AI with ERP, SAP BTP and SAP BDC

00:00:01: The podcast accompanying the E-three cover story is a critical and constructive discussion for the SAP community from the perspective of the E-three editorial team.

00:00:09: Two AI avatars are equipped with all the sources from the E-three editorial team and explain the challenges, tasks, and pitfalls to existing SAP customers.

00:00:18: The respective topics are discussed critically but constructively, with lots of tips and tricks.

00:00:25: SAP AI, or Business AI, is based on the Business Technology Platform SAP BTP.

00:00:33: The Predictive Analytics Library, SAP PAL, forms the basis for embedded machine learning in Sforhana.

00:00:42: LLM and Generative AI are integrated via the BTP's GenI Hub.

00:00:48: This is where AI models from SAP and hyperscalers are orchestrated.

00:00:53: Event note, AI, ABAP, and SAP Juul.

00:00:58: are also central topics at the SAP Communities Steampunk and BTP Summit on April twenty-second and twenty-third twenty-twenty-six in Heidelberg, Germany.

00:01:09: But now to the E-three episode.

00:01:11: AI with ERP, SAP BTP and SAP BDC.

00:01:18: Welcome to the deemedive, where we unpack complex research and turn it into clear, actionable insights for you.

00:01:24: Today we're tackling a massive topic, SAP and the whole world of artificial intelligence.

00:01:30: We've gone through a pile of material from E-III Magazine and SAP Press, and our mission is really to figure out the strategic role of generative AI and see if SAP's big new assistant, Juul, is actually the game changer.

00:01:43: they say it is.

00:01:44: It's a huge topic, and honestly, the timing couldn't be more critical.

00:01:48: The consensus across the entire industry is, well, AI isn't just some future tech anymore.

00:01:52: It's the absolute key to making sense of all the big data we have to enabling the IoT to real automation.

00:01:59: And the promises are, I mean, they're enormous.

00:02:02: We keep hearing this headline number from SAP's CEO, Christian Klein.

00:02:06: He's promising up to a thirty percent productivity gain.

00:02:08: Across everything.

00:02:09: Thirty percent.

00:02:10: Yeah.

00:02:10: Driven by automation and smart analysis.

00:02:12: That's not just an improvement, that's a revolution.

00:02:15: It is a revolutionary promise.

00:02:17: But, and this is the tension we found in all the sources.

00:02:20: There's a paradox here.

00:02:21: On one hand, you have experts calling Gen AI a Gutenberg moment, you know, a shift that just fundamentally changes everything in IT.

00:02:28: Okay.

00:02:29: But on the other hand, the big question we have to ask is, Is SAP building that revolution?

00:02:34: Or are they just getting really good at writing the coattails of what everyone else invented?

00:02:38: And

00:02:38: that thirty percent goal really hinges on the answer to that.

00:02:41: All right, so let's start with the core technology here.

00:02:44: Large language models or LLMs.

00:02:47: Given that Gutenberg level potential you mentioned, I was a little surprised to see how cautious SAP has been about building its own massive from scratch LLMs.

00:02:57: And that caution is actually the first clue to their whole strategy.

00:03:00: I mean, they figured out pretty early that the cost and the sheer talent needed to build a foundational model that could compete with, say, open AI or Google.

00:03:08: It's just not feasible.

00:03:09: So their play is integration.

00:03:10: Exactly.

00:03:11: Integration and building an ecosystem, they provide the official gateways, the interfaces to all the major LLMs, and it's all channeled through one place.

00:03:20: the SAP Business Technology Platform, or BTP.

00:03:24: I

00:03:24: also saw that they're backing some specific players, particularly in Europe.

00:03:28: That's right.

00:03:28: They've put millions into a German startup called Aleph Alpha.

00:03:32: It's a hedging strategy, really.

00:03:33: It secures regional access, maybe fosters a bit of competition.

00:03:37: And if you look at their own documentation for Juul and so on, the focus isn't just on a chatbot.

00:03:42: It's on these new concepts, things like... Document AI and critically something they call jewel skills and agents.

00:03:49: right those skills and they seem to be the actual mechanism for getting that thirty percent productivity elite, but it feels like there's still some Strategic ambiguity.

00:03:59: our source has pointed out some criticism that top execs like Jurgen Muller and Thomas Sauer seek don't seem to have a clear long-term ERP AI master plan, you know a roadmap that goes out to twenty forty

00:04:10: And that's the heart of the critique of this evolutionary approach.

00:04:14: They have fantastic integration tools for sure, but the big overarching vision for how AI is going to fundamentally redefine core business processes, it still feels a bit fragmented, almost reactive, which brings us to the key piece of tech that makes their whole partnership model possible.

00:04:30: The HANA Cloud Vector Engine.

00:04:32: Exactly.

00:04:33: Okay, let's get into the vector engine because this feels like the linchpin.

00:04:36: This is the thing that turns a generic public AI into a secure enterprise-grade tool.

00:04:42: Why is it so important?

00:04:44: Well, think about the single biggest risk of using GenAI in a business setting.

00:04:48: Hallucinations.

00:04:49: It's

00:04:49: just making stuff up.

00:04:50: Right.

00:04:51: Giving you an answer that sounds great but is factually wrong because it's based on public internet data, not your private sensitive ERP data.

00:04:58: The vector engine is the security guard.

00:05:00: It's the context layer.

00:05:01: So it's more than just a database.

00:05:03: Oh, much more.

00:05:04: Its whole job is to securely combine the amazing language power of an external LLM with the very specific high integrity data that's locked inside your ERP system.

00:05:14: It does this by turning your business data into mathematical vectors, a format the AI can understand and reference.

00:05:20: So it grounds the AI's answers.

00:05:23: And

00:05:23: perfectly put, it grounds the results in your business facts.

00:05:26: It's the bridge that makes external Gen AI trustworthy enough for the high-stakes world of enterprise planning.

00:05:31: And this entire operation... the LOM interfaces, the vector engine, the dual agents, it all runs on one foundation, the business technology platform, BTP.

00:05:40: We hear that mantra constantly, never without BTP.

00:05:43: So, for everyone listening, what is the BTP's actual function in this new AI world?

00:05:49: The BTP is the strategic heart.

00:05:50: I like to call it the innovation garage.

00:05:51: The innovation garage, okay.

00:05:52: It's

00:05:53: this dedicated cloud-based framework that sits right next to your core S-for-Hana system.

00:05:57: It can run on any of the big hyperscalers, AWS, Azure, Google, or on SAP's own cloud.

00:06:04: And its location, being adjacent to the core system, that's key to the whole clean core strategy, isn't it?

00:06:10: keeping the main ERP pristine.

00:06:12: Precisely.

00:06:13: BTP is what makes the clean core strategy a reality.

00:06:17: In the old days, if you needed a custom report or a special feature, you'd have to modify the core ER key code.

00:06:23: Which was a nightmare for updates.

00:06:25: A total nightmare.

00:06:26: Now, BTP lets you build all those custom extensions and AI tools outside the core.

00:06:31: You can innovate all you want without ever worrying that you're breaking your Sforhana system or making it impossible to upgrade.

00:06:38: So it's not just about moving the mess somewhere else, it's a governance layer.

00:06:41: That's a great way to put it.

00:06:42: Yeah.

00:06:43: And for AI specifically, the BTP is home to the generative AI hub.

00:06:47: Think of it as the official front door.

00:06:49: It's the integration point where you plug in all those pre-trained LLMs from OpenAI, LF-Alpha, and others.

00:06:56: It's the central switchboard for external intelligence.

00:06:59: Now, for all this strategic importance, our sources showed some, let's say, friction around BTP adoption.

00:07:05: DSAG says about twenty-four percent of its members are investing seriously, but there's a lot of talk about the cost.

00:07:10: It is.

00:07:11: Customers get the strategy.

00:07:12: They appreciate the clean core idea, but the cost structure, it's often seen as a problem.

00:07:18: The expenses for development, for quality assurance, just the general cost of running services on BTP can be high.

00:07:25: Especially if you're investing heavily before you have a clear productive application for it.

00:07:29: Right.

00:07:29: It's an investment in the future and stability, but that kind of foresight doesn't come cheap.

00:07:35: Okay.

00:07:35: Let's shift from the platform to the intelligence itself.

00:07:40: AI agents.

00:07:41: We hear these huge promises about autonomous agents that will manage whole business processes for us.

00:07:47: What's the reality check on that?

00:07:49: Well, the research from IDC shows there's a lot of market confusion right now.

00:07:53: You have something like forty-seven percent of organizations in Europe claiming they're deploying agents at scale.

00:07:58: Almost half.

00:07:59: But the tech underneath is still pretty immature.

00:08:01: Things like reliability, transparency, security.

00:08:04: They're not quite there yet.

00:08:06: The sweet spot for agents right now is in workflows that have very little or even zero automation to begin with.

00:08:12: If I'm a user, what does a jewel agent actually do?

00:08:14: How is it different from just a chatbot?

00:08:16: A simple chatbot answers a question.

00:08:18: A jewel agent executes a whole process.

00:08:21: It's like a mini bot that's been taught a specific multi-step business skill.

00:08:26: You mean an example?

00:08:27: Okay.

00:08:27: Instead of you manually creating a purchase requisition, sending it for approval, then starting the payment, you just tell a jewel agent, I need a hundred widgets.

00:08:35: The agent then checks inventory, finds the best supplier contract, creates the requisition in Ariba, handles all the data entry, gets the approval.

00:08:43: It does the whole sequence.

00:08:45: It performs the skill.

00:08:46: Yeah.

00:08:46: That really changes the scope of that thirty percent productivity goal.

00:08:49: It's not just about doing one step faster.

00:08:52: It's about eliminating the entire chain of manual steps.

00:08:55: And LLMs are being used for more than just text generation too.

00:08:58: Oh,

00:08:58: absolutely.

00:08:58: The real value is using them to securely analyze complex internal data.

00:09:03: An executive could use it to understand a convoluted supply chain process.

00:09:07: The legal team could use it to scan a thousand page contract for risks.

00:09:10: HR could use it to create personalized development plans.

00:09:13: Exactly.

00:09:14: Shorten onboarding, draft plans based on actual performance data.

00:09:17: The applications are huge.

00:09:19: And it's important to remember that a lot of the intelligence we see today is built on more, let's say, traditional machine learning or ML.

00:09:26: Precisely.

00:09:27: I mean, look at procurement in Ariba.

00:09:29: You have intelligent purchasing assistance, using ML to guide users, using semantic search to find the right item, and checking it against contracts in real time, or invoice processing.

00:09:40: The OCR engines that read invoices are just sophisticated ML models trained on millions of documents to handle any format you throw at them.

00:09:48: It's what replaces all that manual data entry.

00:09:50: So with all this new stuff, what about the older analytical tools?

00:09:55: For customers deep in the HKANA ecosystem, is something like the Anapredictive Analytics Library, PAL, still a thing?

00:10:02: It absolutely is.

00:10:03: The source material is very clear that the Anapredictive Analytics Library, PAL, is still a core part of the ANA database.

00:10:10: While the big trend is definitely toward BTP and external LLMs for that broad generative intelligence, PL is still vital for classic statistical predictions that need to run super close to the core data.

00:10:21: This

00:10:21: brings us to a really difficult point.

00:10:23: where strategy hits the pavement.

00:10:25: For existing SAP customers, this whole AI transition is tangled up with the S-for-anta migration, and our source has described that migration as, well, exhausting.

00:10:36: The complexity is just off the charts.

00:10:38: That's why you see so much reluctance from customers.

00:10:40: The migration is seen as expensive, time-consuming, and a huge drain on resources.

00:10:45: But the real problem is that most companies don't know how to turn their old classic processes into intelligent, AI-ready workflows.

00:10:53: They

00:10:53: know what they have, but they don't know how to make it smart.

00:10:56: Exactly.

00:10:57: And then you have the data problem.

00:10:58: What do you do with twenty years of business history?

00:11:00: You just drag it all into your new S-Force system.

00:11:03: This is one of the most controversial points we found.

00:11:06: The risk of moving all your legacy data is that you import what experts call the dependencies of the past.

00:11:11: Old logic, messy data structures.

00:11:14: All

00:11:14: of it.

00:11:14: Right into your pristine new S-IV system.

00:11:17: To avoid this, experts like Thomas Failer are pushing a new idea.

00:11:21: Separate instead of migrate.

00:11:23: Wait.

00:11:23: Separate instead of migrate?

00:11:25: That goes against everything we've been told about ERP for years.

00:11:28: So you just... leave the old data behind.

00:11:30: Not exactly.

00:11:31: You don't bring it into the operational core.

00:11:33: The recommendation is to extract all of your legacy data, everything, including the deep ADK archives, and park it unchanged on a separate high-performance platform.

00:11:44: This lets your new S-Force system start completely clean with AI-ready structures while you still have all your historical data on the side for reporting and analysis.

00:11:53: It's a radical choice, but it might be the only way to get a truly clean core.

00:11:57: And as if that wasn't complicated enough, the licensing model for S-IV is a whole new world of complexity.

00:12:04: It

00:12:04: really is.

00:12:05: The shift to usage-based licensing, especially with programs like Rise, creates huge new risks.

00:12:11: Customers are used to paying based on user authorizations.

00:12:14: Now they have to pay based on transaction volumes or how much of a specific service they consume.

00:12:18: Which

00:12:19: means if you get your forecast wrong, you could be paying for a lot of software you don't use.

00:12:22: That's

00:12:23: the risk of shelf wear.

00:12:24: And it's a big one because accurately predicting usage this new ecosystem is incredibly difficult.

00:12:29: Let's

00:12:30: widen the lens for a second and look at SAP's ecosystem.

00:12:33: Who are they partnering with beyond the big LLM providers, especially around their data platform, SAP DataSphere?

00:12:40: They've announced some major data partnerships.

00:12:42: The big names are Calibra, Confluent, Databricks, and DataRodain.

00:12:46: The whole point of these collaborations is to help you build a single unified data landscape that can seamlessly and securely blend your core SAP data with data from all your other third-party systems.

00:12:58: That partnership with Databricks seems particularly important for anyone doing serious custom AI work.

00:13:04: It is.

00:13:05: That one is all about making it easier to train your own custom ML models using the Databricks platform, but keeping them grounded in the context and governance of your SAP data.

00:13:15: We're also seeing a big push for flexibility through open source.

00:13:18: Suze, for example, has an AI program focused on letting you use secure open-course GenAI models, which gives you more choice and control over your data privacy.

00:13:26: And speaking of privacy, we have to talk about the geopolitical elephant in the room.

00:13:30: Data sovereignty, especially when you're using these massive US-based cloud hyperscalers.

00:13:35: This is a huge regulatory minefield that every customer needs to understand.

00:13:41: The US Cloud Act basically changes the rules of the game.

00:13:45: It says that U.S.

00:13:45: authorities can demand data from a U.S.

00:13:47: cloud provider no matter where in the world that data is physically stored.

00:13:51: Even if it's in a server in Frankfurt.

00:13:53: Even if it's in a top security EU data center.

00:13:57: Your European data center effectively becomes a branch office of a U.S.

00:14:01: company subject to U.S.

00:14:02: law.

00:14:03: And Europe's response to this is the EU Data Act.

00:14:05: Exactly.

00:14:06: The data act was adopted in mid- Twenty-Twenty-three and it's the core of Europe's digital strategy.

00:14:12: It brings in a whole set of new rules about data portability and interoperability.

00:14:16: The goal is to make it easier to move your data to prevent vendor lock-in and really to strengthen Europe's digital independence.

00:14:23: It's a big compliance headache for companies at first, but it's seen as essential for the future.

00:14:27: So when you put this all together, what does it mean for you?

00:14:30: SAP is clearly positioning the BTP as the one indispensable technological foundation for everything.

00:14:36: The Gen AI Hub and the Vector Engine are the tools they've built to securely deliver on that huge thirty percent productivity promise.

00:14:43: But the thing is, the success of this whole strategy depends less on SAP's code and more on your willingness as a customer to make these big strategic moves yourself to truly embrace the clean core and to get your arms around those really complex data migration and licensing challenges.

00:14:58: We started this deep dive by talking about the Gutenberg moment, this massive disruptive change, and contrasting it with SAP's more evolutionary approach.

00:15:05: So that leaves us with one final provocative thought for you to consider.

00:15:09: If GNI really is changing the fundamentals of IT, will SAP's current incremental strategy of building frameworks and partnerships be enough to deliver that revolutionary thirty percent leap?

00:15:19: Or does a revolution to this scale demand that missing comprehensive master plan?

00:15:24: And maybe more importantly, what's the one step you should take right now?

00:15:27: perhaps by seriously considering that separate, instead of migrate idea, to make sure your own data infrastructure is actually clean and ready for this new AI First World.

00:15:35: This was an episode based on the cover story from the October, twenty-twenty-five issue of E Three Magazine on the topic of AI with ERP, BTP and BDC.

00:15:49: The SAP Business Technology Platform, BTP, and SAP Business Data Cloud, BDC, are essential interface for all SAP AI applications.

00:16:00: The topics of Business AI and Gen AI Hub of SAP BTP will also be covered at the two SAP Community Summits on April twenty-second and twenty-third twenty-twenty-six in Heidelberg, Germany, and on June tenth and eleventh twenty-twenty-six in Salzburg, Austria.

00:16:18: We look forward to welcoming you in person at these two SAP Community Summits.

00:16:23: Secure your ticket now on the E-three magazine website E-threemag.com.

00:16:29: Thank you for your interest.

00:16:31: Best regards from the E-three magazine editorial team.

00:16:34: See you in the next episode of the E-three Cover Story podcast.