Earley AI Podcast - Episode 87: AI-Enabled Enterprise Data Migration with Dominik Wittenbeck

In this episode, Seth Earley speaks with Dominik Wittenbeck, Group CTO at SNP Group, a 1,600-person global software and solutions firm with 30 years of SAP-centric data migration expertise. They explore why AI is only as good as the institutional knowledge behind it, how agentic AI is transforming high-stakes enterprise migrations, and why organizations must treat data migration as a strategic opportunity rather than a cost-reduction exercise. Dominik shares hard-won insights on semantic architecture, governance, and what executives consistently get wrong when applying AI to critical enterprise processes.

Key Takeaways:

• AI is not a silver bullet for data migration - it requires deep, domain-specific knowledge to produce deterministic, auditable results.
• Enterprise data migration is a team sport requiring cross-functional specialists; AI accelerates the work but cannot replace that expertise.
• The real opportunity in migration is not just moving data - it is cleaning it up and optimizing processes while the organization is already changing.
• Agentic AI is transforming the full migration lifecycle, from pre-sales solutioning and blueprint generation to rule creation and automated testing.
• Governance established once without ongoing enforcement decays quickly - organizations must build continuous oversight into critical processes from the start.
• Value mapping, not just structural mapping, is the dominant challenge in SAP migrations, and AI can significantly accelerate semantic alignment work.
• Executives should focus AI investments on problems that truly matter, not easy wins - meaningful impact comes from finding where differentiation really counts.

Insightful Quotes:

"In order to run complicated systems which have a critical impact on your business, they need enough grounding. You actually need to feed the knowledge into the agentic system that you're building on top of, in order to make sure that you get deterministic results in the end." - Dominik Wittenbeck

"Rather than re-architecting the whole thing, try to identify what the critical processes really are, that if they are not exercised correctly, really hurt your business. Find where the value lies - or if you can't find that, find where your risk lies." - Dominik Wittenbeck

"Sometimes cheap is quite costly, and sometimes slowing down speeds things up. If you're moving stuff from one system to another and you say, we'll clean it up later - that's never going to happen. It's like moving from one house to another with an attic full of boxes and junk." - Seth Earley

Tune in to discover why successful AI-driven enterprise migration depends less on technology and more on institutional knowledge, governance, and treating transformation as a strategic opportunity.

Links

LinkedIn: https://www.linkedin.com/in/dominik-wittenbeck-61a64669/

Website: https://www.snpgroup.com

Ways to Tune In:

Earley AI Podcast: https://www.earley.com/earley-ai-podcast-home Apple Podcast: https://podcasts.apple.com/podcast/id1586654770 Spotify: https://open.spotify.com/show/5nkcZvVYjHHj6wtBABqLbEiHeart Radio: https://www.iheart.com/podcast/269-earley-ai-podcast-87108370/ Stitcher: https://www.stitcher.com/show/earley-ai-podcast Amazon Music: https://music.amazon.com/podcasts/18524b67-09cf-433f-82db-07b6213ad3ba/earley-ai-podcast Buzzsprout: https://earleyai.buzzsprout.com/

Podcast Transcript: AI-Enabled Enterprise Data Migration, Governance, and the Knowledge Foundation

Transcript introduction

This transcript captures a conversation between Seth Earley and Dominik Wittenbeck about the high-stakes world of enterprise data migration, where a single hour of downtime can cost millions. They discuss how SNP Group has spent 30 years codifying expert knowledge into structured, reusable repositories, why that foundation is essential for effective agentic AI, and what executives must consider when approaching SAP modernization, semantic architecture, and enterprise governance in an AI-first world.

Transcript

Seth Earley: Good afternoon, good evening, good morning, whatever your time zone is. Welcome to the Early AI Podcast. In our podcast series, we basically talk about what organizations are doing to be successful with AI, what the challenges are to that success, how to get from proof of concept to production, how to deal with data issues, how to deal with agent issues, security, compliance, all of those types of things.

Today, we're going to talk about something that's very high stakes for organizations - enterprise data migration, and how it impacts AI initiatives when you're doing a merger, a divestiture, an SAP platform upgrade. The data has to move. And if something goes wrong during that process, there's certainly significant impact to the business, and it can be very consequential. When we try to leverage AI in all of these initiatives, we introduce new factors, new risks, new compliance issues, new challenges that didn't exist in prior iterations.

My guest today is Dominik Wittenbeck. He's Group CTO of SNP Group, a 1,600-person global software and solutions firm that has spent 30 years doing SAP-centric data migration. Dominic has been with the company for 25 of those years, and he's spent most of that time doing something that will resonate with our listeners - talking about what's inside the heads of experienced consultants, codifying that into structured, reusable knowledge. SNP now has what they call a transformational repository covering millions of data points across the SAP data model, plus guided procedures with 1,500 discrete steps for running a migration. This is 15 years of formalized enterprise expertise. And now they're building agentic AI systems to work on top of all that knowledge. Dominic, welcome to the show.

Dominik Wittenbeck: Thank you for having me, Seth.

Seth Earley: So, when you talk to executives about applying AI in data migration and replatforming, what are the biggest misunderstandings that you've come across?

Dominik Wittenbeck: I think mostly it is that AI will solve basically everything. The field that we're in is born out of expertise of two decades plus. I spent most of my professional career extracting knowledge from consultants and trying to formulate it in a way that actually fits into software. The biggest misconception is that AI is going to be the silver bullet that heals everything out of thin air. In order to run complicated systems which have a critical impact on your business, they need enough grounding. You actually need to feed the knowledge into the agentic system that you're building on top of, in order to make sure that you get deterministic results in the end.

The misconception is that this all becomes very easy - a save time, save cost type of problem - where in reality, what you really want to make sure when you're doing data migration is that you're doing it correctly, efficiently, safely, and auditably. Because ultimately, your system is going to get audited, and whatever has been done incorrectly in there will surface at some point.

Seth Earley: That's a great point. It is about risk mitigation, about doing it correctly - not necessarily about doing it as cheaply as possible. Sometimes cheap is quite costly, and sometimes slowing down speeds things up. Data migration requires teams of specialists who understand the source and destination systems. There's too much complexity to hold in one person's head, and that's a classic expertise bottleneck. How has that scaling constraint shaped your approach to AI?

Dominik Wittenbeck: Migration is a team sport. You always want cross-functional, interdisciplinary teams that mix people from the business with technical people. In data migration, you're modernizing a system full-fledged, touching all processes and all data. You have to get a lot of people into the game, both from the customer side and from whoever is doing the migration.

How does AI actually help? It starts even way before you hit any data transformation tooling. When you're presented with a migration opportunity, a customer starts with an RFP showing where they want to go - but it doesn't necessarily state where they come from. For 15 years or so, our approach has always been to get as much transparency into the as-is situation as possible, so you can evaluate very precisely how to actually get to the to-be situation. We always run a scan on source systems to truly understand the facts and figures - which modules are in place, which processes are run, which data is on those systems, and to what degree.

With AI, even at the very beginning of a project, when you're fleshing out the solution, you can use AI to bring all of that together and produce a very distinct, concrete, customer-specific survey that ensures all the key decisions are made at the earliest possible point. And that's not even getting into the downstream platform that actually does the data movement.

Seth Earley: Give me a before and after. If you did a project a couple of years ago with minimal AI versus what it looks like today using agentic approaches, what changed?

Dominik Wittenbeck: Let's run through a project virtually and see at what points things have changed because of agentic capabilities.

At the pre-sales and solutioning phase, one of the first things you do is craft a project timeline, a project methodology, and a migration blueprint. Crafting that blueprint, if you give AI the proper template that we've formulated over years, is something that's much easier to do now - you can get a good proposal to begin with. I always treat what AI produces at that point as a proposal. Ultimately, you as the solution engineer or enterprise architect have to take responsibility and accountability for the end result.

Next, you take that blueprint and build a data pipeline from it. Our platform has guidance procedures - thousands of interdependent steps covering almost all the eventualities that might happen. With the blueprint in place, we can dynamically switch on and off the relevant pieces. The pre-configuration that would typically be done manually is now significantly accelerated.

Then you craft an individual rule base. When you have the procedure and know what steps to take, you need to adjust the templates - maybe you're changing the chart of accounts, reorganizing profit centers or cost centers. These rules can now be scaffolded with the help of AI. Still, we're using the underlying system and the way it's been executed, but the rule generation and pipeline crafting - that's what gets accelerated.

Then you run the pipeline. What you really want to make sure is that it runs deterministically - the same inputs produce exactly the same outputs, because that is a necessity for an auditable system. And at the testing and validation end of the process, there's a whole lot more AI can do now versus former times when much of the testing had to be done manually.

Seth Earley: So you're also looking at process engineering and optimization alongside the migration itself?

Dominik Wittenbeck: Most of the time, customers are looking for these optimizations. The initial spark typically comes from a customer's desire to become an enabler of the business. While you're going through this change anyhow, don't do another project a year down the road - inline the improvements right away. That way the time to value that you provide as an IT leader to your organization increases.

You also have a once-in-a-lifetime chance, as a CIO, to make things clean again. But you have to choose carefully, because the more change you incorporate, the more burden the business has to take on within the project. Cater the scope to what your organization can actually digest. The change management required to get people on board gets harder the more change you accumulate in one project - and that is something independent of AI.

Seth Earley: What are the big drivers and catalysts you're seeing in the market right now?

Dominik Wittenbeck: From our perspective, the biggest push we're getting is from the SAP S/4HANA wave. SAP is one of the biggest ERP vendors globally, and they announced a shift from ECC to the S/4 platform roughly 10 years ago. The reason it's taken so long for organizations to adopt is that they've invested 25 years optimizing their old systems, and one of SAP's key selling points was always the high degree of customization you could apply. When you have to restructure that - potentially moving to a more standardized application set - you have to think hard about what you can let go and what you should take with you.

Modernization is the dominant driver, fueled by two things. One is that maintenance for the old systems is going to run out - if that deadline hits you, you're in a bad position, because you have to do it no matter what and can no longer choose the pace. The other side of the coin is that the new systems are highly AI-enabled. If you get all your historical data reshaped into a form that new AI use cases can readily use, that opens up new business models and capabilities. So those are the two main drivers. Mergers and acquisitions happen all the time, but modernization is really where things are peaking right now.

Seth Earley: How do you think about semantic architecture in the context of these migrations? The translation of business concepts across different systems and levels of granularity is enormously complex.

Dominik Wittenbeck: If you are in an SAP-to-SAP data migration, you're actually a lucky person, because SAP, being a German company, has over-engineered and very precisely formulated its data model. You find very little generic and shapeable terminology. What you find instead is around 100,000 database tables, with an average of 250 fields each, and those fields typically have multitudes of distinct values. Once you've had the chance to work with that system for 25 years, as we have, you learn a lot about how that data model works. The terminology, methodology, and taxonomy are pretty much set at the structural level.

What I typically refer to as value mapping versus structural mapping is where things get interesting. The structure in SAP is pretty fixed. The values are very fluid. Value mapping - finding how terminology maps from point A to point B, from system A to system B - is one of the dominant exercises in the blueprinting phase. You sit with customers in workshops saying, this value meant this before, I don't find it in the target system configuration, I found these other values that might have the same semantic meaning - is that really the case? AI can help a lot here, because the evaluation mostly happens at the text level, where semantics are implicitly encoded.

When people want to move into SAP from a non-SAP system - a classic re-standardization case - you often encounter homegrown tools built by people who are long gone or retired. Here, you need to gain an understanding of the source system from scratch: where does the product actually sit across which tables, how does it relate to purchase orders and invoices? Gaining that semantic understanding of the system is the predominant challenge before you can move it to the target.

Seth Earley: How do you think about governance and maintaining control as these processes become more interconnected and autonomous?

Dominik Wittenbeck: There are various ways organizations tackle this. Some basically let it flow with little governance over how processes evolve. Others try to stay on top of it, which typically leads to redocumentation projects - let's put all our processes into one system, get them well documented, and so on. In my experience, many times these are one-time efforts. After they're done, everyone pats themselves on the back. And then what's lacking is the downstream governance that ensures the consistency improvements don't decay - which they do, pretty quickly, once you lose sight of what you're actually doing.

Rather than re-architecting the whole thing, try to identify what the critical processes really are - the ones that, if not exercised correctly, really hurt your business. Not everything is at the same level of risk. Find where the value or the risk really lies. Think about the last escalations you had because a system wasn't working or a process got stuck. Identify those points. Find a remediation. Establish governance at those exact points so you're sure that in the future, those things don't break.

It's like test-driven development in software - you stand up the contract for how it should work before you do the implementation. Few organizations actually do that because it comes with cost and less spontaneity. But at a bare minimum, when you find a critical failure point, make sure it gets tested going forward so that same problem does not happen again. It's about finding balance, and it takes experience to find it. But establish governance where it matters most - and then you have other problems to take care of the next day, not the same problems again.

Seth Earley: Dominic, final words for executives thinking about what to prioritize moving forward?

Dominik Wittenbeck: Find a problem worth solving. Don't find the easy problem that gives your organization a quick boost. Find something that really matters in the end, and try to put automation on top of that, put governance on top of that - and if that ends up being a transformation project that modernizes your systems, so be it. Try to make a meaningful impact by looking at your organization holistically and finding the spot where differentiation really makes an impact.

Seth Earley: That sounds great. Dominik, thank you so much for your time today. It was great to have you, and I appreciate your thoughts and your insights.

Dominik Wittenbeck: Thank you so much, Seth, for having me.

Seth Earley: And thank you to our audience. We will see you next time at the next Earley AI Podcast. Thanks, everybody.