the technology to to our customers and then having an impact with it. So uh yes, it's a very good collaboration. I think we see more and more projects where together we are successful, uh I would say. Yeah.

Internal Adoption And Change Management

Patrick Boone 1:40

Yeah. Absolutely. Maybe there are indeed two aspects. Uh you've got the external thing and then your internal transformation. Let's focus first on the internal transformation because that's a challenge for all, also for us, privacy. How is that going? Because you have different profiles than we may have. Uh is that an easy ride?

Marijke Schroos 2:00

No, I think you never Ali never underestimate the change management, even in a company where the passion is very high for IT and tools and uh and what we bring to market, of course. It's still a lot of change management because you're used to doing things the way you're doing it, and there is so much now to um to learn and to unlearn. Um so it was not that easy. I thought as well, when we were introducing Copilot, oh, everyone is gonna go full in. It was not the case. Yeah, everyone was like testing it, but not taking it to the full advantage of the of the tool. So we um we actually set up a governance, and then um I had like weekly reports that I was sharing with all people managers on how the daily adoption of Copilot, for example, uh was going. And uh we saw that uh typical things like leader communication, making sure that you share prompts, community feeling, even in a company as Microsoft was very important to get the uptake to at least 90-95% daily usage. Some people take holidays. We see now that it's uh it's even becoming higher because also in private life, people are getting used to just use Copilot for uh for a lot of uh things they were not thinking about that much before.

Patrick Boone 3:14

Which is funny to see because we uh we're facing with the same balance.

Michiel De Keyzer 3:17

It's very similar, yeah. We we see the same at our clients, but also with us internally. It's not enough to just make the tools available. Uh, there are people that take it on board like naturally and start doing it, and these are also your champions, but but it requires an effort to get people using it, using it properly, and really having an impact with that. You need to hold their hands, you need to have training, you need to foresee coaching, sharing as well, community building. I think all those things we see we see at PwC as well, and also at our clients, yeah, at the client side. Yeah.

Marijke Schroos 3:46

I had a very interesting uh chart at Jay Perak, who is uh our worldwide EVP for uh for Core AI on adoption, is that there are actually two um types of persons at Microsoft. Um and actually there's like one one in the chart is going like this, like small increases, and those are the ones that are saying, wow, this is just so useful. We're using it for everything, and we're look at this prompt, and it's giving what I need, and then you've got an adoption like this, and also the increased value towards the businesses like this, and those are the ones that are challenging the tools and are actually saying, I would expect that it does this already, and uh, shouldn't we do better prompting? Shouldn't we do better integration of data or whatever to really get to the to the limit of what the tools can bring? And that is where the real drive is. So why you were saying, okay, there are some people very excited at a certain moment. It's okay then to have people as well challenging the system quite a lot because it's giving you a higher degree of adoption and also higher value towards your business.

Michiel De Keyzer 4:50

Yeah, and it's funny because with every technology implementation, you would always foresee traditionally a training program, a change management trajectory. And for some reason, sometimes with AI, this is forgotten because people assume that it's being used already by everyone in their personal life, for instance. But yeah, using it in their professional life is still something different, and it's also not the case that everybody uses it in their personal life. And I would even say that change management in the in the con in the context of AI is even more complex because you have the trust aspect. This knowing that AI has some biases, that there is uh this is a black box, so there is some sometimes mistrust. Um, and also it's changing so fast. So the the fact that it's it's not a stable solution, let's say. So it's changing very rapidly, and as such, that also comes with the challenges when it comes to change management. So it's even more important, I would say, uh, than with maybe traditional software implementations.

Patrick Boone 5:40

So yeah, we tend to um overestimate also the short-term impact and underestimate the long-term impact, which is of course an open door, but it is definitely also true in AI. Yes. But the problem that I see often is that it's incredibly difficult to quantify the impact. If it's now efficiency or yeah, top line, what is now really driven by AI. Is that something that you're also struggling with?

From Play To KPIs And ROI

Marijke Schroos 6:02

It's changing quite rapidly. I think the first excitement about AI was just using it, trying out, seeing, playing with it, um, and maybe not always to the business KPIs that you want to get out of it. I think part of the change management process is also to make sure that you start with, okay, this is a process. What would we like to what is the problem about this process? How can we increase the KPIs and how can AI help uh achieving those, with then very clear ruling on how you're gonna uh set it up? And as soon as you start thinking about which problems or which opportunities do we want to address and how AI can help with that, it's a totally different uh discussion. And then you see the true business value. And most of the cases that come out of these exercises, hackathons or in in any kind of format or form, um, when you when you do that, you see typically um uh either uh ROIs or um business value increasement or productivity increasement between 30 and 50 percent easily, and then it makes also much more sense for a business owner to start investing in uh in what AI can really bring to the business.

Patrick Boone 7:13

Okay. And what's then the role of a Belgian or a Baylux organization in the global uh Microsoft uh mastodont? Because of course it's uh it's a mega organization.

Marijke Schroos 7:26

Yes, um, I think a very important role. So of course we are representatives of Microsoft to the market, but also representative of the market towards Microsoft. Um, so meaning that we translate whatever the engineering teams are developing into okay, how can we serve the customer needs? But if it's not enough or there are improvements to be made, that we can also funnel it back to engineering so that it's an uh it's a loop. Um and I think being very close to the market, very close to the customers in all kinds of forms, because we've got the very small customers and the very big ones, we've got public sector and we've got the commercial uh companies, very regulated industries, and more global-oriented commercial industries to make sure that we capture everything what they need and get it back to the uh the mothership, is I think very important.

Patrick Boone 8:15

Because how many people do we have in Belgium?

Marijke Schroos 8:17

In Belgium we are 550 in total, about 350 are really addressing the Belgian market. The rest is more like uh what we call uh LGEs, uh so the regional guests, uh like for example Philippe Rogge, who is actually the worldwide public sector lead, he just uh came back to Microsoft, but he lives in Belgium, so he's is he's noted in Belgium, but not really addressing only the Belgian market, of course.

Patrick Boone 8:42

Okay. And ideas like alliances, because again, we are one of the alliance partners. That's really core of the strategy, I guess.

Belgium’s Role And Partner Ecosystem

Marijke Schroos 8:51

Yes, that's always been the case. Microsoft was from day one a partner-oriented organization. And I think even more now with AI, uh, we want to partner with uh with others. I think for uh for companies like PwC, the value of working together with Microsoft has only increased because, of course, with AI you touch every single business aspect as well. So you have a lot of contacts with the business parties, in and we used to be more at the IT side, but with AI, of course, it all comes together, it's business decision making um in all kinds of uh you know forms, and I think that also increased our partnership quite a bit.

Michiel De Keyzer 9:28

That's the transformative part, indeed. Yeah, so I think uh the complementarity is that we know the industry declined sometimes very well, the processes that they run within certain functions, being finance or HR or whatever. And so, yeah, to make to make sure that the technology has an impact, you need to make it land within the organization, within the processes, and that's I think where we can help. We can look at the opportunities within the processes, but then the products or the the technology makes it come to life basically. Uh, and and so that is one plus one is three. So very good collaboration, and we have some very nice examples already, both in public and in private sector. So where we've managed to do that.

Patrick Boone 10:04

So yeah, who's then typically the the audience for starts of discussions within companies?

Michiel De Keyzer 10:10

It's uh it's all kinds of audiences, I would say. Well, that's how do you start that discussion? That's yes, but what surprised me maybe a bit personally is uh so I've been I've been uh driving a little bit the AI journey uh for PwC and also towards our clients uh from let's say the launch of ChatGPT where everything has started. Uh and what we said, what surprised me a little bit is that it's not at all an IT discussion. Uh the workshops that we've had are uh largely with CFOs and their teams, CHROs and their teams, procurement teams, legal teams, and so on. So all business teams. In a lot of cases, uh the CIO or IT is in the room, but they're not saying that much. It's really a business-driven discussion. Um, and and that's where it starts. And and and what has also uh what is also uh remarkable for me throughout that journey is that uh the it has never been that easy to brainstorm about where we can use AI because people are using it in their personal lives, they have played around, it's so easy to start using it and to try things out that people already copied IDs to the to the to the table. Not always IDs that that you would say in the end we will solve it with AI, but that's and that's also for Pwc. That's not really the end game. I mean, we are there for the transformation, and whatever ID that can help transforming, let's say, the way you are doing things, whether it's with AI or without AI, is a good idea, I would say. And and and and that is yeah, that is uh a nice, let's let's say, uh, discussions to have with the client. So and that's the starting point, but then yeah, the difficult thing is then of course, how do we bring it to the next level? How do you start making it tangible, building stuff, implementing it well with everything that comes with that, like change management and so on? So that's I think still also where the market is a bit maturing, in my opinion, uh, but where steps are being taken and more and more it becomes concrete.

Patrick Boone 11:48

Which is also linking to the question as to why are they doing it? Because yeah, you're not just using copilot, because the others are doing it. There needs to be a purpose, I would say. Yes. Otherwise, it remains at the playing uh stage. Yes. And do you see a maturity increase there?

Marijke Schroos 12:02

Yeah, absolutely. You see very, very good uh usage, not just on copilots and not just on M365 copilots. We've got many co-pilots, of course, but we see it in all kinds of areas, like uh with a developer skit up co-pilot, it's like going very uh very rapidly in the uh in the adoption. Security co-pilots is typically also something that uh works really well, or their own developed um generative AI solutions based on the processes and the solutions they have in place. But I think it's crucial now, uh because you were referring to the fact that um most of the employees are already using it as a consumer. It's really important that companies are looking into it because if they start to use it as a consumer with uh with the business data, uh without taking into account all of the things that also the guardrails that you need to do to to be set up, then it's I think a very vulnerable uh uh situation you're in as a company.

Business-Led AI, Not Just IT

Patrick Boone 12:58

Which is also very explicit on in your mission statement as the ethical use of AI, and that's indeed true. And without even addressing the environmental uh aspects of that act. Um is that something that you see also within the organization uh of Microsoft that people are starting to say, mm, how far do you want to go?

Marijke Schroos 13:20

Uh I think I we we are you for us for our own usage and of course what we bring to markets, we have uh defined very good governance and also the guardrails. So we have a lot of uh principles in which we we navigate. Uh of course, like we and there are few things like data classification or documents. I cannot just send it to you. I really have to go through uh quite some steps because otherwise it's it's secured. So I cannot because because it's important, of course, if you're using the data in AI models that you you're 100% sure that though that that that data can be can be shared. But also on the setup of the models, there is a lot in there that we define, like uh it cannot be uh harassment, it cannot be etc. But it's of course we can do it already uh as a base in the systems, but it's still the customer that needs to make decisions there. Um because if you say, okay, we don't want to have sexual abuse, for example, or any language in there, but you're working for the police specifically on those cases, then you need to turn it down as well, and you need to be able to say, okay, no, those are explicitly the ones that we want to see popping up. So it is important as a company to really think about it. Okay, what do we want to get out of it? What are the rulings we are gonna use? Um, and then use it very responsibly. Also to avoid bias, for example, if a developer, if it's only one developer who is developing something at Microsoft, it has to go to a specific through a specific board to see whether there's no bias in there. Um, so there are many, many different things, and we are very vocal also towards all of our employees on what we're doing there and how it works. We've got a lot of training as well to make sure that they understand very well. So, on those areas, I don't think there is much doubt that we're doing the right thing, and we are set we're setting up the tools to do the right thing as well.

Michiel De Keyzer 15:10

And also, that is not only a tooling, uh tooling questionnaire. I think it's about education of people, finding guardrails, uh yeah, as as part of building trust in how you use it, how how how how you can trust the outcome, but also yeah, I mean trustworthy and and the responsible user of the AI. I think it's also something where where a PDBC can help to define those guardrails to help people educate on what what you can or cannot do or shouldn't or should not do, define the policies but train the people in the and also how reliable the thing is that yeah comes out of the machine. Indeed, indeed. What what questions you should ask yourself indeed if you get a certain outcome on bias, for instance, uh and and have a critical mindset looking at it by by looking at it and need to uh checking some of the sources as well.

Responsible AI, Guardrails, And Trust

Patrick Boone 15:52

Yeah if things are not uh because there are people uh advocating that we need to tone it down a bit, and it doesn't have to be less high performing as a as a as an approach, but that it basically should guide, definitely in the educational system, it should guide the people rather than providing the end deliverable, which of course is a question if you are 15, 16, will you accept that it's basically suboptimal and it doesn't provide you an end product and only prompts and what have you? So I think yeah, the truth is probably in the middle.

Marijke Schroos 16:20

I think so, but it's uh you could you could easily set that up uh that way that the responses would come to to really stimulate to the the the the critical consciousness. And in a certain course or at a certain age, it might be very interesting to have this as well. That's also generative AI, to use it to train their critical consciousness as well and to see whether they are taking uh the right decisions. And I think it's really important if you want to get the best out of the new technology that we are very conscious about it and that we make sure that our children are also uh educated in a way that they can handle it and can only use it for the good and not for the bad.

Patrick Boone 16:58

And other links with the educational system.

Marijke Schroos 17:01

Yes, we do a lot with education, a lot, and uh I also have a lot of discussions uh with the education field uh because I think it's important, it's new for everyone. So we need to make sure that we provide whatever we can provide them with to uh to make the right decisions, and they are well uh well, they are the educators, of course, they know what is good for the children as well. So we need to go hand in hand to say okay, what is possible and what what is needed and bring that together.

Patrick Boone 17:30

Instead of pushing it back.

Marijke Schroos 17:31

Yes.

Michiel De Keyzer 17:31

Which is uh and there's a lot of work there, I think, in our Belgian education system. My wife is a teacher, so uh there's a lot of work to also uh teach the teachers, let's say, so they can then teach the students and and true show that it's something that you should embrace, not fight, but embrace it in the in a in a responsible way, let's say, and I think uh yeah, there's a lot of work uh there to be done.

Patrick Boone 17:51

So but we're not on I think you say in Belgium, but we're not unique, I think.

Michiel De Keyzer 17:55

No, uh exactly not probably it's not a unique problem for Belgium alone, yeah.

Patrick Boone 17:59

Indeed. So what's next, you think? We have we talked about in previous sessions on agents. But what what's uh yeah, difficult to predict? You don't have a crystal ball, but in two, three years' time.

Marijke Schroos 18:09

I think the scale and the speed will only go up. I think each and every uh process will be fully agentic, so meaning that you have your agents in your organization. So I think people need to be able to prompt, but also need to manage agents and have a very good AI governance in place. And then I think the way we work will be totally different from what we do right now. All of the repetitive tasks, all of the things we can outsource will be we will have agents as well next to our normal colleagues. If you then add quantum to it, you uh you see that um the power of AI will only be will increase quite a bit. And if you then look into um applications in pharma, for example, to find new medication, uh, what is impossible now because there are too many components available and it's too hard to find the right components. Thanks to AI generative AI in combination with quantum, it will be very easy to get the right the right components and bring them up. So I'm I'm a positivist and I'm a very big believer in technology and when I'm really sure that it can help solve quite a few of the world problems. Uh I really believe so. So I think yeah, the combination of the quantum and the AI that we and then the right usage, of course, making sure that we educate well and make sure that we keep ourselves uh accountable for uh for uh for you for using it for the good, I think that's really uh the next but contrary to AI, quantum will in fact not be visible to the user, uh no, it will only be visible in the results.

Patrick Boone 19:42

In the results, uh so and is that something that we are also looking into?

Michiel De Keyzer 19:46

We're definitely looking into it from different perspectives. Uh uh for instance for also quite post-quantum security, these kind of things if we prepare organizations to be secure, that we already take that into account. So these are things that we are already looking for. Into yes, yes, definitely, yes.

Patrick Boone 20:01

Maybe a question on because you talked about private public sectors. Uh yeah, we had difficult discussions in terms of budget uh recently. Um are we lagging behind uh in the public sector or are we contrary basically front running? So my question is who is going faster than the other one, or is it difficult to compare?

Education, Skills, And Critical Thinking

Marijke Schroos 20:25

It's it's always a bit difficult to compare, of course. Um but I think Ali I'm quite proud on how Belgium is doing. Uh of course, Belgium is also a public sector. Belgium doesn't really exist. You've got many different organizations and they have different speed as well. Um but if you look what the Flemish government is doing, but also how police is looking into it, uh, how justice is doing. So I I think defense there are many, many, many great examples where we are really front-running uh on the on a lot of other countries. So I'm happy that we are embracing it that much and that we are looking into what the possibilities it brings. And budget constraints can be a very good one to really look into how uh how AI can help, how can we look into budget spend, take better decisions there? But how can we create more budgets by maybe fraud detection, these type of things, or how can we increase the productivity of people? I think there are so many themes coming out of that challenge that it becomes an opportunity also to address with technology.

Patrick Boone 21:26

Yeah, because you talked already about return on investment, but that's an effect also what we are doing in a number of cases, building a business case for change because it's not just a gimmick. No, no one is using AI, we need to do the same.

Michiel De Keyzer 21:37

There needs to be something in return. Yeah, and that's something that we try to bring into the equation as from the start, as from the initial brainstorming. It is not not a showstopper if you don't have a full view on the ROI as from the start, but at least it's something that you need to think about uh as from the start. I think the important thing uh in in that equation is that uh yeah, that you also need to know what impact you want to have. So it's not only about uh knowing where the impact is, but maybe you should think in the beginning where do we want to have impact and find the impact that you want to have, plant a flag, there to plant the flag and work towards that flag. Maybe you don't get there, but okay, at least you will have had an impact then and then you also are sure that you move in the right direction. There might be pleasant or unpleasant surprises along the way, but I think that's an important one.

Patrick Boone 22:17

Yeah, great. Maybe a bit of a rapid fire. So basically it's short questions, you don't have to uh think too long and yeah, one word, one sentence max.

Marijke Schroos 22:27

One word or one sentence.

Patrick Boone 22:28

Yeah, you can choose whatever you want. The first AI app that you use in it on uh every day.

Marijke Schroos 22:35

Um that would be my my M365 chat on mobile. So when I when I'm I'm like getting awake, I already ask what did I miss. Uh it's a spromp that I always use. What did I miss over the last 12 hours? Because of course we are a US-based company as well, so a lot is happening there. And then it looks into my team's chats, my calendar, my email, uh, just to see okay, are there things that you need to follow up rapidly? Uh that's the first one I I use. Okay.

Patrick Boone 23:04

A bit of publicity, that's okay.

Marijke Schroos 23:06

Sorry, but yeah, it is the first one. Yeah, but it is, it is what it is.

What’s Next: Agents And Quantum

Michiel De Keyzer 23:09

It's the same for me. I mean, but uh more more for forward looking, what is on the agenda for today, uh more in that aspect, but indeed it's it's exactly the same for me. Uh 265 copilot term.

Patrick Boone 23:21

The biggest mistake that you made using AI so far.

Marijke Schroos 23:25

Using AI so far, um, I think from a company perspective, I'm I I the change management, but also for myself. Um making sure because it goes fast and the possibilities are endless, uh, to make sure that you spend enough time to reflect and to say, no, no, no, I'm not gonna do it that way, I'm gonna do it this way. I'm just gonna ask it. And I became a very good prompt uh engineer, really good. We had some competition in uh in uh Europe, Europe North, and I was actually the winner because I'm getting very if you are if you're training yourself and you're really saying, no, no, I'm just gonna reflect, what do I really want? To which or for which audience, what do I then you get the results are just uh amazing. So I think that was at the beginning a little bit, yeah, the fast, the fast the furious, you know, like quickly, quickly do it and then and then get some results out of it and already be like, Wow, what are results? But if you really take it uh at heart, I think it gives a big difference.

Michiel De Keyzer 24:23

For me, it's probably showing it to my kids. My biggest mistake, the power of ChatGPT, showing it to my kids, because they asked for my phone on a daily basis to play around with ChatGPT or to create a picture. Uh so that's that's the thing without you. They probably would have done it without me as well.

Patrick Boone 24:38

Uh but okay, I still blame myself. Maybe a last question I have is um you're also heavily involved in diversity and inclusion initiatives. Um and of course, the haves, have nots, the yeah, the digital divides can only accelerate, I'm afraid, with AI. Is that a concern that you have?

Marijke Schroos 25:00

I don't think it's gonna broaden. I think it can be much more narrow because on the if you look at inclusion, the fact that you have AI also includes a lot of um groups that maybe don't have access now to uh technology because you can do a lot with voice, you can do a lot with so there are many uh new scenarios um to audiences that maybe were a little bit left behind uh before. I do believe that we need to um make sure that the digital skilling is going up in uh in Belgium, so making sure that everyone knows, but the the the entry to the technology becomes a lot easier. So I think it's gonna close the divides more than it will be important.

Patrick Boone 25:40

You're an optimist, that's good. Yeah, I agree. I need to be.

Michiel De Keyzer 25:42

Um it's our obligation.

Marijke Schroos 25:44

Yes. But I really think.

Michiel De Keyzer 25:49

The information, all the underlying information can be. We did uh we did a small uh book with the Flemish government on parliamentary information. It's all in Dutch, but you could ask the questioning Ukraine or Arabic or whatever and still get an answer based on the Dutch information of what has been discussed in the parliament. So the barrier of languages becomes much slower, and that is, I think, a very important aspect or or let's say enabler for inclusion. So yeah, I'm also an optimist. Uh in other words.

Public Sector Pace And Business Cases

Patrick Boone 26:17

Okay, with that and with these wise words, I would again like to thank you. Oh you're welcome. And uh also thanks for tuning in, of course, uh to yet uh another episode of uh AI Unscripted. And as you have seen, it doesn't really matter if you are now uh focusing on AI as your uh core or one of your core businesses or not. The problems are basically not problems, they are challenges. They are not really challenges, they are opportunities, and it's that that we basically need to uh focus on. Uh and yes, uh we will also focus on that in the next uh set of episodes. So tune in to the next uh episodes of AI Unscripted. Thank you, and bye bye.