#579 AI Meets ERP Transformation: Dominik Wittenbeck on the Future of SAP Data Migration

Mehmet: [00:00:00] Hello and welcome back to a episode of the CTO Show at Mehmet today. I'm very pleased joining me, Dominik Wittenbeck. He is the group, CTO, , at SNP group. Dominik, he is a veteran in, in technology. You know, he, we've done, uh, a lot of projects around digital transformation, but mainly, you know, we're gonna talk today about ERP and.

As you know, I'm telling the ga, the, the, the audience. There's a lot of vendors in this space, but you know, Dominik is specialized in SAP mainly. So this way we're gonna talk about SAP. Some people they call it sap. So just for people to know. Without further ado, you know, I don't like to waste much time from my guest, uh, Dominik.

So tell us more about you, your background, your journey, and what you're currently up to, and then we can start the conversation from there. So the floor is yours. 

Dominik: Perfectly thank, uh, thank you Mame for, uh, having me in the show. Yeah. So what do I do? Who I am? Um, [00:01:00] well, I am, as you said, in the SAP space, um, specifically in data migration, to be specific.

So the company that I'm working for, SNP with like almost 30 years, uh, of experience in SAP centric data migrations and, uh, what does this leave me with or what does this mean for what I do basically. I'm basically in the field of, while selective data migration to be specific data migration. The SAP space comes well with a couple of different twists and turns and focus areas basically.

So right now, many of you know, probably know, probably know that, um, SAP has put out S four for, well, almost a decade now, even beyond. And, uh, many of the customers. Are on the journey to migrating to S four hana, but that's not like the only thing that SAP centric data migrations typically are, uh, about. I mean, any type of modernization, um, of system landscape that you might come across, uh, is falling into this area as well as, of course, mergers and acquisitions, restructurings and all these sorts of things [00:02:00] basically fall under this category.

So. What I'm, uh, what am I doing specifically in this area? So I am, um, inside of r and d in our organization. And while you might think that data migration is probably something which is very project specific and therefore would probably be classically a services business, whereas s and p have been focusing on providing a platform for facilitating these types of data migrations basically.

So my goal for the last, I'd say. And my business in this, in this particular case, 15, 20 years or so, is probably like making sure that the experts that are out there in the field that are either with, uh, my company, uh, or with the co uh, with, uh, affiliated companies such as like the biggest iso. So, um, they have the most, the, the best tools basically available.

Um, the best sources of information available, the best methodology available, uh. If possible, all packed into software to make their [00:03:00] lives easier, because at the end of the day. People in data migration consultants, data migration, they are, I wouldn't say rare, but there are not so many of them. Right? And the pressure now nowadays in, in, in markets really increase, not just for the modernization piece, but if you think about like, Hey, you need to move outta specific region very quickly because there's conflict happening in this region and you need to do carve out or restructuring.

This puts immense pressure on organizations to be quick on their feet. To execute these types of data transformation and make them sure, or make them realize basically in, in their system. And yeah, so I'm basically with me and my guys trying to build tools, uh, and methodology to, to help the people in the field executing these types of projects to be well as quickly and as reliable as possible.

Mehmet: Great. And thank you again, Dominique, for being here with me today. I'm gonna start, you know, from where you kind of ended, and this is a curiosity kind of question, why you mentioned that [00:04:00] experts are rare. Why? Like, is it not that attractive? Is it because something hard to, to, to learn? Like, what's the reason for the scarcity?

I would say in the field, in this, in in, in this domain. 

Dominik: I don't know about the attractiveness of the situation. Uh, probably, I mean, you don't know what you're getting yourself into when you start entering the job, but to be honest with like the seniority degree of people that I see, and many people that I'm working with are 15 years in the business, 20 years in the business.

I don't think it's like people are leaving it. But it's very hard to get knowledgeable in this particular field because it's a super lot of factors being involved in data migration. It's not only the technical part around this, and that alone probably is good for like two or three roles to fill, but it's also the organizational part.

Change management is a big, big topic that's always involved. So people who are really senior in the field and know what they're doing and what the nuances are that they need to like listen to. React to. [00:05:00] They are rather, well, they're just not enough for the amount of work that's basically there. As I already said, like the, the, the change rate that's happening, right due to external factors, economy and the regional conflicts and so on, so forth.

It just increases. So that puts even more pressure on the few people that are in this particular industry and business. 

Mehmet: Great. Now let's talk about, I would say the real problem here, right? Or like what, what organizations, um, face, right. So I, I was working in, in, in IT department long time ago, in previous lives, as they say.

And the word migration used to scare us, right? Whatever migration, it's like, you know, sometime even you do migration from a, you know, on, on the infrastructure level from one platform to another. Uh, I've seen like teammates who are doing migration on the database level, so [00:06:00] from one platform to another platform.

So data migration, usually it's, it's a complex project by itself, right? Um, so. And today we have also the, the, the theme of AI also as well, right? Yeah. So what makes data migration different from, from many other, like other AI use cases? I would say. 

Dominik: So, lemme try to answer your first part of the question first.

Sure. So why is, why, why is data migration special? Well, typically data migration is a result of change that's happening in the organization. And typically I know about you, but for most people, including myself, change at the very first, main, uh, moment is scary because it's different from what you already know.

So per se, data migration as being in the midst of this and. Basically a result out of, hey, we need to migrate to set to a new platform. We need to introduce a new tool, or we are reviving our [00:07:00] processes or streamlining our processes. For many people, that means that their jobs are going to change and at the beginning they don't even know whether that's a good change or a bad change.

Um, many people are embracing change, specifically the ones which are overload. Per se, right? They hope for better when streamlining and, and, and well getting stuff rid of stuff basically. Not all are, but at the, at the core of it, data migration always is something that is following an organizational change, making that really at the heart reality of it.

Because before the actual data migration, meaning before you actually did adaptions in the systems that you work in and on, day in and day out. This thing might be theoretical, you might still be thinking like, yeah, well this may pass by me. Right. I might, I might get by without even worrying. Um, but in the end, data migration is what makes this real.

On the systems. You are moving your customers from A to B, you are changing the processes to align to the new chart of [00:08:00] accounts that you supposed to have to align to a corporate structure or something like that. That of course, is something that will then. Really affect your job. If the data, if not for data migration, many things will not be as like present, which is also why, well, people sometimes overlook it a little bit as being something which is at the end of the spectrum, but really what data migrations typically surface is like many of the.

Misconceptions that you made before many of the, um, well, um, corpses that you buried in your virtual IT backyard, um, basically all sorts of data, uh, ambiguities, and that data quality comes up all over a sudden and you have to treat it. Now that you're moving up to the, to the other system, you can't avoid it anymore.

That makes migration actually more than anything else, team sports, which is going across the whole organization typically. So it's not only the technology team that has to work with it of the customer to actually make the data [00:09:00] move and the justice to the needs of the respective program, but it is also alignment with the business much more than anything else because in the end.

The business, uh, are the ones who have to execute their daily work on the future system. So what is one of the common problems is misalignment specifically between the IT part of the equation and the business part of, of the equation. So what I see many times, and this is where AI actually comes in for like a first use case in data migration, is.

How this typically starts is you get an RFP as a vendor, right? Which is basically like, Hey, this is what we wanna do, this is what we wanna change. It typically hasn't too much of detail basically in it, but it's supposed for you to give some sort of, uh, well, uh, rom basically in, in the end, right? Um, so some type of effort that goes towards this, this specific, um, um, project that is about to happen basically.

What you then [00:10:00] normally do, if you can, you speak back to the customer, clarify ambiguities. You clarify things that are not explicitly written down in the RFP, and that typically happens at the beginning in the pre-sales phase, right? When you're trying to, to sell the project, whatever type of business you have, whether you complete service, business, or software, or a mixer of anything.

What I saw happening most basically is that you clarified all these things with a customer upfront and then the project team comes in basically. They are not too much attached to presales. I mean, in a good organization they are somewhat attached to presales, but maybe they're not, right. And then they start asking the customer the same questions over and over again.

The customer was like, Hey, didn't I just tell you that? Right. So it's really like a lack of process really, where I think AI can first and foremost help because you what, what I like to do when going into such workshops typically is. I like to go ahead and say like, okay, this is like the overall picture of things.

What are the, these [00:11:00] are the questions that I would typically ask in the situation. Is there something that I'm missing? So that's the first checkpoint. And here, to be honest, I mean, typically you would like roll this around a challenge with colleagues, right? That have seen a similar experience. But what if you're just starting out?

What if you don't know anybody in the, in the organization so far, you still have the task for getting all this information out of the customer. So then you might as well use AI to by get, make sure that you have some level of completion in regards to the questionnaire that you're actually fighting next, I mean.

You run this questionnaire and the customer will most likely give you much more insights than your questions, unless all your questions are yes, no type of closed questions. They will give you much more context basically than you initially asked for. When you have to comprehend this context, you have to write it somewhere down.

So. Next case of AI using the transcript and bringing it basically into something which is really a specification of the, of the, of the project that is supposed to be happening here. So really at that [00:12:00] point, AI is really helping you to prep for the workshops, but it's only helping you to consolidate basically in the end and create something which I'd say a specification is a communication artifact with the rest of the organization.

Because once you align on it. You can always say like, Hey, this is what we stated. This is how we said it's going to go down. It's not that it's unchangeable, right? But it's basically a, an a manifestation of an agreement that you have amongst all the parties, which is why being completely and being precise here is something which has been super important.

But AI really helps you to like even find the little gaps that you had before and close them also. 

Mehmet: If you allow me, Dominik, I would say my 2 cents. As someone who work in presales again in, in previous lives, um, I think whatever the project is, you know, the moment you start to have the confidence that you're gonna be part of this project, I think the sale, I, I'm talking about presales and [00:13:00] sales teams, you should directly align with your, you know, implementation team because.

I saw it many times and unfortunately, yeah, like what happens is like these guys, they say, okay, we're gonna just need from sales perspective, and I can understand this, just close the deal. And you know, by the time it, it closes, like it's not my job anymore. Let the implementation team happen. So the way, and I think AI, as you mentioned, can help a lot here, like at least, but back in the days when there was no AI and note taking and all this, so.

I tried as much as I can to put all the information as on behalf of that team. So the ultimate scenario would be like you have someone from the implementation team during the presales phase also as well, right? So these guys, they are up to speed. You don't need to go and, you know, repeat the things now.

Thanks, you know, to the technology advancement and now note taking. I think it can. Remove some of the burden there. So mainly to [00:14:00] get the things, but yeah, like the best thing that can happen is when, uh, you know, like, uh, you, you, you have this complete alignment between the implementation team and the pre-sales team.

And I'm saying this from a practitioner perspective, someone who've been, uh, on, on, on, uh, on, on that table now. If we want to jump on, uh, you know, the migration itself, so you focus a lot on selective data transition, SDT, what is it and why do companies choose it? 

Dominik: Well, selective data transition, and sometimes you might have heard the term blue field, um, what you traditionally intuitively do when you are introducing a new system and a new or, or, or change or something like that.

There. There is a reflex in organizations to do something that we typically call greenfield. We start a new, we get rid of all the old stuff, basically. But once you really start it, [00:15:00] you recognize that, oh, I have to like now understand the context of what I already did in the past systems, because maybe you have legal obligations to keep specific data around for a certain point in time.

Um, you probably have invested a lot in the past system unless, I mean, you're a startup, you're completely starting a new, I mean, then you're, you're Greenfield, but if you are a established company, you probably have a system of record. You, you p system or other, other, other system. There is, and you have invested a lot into, into this system.

You have probably been been tuning it. You have probably been. Implementing your specific processes in that, that actually also like, will separate you from your competition in the market. So you wanna retain this type of intellectual property that you put into the system. Um, and if you put these things basically together to retain what you already have to, well, we always say keep the best, transform the rest to embark to the new system, to embark to the new processes, but keep what makes you special as a [00:16:00] company, what makes you unique and what does.

Build your success basically in the past, then you are mostly up for a game where you need to keep some existing configuration. You need to migrate historical data selectively for most part, because you wanna get rid of some things. I mean, you wanna clean up stuff. At this point, um, and this is where you actually do a selective data transition, where you move data that is relevant for you and your business into the new system, and you transform it along the way so it complies to the new processes and the new wealth setup of the new environment basically.

So. As I already briefly mentioned, I mean this is true for modernization cases. Very, very, very, um, obviously, right? You, you go from EC to S four. S four is is a different thing. Uh, in the meantime it is. Uh, but you want to keep basically your, your investments. But also if you have a company and you, you do an m and a for instance, you also wanna like the completely destroy their IT infrastructure.

You wanna [00:17:00] take what's best, right? But put some governance on top of that, that is corporate at this point. And selective data transition is exactly where you can find this middle ground where you need between going almost the complete way green and only take very little selective history to almost going completely brown or conversion, and basically take almost everything that you had and change only little bits and pieces.

So what SDT or Bluefield really gives you is the flexibility to stretch between these two parts. Seamlessly to, I always say like what your organization is able to digest because change is something that's happening and not all the companies are super good in adopting change. Depending on how centralized your structure is, how liberate your structure is or regionally, regionally organized your structure is that really like depend makes it really like different for how change ready your organization is overall.

And STT gives you basically a lever which you can put from one side to the [00:18:00] other to perfectly match your organizational needs with what you actually can do as 

Mehmet: Dominik what can, can go wrong, you know, uh, if, if data migration is underestimated and, you know, walk me through also maybe some, some of the risks as well that you have seen from, from your experience.

Dominik: Yeah. Well, I mean, data migration is, is mostly on the radar to be honest. Um, most of the people initially at least go in with, Hey, that's an afterthought where once you realize that actually this is moving your core ip, this is adapting to new processes. That is where, where things are basically changing.

But what can happen if this goes wrong and if it's underestimated, I mean, very easily. Timeline stretches, right? If you are under-estimating the data migration pieces, and then during data migration, all these ambiguities and the data quality lacks that are basically happening there, um, they are surfacing.

And then these data migration guys, they start asking nasty questions, [00:19:00] Hey. We found this particular thing in here. So I don't know, let's, let's take an example, which I'm currently running through with, uh, a data migration that I'm running at this point. Like we are migrating business partners and I see like, Hey, I have to migrate tax numbers.

Now, the tax numbers in the old system. They didn't have any check algorithms basically to it. So it's like completely all over the place. The new system is enforcing this check algorithms now, basically now, uh, in the data migration, go to the business. Say like, okay, hold on. We have a problem. I'm not get, I'm not able to get the records into the target system because they are not complying to the data integrity rules that are basically there.

Do you have option A, B, or C, how to deal with this? Now, data migration has a tendency of surfacing these types of insufficiencies that are buried in the old system and making them obvious. And if you underestimate this type of thing, and this typically means that the timeline stretches you effectively.

Likely then to move the go live because of all the decisions that you didn't know you need to take and need to like align with all the other [00:20:00] guys inside of your corporation until the decisions are finally made. And if you don't do, and one of the things that might be a shortcut is specifically if you're going towards a database.

Only migration. Well just let's take data over as is and just let's eliminate all this validation rules. They will pop up later in your proce in the processes, right? But only, uh, across time. So another result, which typically can happen then is there is bad data quality, which you should take the chance of cleaning this up for once.

Ex experience shows that from this point on, it's going to go downhill anyhow because all the governance that you try to take in place with, Hey, let's have a fresh start. It's like this time really, really do it right. Uh, typically it starts to crumble once like operational daily business basically hits in.

So at least take the time at this point and clean up what you actually can. 

Mehmet: Great advice. Now we can't skip ai. We're gonna talk about ai, but before going deep, about the AI use cases, [00:21:00] um, like what would separate or let's say, what's realistic, you know, today and what's aspirational? 

Dominik: Well, I mean, I'm working in this, in this field for so long, and to be honest, for me, AI was always something which I.

It is helping me along the way. When I get up at night at like, I don't know, 5:00 AM or something like that, nobody's there in my time zone, at least that I can bother asking a question and I could basically get back in front. So I mean, optimally. There would be an end to end AI orchestration across the whole project lifetime, basically from the beginning.

So you're selling the right thing according to the RFP. You're checking all the notes, basically all the way through the construction of the actual, um. Of the actual data migration rules that through the execution, the root cause analysis and all these types of things would be end-to-end, basically, um, um, orchestrated [00:22:00] and basically, um, supercharged with ai.

Now that's probably a little bit too farfetched because the field is vast. And of course when implementing such a system, and we are of course, uh, as SNP right now in the midst of developing such a system, the end-to-end AI orchestration is still something which is aspirational somewhere in the future.

But for individual pieces, and maybe we touched already a little bit about that, right? With the note keeping at the beginning and making sure that you got the right questions when you go into the surveys and into the customer, uh, workshops and so on and so forth. So this is where I can already help in many, many places more.

Right. I would say it is yet still a little bit spotty, but the big aspiration is to make it all integrated work as one system. Basically something that is helping you to govern this thing or govern this thing, um, where governance of people and processes might be even lacking from time to time. [00:23:00] 

Mehmet: Dominik where really it's helping, let's say the presales team.

Is it like re you know, responding to the RFPs or RF Qs or like, is it, uh, you know, you just feed it with some knowledge base from, you know, maybe previous, uh, projects that you've worked on. And there are some people who are skeptical still about the use of ai, especially with hallucinations and, uh. Other, other issues that we see in in ai?

You know, I tried couple of things, although like I don't have to do it in my day-to-day work, I would say, but I tried couple of, um. You know, tools and, uh, kind of mini projects that I assigned to myself. I said, okay, imagine I had access to AI when I was pre-sales and you know, really the results aren't bad.

Like, you know, but still you need the human in the loop, which all my guests are focusing on it. So like, tell me what's working well today when it [00:24:00] comes to, to, to utilizing AI in. B 

Dominik: before I start in that. I mean, you sure 150% probably, right? MeMed that as a human, you stay accountable. So AI is a tool for speeding you up or for making your work easier or even doable, um, in some, in some regards.

But you are ultimately responsible and accountable for the results that it basically produces, which is also why. You can misuse ai if you use it blindly. I mean, you have no way of validating what is a good result, then you probably out there in the open because you cannot follow up on anything that's basically happening.

Once you know what a good result should be looking like. Then AI can really like, support you here. So regards to your specific question, like pre-sales teams, I mean, you already mentioned that uh, very rightfully, you basically, although a project is a lot size of one because I [00:25:00] mean, it's a project, it has very many individualities.

To it, there is patterns that can be recognized. Right. And you typically, like you said, you feed it basically with a knowledge base, put a rack on top of it basically, so you can retrieve, um, basically everything in it. But one thing that we particularly do as s and p is for, for what we do at the beginning of each of the projects, we are scanning the source environment, the source landscape.

We try and basically get an x-ray of the source system because. I mean the, the, the, the ground truth. What all AI needs to be grounded into in order to like stop the hallucination to the largest portion of it is that it is, um, fed with facts. And of course facts reside in knowledge basis, although sometimes maybe it's arguable that, um, it's really facts.

One more, it's like. Well knowledge that's been been gathered, um, and what really works well. But we also mix in analysis of the source systems into this that really gives us an x-ray of the source of the, of, of the systems that are [00:26:00] basically enhanced, uh, to be transformed. So that's a process that we also feed into the whole procedure.

Um, we would do so actually, and we did even do so before the age of ai. So all the service that we do and all the workshops that we run are fed with this analytic information that will tell you. These are the business objects that have been used. You have so and so many invoices in history. For instance, in this particular entity, in this particular company code.

You are using this and that type of process in this other entity that you are leaving behind, are you sure you want to do that? So that's type of grounding that is very organizational specific, very system specific most of the time. I mean, it's not only needed for ai, but it's also um, um. I'm losing my thought here a little bit, but, um, it's, it's, it's really helping you to ground the whole conversation onto something which is, which is facts, which is not like how I think the system would behave, but how [00:27:00] the system is actually being used today.

Because ultimately that's the data that you're going to hit when you start doing the. 

Mehmet: And I think this would help also in the handover, uh, Dominik as well. Like, you know, I'm imagining if, if we have the common database, whatever you want to call it, about the project, I mean this particular project, so can I imagine?

Uh, as part, uh, as a, you know, one of the team members of the delivery team, I go to kind of an LLM and ask, Hey, can you walk me through what the presales team have done in regard to one to three? So instead of, you know, wasting time on emails and probably meetings to, to align, w would, would that also kind of, you know, seeing this?

Dominik: Absolutely. Absolutely. Ma. So, um. What this tool produced, again, also even before AI actually hit the [00:28:00] game, really is a structured set of documents. So if you are a services company, you likely have best practices. You have documents like for instance, a blueprint, and the blueprint would have very specific chapters in there would tell you like, Hey.

What's the scope? Which objects are basically in what timeframe am I carving out of the system or selectively migrating? What is the target that I need to go to, and so on, so forth. So having this artifacts basically, and most of them are really documents of some sort that are being produced as an output of this process.

Step right here. Of course with all this has been, the handover have always been the handover into the, into the delivery team. Um, however, of course with ai, now you feed the pro, you feed the document. Um, into, into a chat and you can ask very specific, uh, questions about that without needing to read this typically 180, 200 page document that this is, uh, in the end and really quickly get to the [00:29:00] answers that you are in your particular role I interested in.

So as a project lead, for instance, you will ask it about. The particular timeframes and maybe the off times that the customer does require, or what is their release cycle. So at what times can't, can I not migrate before going into actually catering for the next artifact, which is, for instance, the project plan.

So here, of course, AI really helps in. If you always have these intermediate steps in between, we have these artifacts which really wrap up a specific state and basically represent, Hey, this is the agreement that we have so far. Now, all the basically follow up things that that happen after that can really use AI to question very specific portions.

To their, to their specific role and project. Is that only one of them? The technical migration lead would be another would be asking about, hey, database sizes and system performance. So, and from there that derived and hey, what would I be advising for a target infrastructure to look like to be [00:30:00] able to cope with this data migration, for instance?

Mehmet: Great. Now let's talk about the transformation itself, right? So the logic. How do you use AI in a safe way here? Like, and there are a lot of, you know, talks now, now, like if you give a lot of autonomy to the ai, it can behave in a certain way, whether that might affect the. Security aspect of it, the privacy, or if it might go and, I don't know, it might go and read the whole data, which is not the thing that we want.

So how do we, how do we make sure that we are using the AI safely? 

Dominik: All right, Meir. Yeah, so I mean, you're completely right. I mean, data migration is one of the most critical parts, um, or the most critical things that you can actually do. It's mostly like open heart surgery, really, right? I mean, you want the, the system go out of the surgery of this restructuring basically, uh, [00:31:00] in a life and healthy state.

And so you have to, of course, make sure that this is really the case. And now AI of course, I mean, you know, it's not a super deterministic thing. You if, if you're like going to. I dunno, chat BT or something like that. You type in a question, you get an answer, you type in the same question, different chat, same chat.

You get a different answer. Symbolically might be the same, but if you think about that, and data migration needs to be audible in the end, needs to be deterministic, fully, meaning that for the same inputs you always get the same outputs. Um, you cannot rely on the AI doing the actual data migration.

Because I mean, that could lead to, it could be source of error. So what we do instead actually, is that we use AI for authoring the transformation rules for speeding this up, for coming from a customer specification basically, or a customer requirement, and translate that into the data pipelines and into the transformation logic that then actually is being executed.

You could get a deterministic result. You get basically a data migration [00:32:00] pipeline, or in our case we call this data migration objects actually. Um, but they then execute completely the deterministically, meaning that if you push the button and you export the data from the source, it's executing by the rules that were created, not by AI evaluating on the fly the data and moving it basically through the pipeline.

So this makes sure that. Why we gain speed specifically in the phase of, um, creation of those rules. Um, we still remain, uh, we still keep the determinism of the overall algorithm, which in the end, an auditor will rely on because if they're checking whether you did that the thing to the system correctly, they will also want to look at what actually was used to change, uh, change the data to extract imports and like transform the data along the way.

Mehmet: Great now, but we know, uh, Dominik, that things can go wrong, right? Um, and it, it, it's normal e even even from, from a human aspect or a machine aspect. Now, let me ask you, where does [00:33:00] AI help most once things start going wrong? Like, can it, can it like, fix things up? 

Dominik: It can help you. It can help you fixing things.

First of all, it can help you find things much more easy than, um, you would regularly take. So let's say you have an error, and I always say everything that had has not been tested is definitely broken. Most of the things that did, you did test is still going to be broken until you went through a couple iterations.

So let's assume things breaks because they do, um, in data migration that typically surfaces as part of the inconsistency of the target system. Or you just don't get the data in because the interface that you're posting the data against just gives you something like, Hey, you wanna move the customer in, but I'm sorry, I require bank data for it.

Now what you then typically do is you do root cause analysis. You go into the system, you try to find out what's actually wrong here. And because of AI being very quick, um, when having large contexts available. Um, you can do this root cause [00:34:00] analysis much quicker than you typically be able to do this as a human being.

Now, of course, it doesn't help with a specific problem if you just rely on general AI knowledge and capabilities because they don't know anything about the systems underneath. So what we built in our platform, Keanu, or building in our platform, Keanu, is a, uh, uh, on top of all the engines that we basically have a layer of MCP.

Model context, protocol servers that actually expose the functionality that as a user you would've had now actually can also give to the agent. So if you want to explore, like, Hey, why is this particular thing inconsistent? It would go into the tool, do the navigation that you typically would do, or.

Actually would rather call an API, um, that basically supports this and give you the insights right away. Also, when the, when maybe that information is not enough, it would correlate it with other information that's also somewhere in the system, somewhere in the log and basically present you with probably a resolution.

So. Probably [00:35:00] this helps a lot. Not probably. This does help actually a lot and I will, would like to, well, it puts basically the consultants on steroids because they can do much more in much less time. And basically this is really improving the quality or the security or minimizing the risk of data migration overall.

Mehmet: Uh, you know, we've, we are seeing more and more MCP integration, so, uh, is it something like, you know, I can, uh, you know, just for the sake of the, um, you know, audience that might not, you know, know the full potential of these cps, right. So, and MCP servers, so, uh. Like, to what extent do you think they can be autonomous and to what extent do you think that they can, um, play major role in these migrations?

Like, yeah. You, you mentioned a couple of of points here, but I would like you to, to, to expand a bit more so people can in general understand the potential [00:36:00] of, uh, these, uh, servers, MCP servers, and of course everyone talking about agents nowadays. So I would like you to, to explain that more. 

Dominik: Yeah, sure, sure.

So generally speaking, what an MCP server is, basically it is exposing the functionality of your application like an API, but to an agent, um, which typically means that you have to provide in this MCP server much more context than a typically API would, would have that you hard wire against when you build a system to system integration.

That foundation ultimately gives you the chance to expose all the functionality in a rich already existing platform to um, to, to the agent. Now, about how much autonomy would I typically give it now in such a delicate thing is data migration. My general way of going about it is. For any MCP action that has potentially changing impact or side effects on the target system, meaning that it could potentially change the state of [00:37:00] the system because it's.

Well migrating data in itself, or it's, uh, trying to like delete a lock or something like this. This would always be actions that would be user guided. Basically, I would rate the confidence of this in this particular scenario low. So you would always ask for permission from the user, meaning the human is in the loop in this scenario.

In other MCP service and capabilities, specifically the ones which do analytical tasks, where it's like read only to the system and they are side effect free, meaning you can repeat them over and over and over again, right? You get new systems, state that you can explore, but you don't have no, um, no chance of harming the system.

I let them go. Basically. I let them flow because I mean, they. Don't, like, could put me into the loop. Yeah. Can I like explore the logs? Um, yeah. Can I correlate that with the, um, data consistency, verification and so on so forth? I don't care about that. Yes, you should do that. But should I actually like set back the data migration data before and retry the import?

Oh yeah. I wanna be on top of that. I wanna be able to control this. So MCP really is of course, a [00:38:00] way of making it full autonomous by allowing everything basically. In the data migration scenario, particularly, I would advise anything that is harmful, potentially harmful to the system. Not to allow directly, but basically use it and be, have, have the human basically stay on top of this.

Mehmet: Great. Now, as a business, how this might, and the AI in general, I'm, I'm speaking here, help me in, you know, situations where I need to do validation and testing. 

Dominik: Yeah, big piece actually. If you look like the overall, like what customers spend their time on, it said, not exact numbers, but it said that 40% of the time is actually being spent, uh, spent in preparing the test, actually executing the test.

So obviously there's a huge potential in AI supporting and helping them. So now the question is like, how do we actually speed up tests? Right. Typically, I mean, the best test is, is a test which runs fully automatically and gives you full confidence that everything basically worked out in the end. But how do you get to an automatic test?

Um, [00:39:00] typically there is test authoring. Um, test authoring starts typically with, you write up a spec where it says like, I do this step. You expect this results, I do the next step. I expect this results. And in a very simple scenario, human would actually go through this and, and, and execute this now. The next level after that, I mean, that's a decade old industry is test automation actually.

So now the question becomes how do I, how can I create tests, um, in a very, very, uh, easy and well seamless fashion basically. Now, AI can of course help for structuring, for instance, process documentation that you have within your company and create a manual test script out of this with just this step approach.

And then that's feed it again. And if you have a platform underneath that's actually capable. Of carrying out, uh, test automation, you can now translate these very distinct steps into something which, um, uh, the underlying test automation software does understand. So there are a couple of vendors out there, but this is [00:40:00] basically like at the brink of where this industry currently is like.

Making sure that from a process specification, I get to something which is executable basically in the end. And now one of the things that typically happens is your tests break. And your tests break because of mainly two things. One, you set it up or when it actually finds an error, which is in there. The second thing is.

It's prone to change because you changed something in the system. Configuration in a test, which was supposed to run clearly before is now not anymore because of you changed how the process is going to work. And this is where, say, of heating capabilities of AI can actually overcome a strict test script which just runs and it basically stops and blocks.

Whenever a button has been moved to a different place or a value is something different, then you can basically. Go to the AI and ask it. Well, I mean obviously I didn't find something, but can you actually try to find where it is now? Can you assess like how the data is being newly correlated or how it's newly is being set up?

Where did that button [00:41:00] actually move? And that, at least for the self-healing process of that, I can also help in that particular step, resulting in the end in a well really disruptive way of how business can achieve much quicker validation times in a project than they would if they did manually. Or they have test automation and have breaking tests all along.

Mehmet: Right now, of course Dominik, you know, such projects are great. Um, I would say opportunities to learn, um, especially migration projects. Right. So, but I'm sure you can share with us, like what do you see, like some of you know, the biggest opportunities maybe organizations still miss after they go live. 

Dominik: Well.

Let's say you're a professional service organization and you pro do project after project after project. In the past, you typically relied on the consultant, basically having a memory of what they did and then apply and transfer this to the next time. So I think [00:42:00] the biggest challenge still today is knowledge loss, loss of I explored and, uh, seen a pattern and learned something in the project and not being able to transfer it into the next project because.

I personally am not staffed. Documentation is unsufficient. Um, or maybe somebody else at the other end of the world in my, in my corporation or in in, in my services organization does the same thing, but doesn't know that this was already done and the solution for a particular problem was already found.

So one of the key things here where I, it would be a massive help. Um, to be honest, a little bit of an outlook into the future here really. But, um, to consolidate the results, document them in a structured way, um, better than the average human would do, which is actually not so hard to do because the underlying information you typically have, right?

You have a rule set, a pipeline, that's the deterministic, right? That is executed by a machine already. So that is a foundation from which you can [00:43:00] redraw documentation. If you don't have at. Um, then you have logs, for instance, right? You have protocols which have been written out, which basically tell you about the system, say also a great source of knowledge, specifically when cases go wrong, if you pair this with a defect management system, and also, I know there's of course, data privacy concerns around that, but if you can act, extract knowledge around that, and also feed it back into a knowledge base that you can, uh, basically can reuse the next time around.

Then of course I think you are in a very, very good state because then you are running into an incremental cycle of always learning the ai, basically becoming your governance engine and making sure that you don't miss anything and you refeed it basically back into your repositories, into your knowledge basis, and the next time around the agent execute.

That knowledge is present and available. Basically, I think this is where most companies today, the knowledge laws that basically gets lost because of people are leaving a specific process or, um, uh, a specific project or so, and are not in the [00:44:00] next one. This is something that AI can really help overcome because it's mostly about re documentation, making sure to put best, uh, back everything consistently into a knowledge base.

And I mean. We all know that AI is really, really great at digesting a lot of unstructured information, summarizing it, and putting it to the format that you actually can read again and, and reuse. 

Mehmet: Right? You mentioned, uh, you know, Keanu, your, your, uh, agent, uh, few moments ago where all this gonna end, Dominik.

Dominik: So, I mean, from an aspirational perspective, all of what we talked about is something that we want to put into the platform. We are currently at the point of building this. Um mm-hmm. We are concentrating on what we already have. So in Keanu, that's a data transformation and data management platform that we s and p build.

I myself spent the last 15 years of my career at least to like build this and most of the time my job was [00:45:00] basically go and find the consultant, eager to like offload the work off their head and help me put it into the software itself. So it has a great foundation already with best practices and transformation content and all this sort of stuff, which is actually the foundation that you need to now put.

CPS put agents on top of to run automation and basically you have structure for how you want to process the project, which the software already knows. You have an existing knowledge, which, which you only need to enrich much better starting point than if you basically start from a, from a blank sheet of paper and you have to invent the processes basically all the time themselves.

You know, refinement is always easier than starting off from scratch here at this space. So we are currently in the stage of building out. The automation layer on top of the tools and the top of the, of the knowledge basis that we have. But ultimately, all the areas that we have been discussing in this session today, of course, would be something I like to tackle sometime in the future.

I'm not saying in the end I wanna have a click button solution that automatically runs. I don't need to [00:46:00] people anymore. I just like people too much to like avoid them in the process, right? What I wanna build in the end really is something that puts. Consultants on steroids makes them much more effective basically, and helps them to cope with the already very stressful work that they typically have.

So when consultants typically come to me, they say, Hey, I have like 2, 3, 4, 5, 6, 7 projects running in parallel, right? And I have to cope with a mental, with a mental load that basically this puts on me. If I can be any help in that regard to get this down, we can also help to scale out. The whole industry basically, to be able to deliver more of those projects, which is very necessary because change is happening actually even faster, right?

I mean, you know, due to the conflicts, right? You have to carve out companies outta specific countries, right? You have system modernization pressure. You need to adopt new technologies much, much faster, right? You need to reshape your business all over. You are buying a company, you are divesting a [00:47:00] company, right?

All these changes somehow have to materialize. And to be honest, at least in my perspective, the rate of change in this regard over the last 10, 15 years has been growing almost exponentially. Right? And it's hard to keep up because they're just not the experts available in that mass that the change is happening.

So, I mean, good that we have now a good opportunity with AI to support this particular field. 

Mehmet: You know, I gonna kind of ask you the final question based on, on, on this. It looks like we have, I would not say two camps, maybe I would say multiple camps, but all they maybe diverge or converge, sorry, into, into few.

So we have the camp of people who are going fully on, you know, the use of AI and, you know, age especially ages like 20, 26 for. For sure. It's, it's the year of agents. Like there is no doubt about it. And you have the people who are [00:48:00] going fully on saying like, you know, jobs are gonna. Be redundant for a lot even of consultants we've seen couple of weeks ago, you know, the big fours and what's happening over there and you know, and then we have the other camp, which they are saying actually companies, they will be very slow and they will not be adopting the technology because actually, you know.

They, they, they don't have the capability. And then you have another, another camp, you know, which is kind of mix of both, like saying, okay, these companies will, you know, they, they'll go out of business actually because, you know, like another company will come, maybe they will buy their, uh, their data and you know, they're gonna build on top of it using the AI agents.

If we want to end this episode, Dominik, into an outlook. For you, someone experiencing this domain, and you know, there's, I don't think someone better than you because you, you, you, your mission is to translate business needs into technology and vice [00:49:00] versa. Like convert the technology advancements into business outcomes.

If you want to summarize and, and close with something regarding where we might be, uh, heading in this and your personal opinion, of course, I would love to hear it. 

Dominik: Yeah. So. I, I think I'm, I'm in the camp of AI won't replace people and won't replace jobs. It will change jobs. It is, mm-hmm. A very massive, uh, disruption.

More than I can actually remember, but. It is basically like the automation that happened a couple, couple years ago. Uh, basically, and so in this sense, I don't think AI will replace jobs specifically. Not in an area which I experienced, where people are constantly overloaded anyhow. Right. And they don't know how to cope with the work.

They will, I, I will just eat the people that. Don't use it. So somebody with AI will always be better or [00:50:00] will become better than somebody not using AI at this space. So you basically have to find your way about how AI makes you a more productive individual or a more productive organization, basically.

And if you can find this lean way, then this is like a, a great, a great way to an average. To me, fundamentally, it's not a super different threat than automation to different jobs basically in the past and is today. You always need to adopt as an individual or organization, and if you stay on top of this, if you embrace it really.

Then you can leverage whatever comes out of this, and you can stay ahead of the game. If you don't, well, then you'll be let bad, and then you also may be laid off at some point, right? Because you just don't bring the effectiveness that now becomes basically like common standard, right? Or like, like, like convenience, so to speak.

Mehmet: Right. You know, and that's why I'm encouraging people, like at least, you know, at least try the tool. Maybe it's your own freedom not to use it. I mean, any tool that, I'm not speaking about [00:51:00] a tool in specific or a solution in, in, in, in specific terms, but at least go and watch a demo. Of course, I know like there's a lot of noise out there.

People, they sometime exaggerate, you know, a few things, but actually some of the AI technology it's in, in production today, it's working like it's doing stuff in a proper way. I can, I can tell from my own experience also as well, there's no doubt. You know, and, and to your point, like I a hundred percent agree with you about the point, like we need to educate ourselves.

And Yeah. And if we just watch and do nothing, yeah. We, we gotta become obsolete. 

Dominik: You, you, you're gonna become obsolete, as you say, basically. For me personally, it's always like trying the stuff out because there's a lot of hype out there. There's a lot of hype out there, and only like fraction of this is really true at this truth of the moment.

I personally always try stuff out to see how it really works, how it fits my workflow, how it improves my life, [00:52:00] basically. And, uh, so this is how I basically try, is just to, to stay on top of the game. Now, this of course, can become very, very overwhelming because, I mean, new models drop like every second day nowadays, right?

So you don't need to be in the top nine and top 1% of the adopters, right? For me, uh, my middle ground always was like, okay, stay in the top. Five to top 10% of adoption. Don't follow every trend because you will be overwhelmed. You will be chasing right things which are not real then in the end. But if you see that something gets a little bit of traction and after you've been in the field for a little bit, you get a feeling of what that really is.

Yeah. Then you can actually start up, you should adopt, uh, early, but you should adopt, be like the very, very first other. Then you have a lot of time. Right. And, uh, yes, a lot of interest in like doing, doing, uh, like exploring and, and do stuff. Then happily do that. But like the average fully employed person with a busy schedule, right?

You don't need to be in the top 1%. It's okay if you're in the top five to top [00:53:00] 10%. 

Mehmet: Yeah. As we used to say back in the days beyond version and minus one, it's okay. It's, it's a stable version. Yeah. It does the job. So 

Dominik: yeah, just nowadays, versions pop out every second day, so it's still a lot of, 

Mehmet: yeah, so I mean like maybe n minus one month, let's say, like, uh, yeah, this is, that's 

Dominik: fair.

That's fair to say. 

Mehmet: Yeah. And AI world like this is how, how I, I personally, I start 

Dominik: two generations. 

Mehmet: Yeah. Crazy times to be, to be, uh, e witnessing honestly. Right. Dominik, really, I enjoyed the conversation. I think it's one of the. Rich, uh, uh, I would say context driven episodes that I have made because we touch both on business and technology in, in, in a full fledged way, I would say.

So really appreciate the time and you know, I know how busy things can, can be for you. Finding thing where people can get in touch and learn more about you and, and the company. 

Dominik: Well, of course, I mean, you can find me on LinkedIn, right? Uh, that's, that's the most easiest, [00:54:00] uh, thing to do. You can probably read a lot of like, uh, press and articles as well.

If you want to, you can actually join our, um, our big community event, which is the transformation world happening July 8th at ninth this year. So that's an annual event. It's great. It's happening in Germany over here, so I know not everybody's able to come, but then you are able to watch the episodes of the streaming, um, of all the, all the stuff that's happening there.

You're going to see a lot of technology advancements that we put forward as a company. But even more important, our customers speak on stage about how they achieved success in projects that they did on top of our platform, which I think is even more important to understand and no, because then you can relate, okay, how does it actually translate to my business?

So if you wanna join that event, go ahead. It's really, really great. 

Mehmet: Mm, great. I gotta make sure I'll put, um, you know, your LinkedIn profile, uh, URL in the show notes, I gonna put also the company URL also so they can learn more. And of course, you know, follow [00:55:00] for, uh, uh, more information about the event that, uh, Dominik would be sharing, I'm sure.

Uh, and of course it'll be on the company website now. Uh, Dominik, thank you again very, very much for, as I said, this rich, uh. Episode from both perspectives, and this is how I end my episodes. This is for the, my audience. If you just discovered us, and I'm saying this, I know, yeah. After four years, people still discovering the podcast, I'm happy for that.

Uh, I need you f. A small favor if you liked, you know, the content. I appreciate if you can suggest it to friends and colleagues 'cause we are trying to get you the latest and greatest with best guests ever, like Dominik. And we are discussing like the latest topics which are relevant to today's, you know, needs in, in, in the workplace and you know, in life in general also as well.

And, you know, subscribe also as well. This would be a big. Push and help. And for the people who keep coming back and they listen to the show, they check their episodes, they send me their emails, they send me their. [00:56:00] Messages on LinkedIn. Thank you very much for doing so because of you. Actually, you know, the podcast similar to 2025.

This is where we, I start to see this real push. We are getting every week in the top 200 Apple podcast charts in different countries. Yesterday I saw like we were like in six countries differently. I need a little bit. From my friends in North America, Europe. You're doing fantastic. I think we need to do it in Germany, in Dominik, so it's for my German France.

So maybe you can give, give us the push over there. 

I 

Dominik: can give them a hint. 

Mehmet: Yeah, because really I'm repeating this at the end of each episode because what we are trying to do is to empower people with knowledge, bringing experts like Dominik, like showing you what's happening in the world, both in the business and technology, and startups also as well.

So I appreciate always this push, and as I say, always stay tuned for any episode very soon. Thank you. Bye-bye. 

Dominik: Bye.