Generative AI is surging, in a dizzying blur of techno-propaganda and possibility. Potential use cases are everywhere (and need enterprise scrutiny).
Questions about managing risk - and the implications for customer data - are surging also, thanks to terms of service missteps by Zoom (see: Enterprise hits and misses - Zoom stirs an AI privacy controversy).
In SAP's case, I've already shared my early look at SAP's generative AI for HCM plans: Sapphire Orlando '23 - are SAP's AI ambitions in line with customer priorities? Inside the SAP SuccessFactors generative AI news. Given the release timeframes and fall event schedule, I wasn't expecting notable AI updates from SAP until TechEd in November. However, I recently had the opportunity to get an update on SAP's AI strategy, via an online briefing with Thomas Saueressig, Member of the Executive Board of SAP SE. Saueressig leads the Board area for SAP Product Engineering; he's the right person to press the AI questions with.
Readers know that I have strong/outspoken views on generative AI. In this case, I'm going to hold my commentary to the end, and start by relaying Saueressig's news and views.
How disruptive is generative AI in an SAP context?
One of my biggest irritations with the generative AI PR festival? Most enterprise software vendors won't ship generative AI functionality until the end of the year. Meanwhile, more mature AI use cases, ready for customers to assess/consume now, get nudged aside. I expressed that irritation to Saueressig in our pre-chat email; that's where our conversation started. As Saueressig told me:
As you phrased it, it's not just only about generative AI... AI is broader, and the reality is: SAP has done AI for a couple of years now. As of today, we have 130 available AI services in our portfolio, so it's already 130 use cases. Just to give you some context, we currently have approximately 26,000 customers using one of these scenarios.
But as for generative AI, Saueressig does believe it will be fundamentally "disruptive" to the enterprise software market:
I truly believe generative AI is disruptive to the market. But also vice versa, it actually means that SAP needs to be able to disrupt ourselves in order to stay competitive in the future. That's exactly where the leadership team and I will talk intensively about how we make use of this technology to really rethink software. It's nicely fitting into our strategy we talked about at Sapphire and before: 'What's the next generation of enterprise software'? For sure, AI was already there. It's one of the key components of how software needs to be radically different in the future.
I've already floated my view that generative AI will be disruptive to enterprise software product categories, licensing, and therefore Magic Quadrant type software thinking, which I believe is mostly legacy thinking (I'm not just picking on Gartner here; many other firms have similar products). Hearing Saueressig acknowledge the potential for internal disruption via generative AI was interesting; I'll return to that shortly.
How will SAP differentiate in a generative AI context?
When it comes to generative AI options, SAP customers will have plenty of choices, including a range of brash new AI vendors. So what makes SAP's approach to generative AI different - or uniquely valuable to its customers? It comes down to SAP's push for Business AI. That means leaving the technical "foundational" aspects to SAP's trusted technology partners:
We want to differentiate on the application side of the house, to embed AI naturally in the software. We don't want to compete on foundational technology for AI. That's where we actually as a company open up, and say 'We want to partner.' That's the reason why we actually partner with many companies like Microsoft, like OpenAI, Aleph Alpha, Cohere, Anthropic, IBM, Google - you name it. [Author's note: as per the link above, SAP has invested in several AI/LLM enterprise startups].
Saueressig says this will enable SAP to set the pace on the business side of AI:
Our approach is we need to take the best and greatest tech which is available - and advancing very quickly - and bring it into the context of our applications, into the business processes, into our business data, and make it super relevant for customers - and reliable. I think that's where our differentiation comes in... Everything we provide with AI needs to be highly reliable, highly relevant, and absolutely responsible and trustworthy.
Saueressig referred to a "unifying layer" across all SAP products, with AI announcements coming at SuccessConnect and Spend Connect this fall: "We will embed generative AI in every product." He doesn't think SAP's competitors can keep up:
You will see a lot of our competitors saying similar things. Where we believe we hugely differentiate in the market is how we do it. That also includes an SAP foundational model. That's actually a big thing. Think about all of our applications, with all the datasets which we have, all the table structures which we have, to bring that into a huge SAP foundational model together, to have this wisdom and knowledge of the entire suite which we carry.
The impact of customer data on enterprise AI
During SAP's July 2023 earnings call, SAP CEO Christian Klein made some intriguing statements about SAP's AI data advantage. During our briefing, Saueressig elaborated on that:
As of today, we have more than 20,000 customers who gave us consent that we use that data for showing in an anonymized fashion, that we use that data to train models - and that's part of our contract... That's the reason we can build up this huge SAP foundational model - most probably 200,000 database tables, to give you some sense of the scale.
But Saueressig says how customer data will be used in real-time is SAP's true AI edge:
We have an AI foundational layer in BTP; we automatically use the prompts that get sent to our AI [systems]. Alongside, we leverage our foundational layer to make it more reliable and relevant for the customer. But we will also pass the prompt via an extension of HANA, which we extend to a vector database. From the same prompt, you get the real-time data of the customer itself, from this specific customer, to bring it into that context as well.
That makes it for sure, super-relevant, because then if you have a question like, 'Will our customer ABC, pay in two days from now?' You take the wisdom of the crowd and all the connections to all of our data models... But you see also the payment history of this specific customer of SAP with their customer, and you can bring that into the context as well, which means you can really rely on the outcome because it has a real data set behind it. And then, for sure, we will protect it. That's an important piece: an ethical layer, which is part of our AI foundation.
Readers may (rightly) ask: given most generative AI functionality won't debut in general availability until the end of this year, what the heck are we going to talk about during the fall event season - AI powder puffs and marketing confections?
My personal goal - dig much further into the specifics:
- How will vendors train their LLMs?
- Will they use third party LLMs as well?
- How will customer data be protected, if that data is crunched by a third party LLM?
- Will customers embrace the use of their data as aggregated data via existing terms of service agreements, or will they want to revisit those opt-ins?
- Will vendors 'pull a Zoom' and try to adjust their terms of service on the fly to enable more model training, and will customers push back - especially if their perceived IP is part of that training?
- How will explainability, one of the most important missing pieces in AI so far, be improved?
- How will AI systems not only minimize their own biases, but be used to challenge human bias patterns inside companies?
So far, vendors have provided unsatisfying answers to these types of 'behind the curtain' questions; that needs to change, and soon. I expect SAP to have interesting and solid answers - but it will take some time to get to all of them. For his part, Saueressig did shed important light on the role of opt-in/anonymized customer data, and the real-time integration of that data into SAP's AI platform. He also said that at SAP's fall events, we'll hear more about SAP's advances with explainability, which he sees as a non-negotiable issue for earning the AI trust of business users.
I question one aspect of SAP's current AI/innovation strategy as currently stated, and strongly disagree with another. The part I question comes down to what we heard on the earnings call about SAP charging a 30 percent premium for generative AI capabilities.
I believe that to maximize the accuracy and relevance of generative AI for enterprises, there will need to be three factors:
- aggregated customer or industry data added to the LLM.
- customer-specific data added to the LLM.
- further training and tuning of that data, utilizing the customers' own domain experts likely with some type of reinforcement learning or iterative tuning of the model.
Saueressig mentioned the first two as core to SAP's approach; we didn't have time to get into the third. SAP's core ingredients for effective generative AI are there, but will customers embrace paying a premium for systems that they themselves helped create the value in, via their own data and model refinement?
Yes, enterprise vendors will need to increase software pricing for generative AI, due to the costs entailed in administering these systems. But three years from now, I believe all enterprise customers will expect - as long as they are on a current release - that AI capabilities are fully embedded and included in the core release (others may be offered as paid add-ons, but third party apps will always be a factor).
SAP is not the only vendor to pursue these premium strategies. But my view is that AI-done-right will be about delighting, retaining and growing all customers, not reserving innovation for premium-only. The cost for a customer of being left behind is too great. But then again, this reflects my view that successful generative AI won't be a packaged solution, but a deep collaboration between customer and vendor.
The labor requirement on the customer side to achieve strong generative AI results is precisely why I refer to generative AI as an 'evolutionary,' not a 'revolutionary' technology, aggravating tech marketers the world over in the process. If I am wrong about that, then I am probably wrong about SAP's premium pricing plans as well. We'll see.
During the earnings call, Christian Klein also said that S/4HANA Cloud was the only way users can access innovations for AI and sustainability. That includes the "green ledger" and generative AI. But Klein also emphasized that those innovations could only be accessed via RISE or GROW.
The outcry and controversy was immediate. During our interview, Saueressig re-iterated these points. I strongly disagree with this, though at the moment it appears to me as SAP market messaging, not yet rigid contractual policy.
There are two aspects of this to unpack. One is the RISE and GROW requirement. This is the aspect I strongly take issue with. Though I do believe these programs can have value for customers (I have documented some of the use cases on diginomica, including Hilti), I believe these hyperscaler management programs should be customer choice. SAP can make the case, as Saueressig did during our call, that SAP can best manage those hyperscaler relationships, and that RISE/GROW will get the most out of SAP's AI and green innovations.
But: requiring customers to go through RISE and GROW to access innovation raises big questions. Since our call, I've researched this. I have yet to find any technical reason that require RISE or GROW to access new innovations. I have a hard time imagining SAP refusing to sell access to premium AI solutions to customers willing to pay, even if they are not in RISE or GROW.
Some of SAP's big partners have their own hyperscaler management programs - does it even work to require SIs to move all those customers to RISE or GROW? SAP's smaller partners certainly aren't going to force fit RISE or GROW onto all their customers. In my view, this kind of talk would make a customer less enthusiastic about doing an on-prem or private cloud S/4HANA upgrade. Why would SAP want to disincentivize major upgrades, by implying S/4HANA is not a defacto innovation platform, not when SAP can continue to make the RISE pitch after the S/4HANA move is made?
I realize that SAP's messaging on this played well on Wall Street, but Wall Street will learn soon enough: the enterprise market is trending in a different direction entirely. I see winning vendors:
- Provide as much customer choice for modernization as possible.
- Give as much access to innovation and cloud services to customers on older releases as possible.
Why? Because giving customers access to modern cloud and AI services, even while on older releases, deepens their adoption and buy-in - and greatly reduces the risk they will procure innovation solutions from competitors.
I believe SAP is going to rethink some of these positions, as SAP has closer dialogue with stakeholders and user groups this fall. DSAG, the German-speaking SAP user group, viewed this as unsettling: On-premise customers cut off from innovations? ASUG CEO Geoff Scott's initial analysis provided useful context to SAP's announcements. My view: SAP can still make an aggressive case for RISE and GROW, without creating confusion about why those programs are mandatory to receive innovations.
In his LinkedIn post, analyst Josh Greenbaum pointed out how SAP innovation can, in theory, be accessed across S/4HANA editions:
A look at the different roadmaps for S/4HANA public edition and S/4HANA private edition make it clear there is innovation coming in AI and sustainability for both editions. ECC hasn't been slated for net new innovation as far as I know for a number of years, but as S/4 private edition can be run on-premise, the term on-premise itself becomes confusing. But innovation is definitely in the plans for S/4 private edition.
A quick look at the private and public edition roadmaps shows that private edition has 11 new features planned through 2025 under the category of AI, vs. only 6 for public edition. Sustainability in public edition has 27 innovations in the roadmap, vs. 35 for private edition. So.. it's clear some degree of innovation in these categories is available to private edition. I also believe that customers running hybrid ECC/on-premise and S/4 public or private cloud can access the AI and sustainability features in S/4 via BTP. I admit that's more of an educated guess, it's sometimes hard to sort through how the three products interact.
Greenbaum's take lines up pretty well with my position: for now, this is about SAP market messaging, not enacted policy. Therefore, there is time for SAP to alter its messaging, before it becomes policy. Now, the cloud versus on-prem issue is different. Across vendors, there is such a thing as functionality, particularly around AI, that can either only be accessed via public cloud, or public cloud services.
And, with new innovations, there are always going to be some tech requirements. Example: Saueressig noted during our call that those real-time customer AI data features have dependencies with certain versions of S/4HANA (I'll look to get specifics during the TechEd timeframe). Obviously, version dependencies can be unavoidable, but in general, if I'm a vendor, I want to make sure as many customers can get access, one way or the other.
In my interactions with DSAG leadership, I've found they are open to give and take on the issue of cloud-dependent or cloud-only innovation. They might not agree, but they will listen and debate it. If I'm SAP, I would reach a firm/clear understanding with users groups such as DSAG and ASUG on which innovations are tied to cloud releases, and exactly why, and what the options are. Where technically possible (and sustainable without custom code), shouldn't any customer that is on S/4HANA have access? If on-prem customers can also access these innovations as cloud services, as SAP seems to be wisely pursuing with their Industry Cloud strategy, why not? These kinds of user group conversations are SAP at its best; they are needed right now. Make no mistake - this is a pivotal moment.
The most interesting part of our talk, for me, was hearing Saueressig discuss how generative AI is disruptive to enterprise software, including SAP. I'm out of space for much detail on that, but this captures the spirit of Saueressig's views:
We need to be able to disrupt ourselves... We need to see how we can we free up all the capacity, that we focus on this, so that we can think differently... I don't want to just take an existing process and put it on steroids with AI. That's also what we do. But that's not sufficient. Sometimes I believe, based on AI, you can rethink the entire process.
Though this was an AI conversation, Saueressig believes we need to have a broader view. Obviously, he is feeling vindicated about SAP's ESG software pursuits, given the surge of ESG requirements, particularly in Europe, and the emergence of next-gen ESG software as a category. But he is thinking bigger:
If you think about next generation of enterprise software, it's not purely about the technology aspect of AI, but also, quite frankly, about aspects like sustainability. We should not forget about collaboration. And the network aspect, as we discussed - the global network of enterprises coming together.
That's a conversation we can't sort today, but we should return to it.
Updated, 7am UK time, with several clarifications for readability and a few more resource links. I also added a couple of points on terms of service questions to my broader list of generative AI questions for any software vendor.