Welcome back to Nerd Exchange, a new ASUG column from Jelena Perfiljeva and Paul Modderman, The Boring Enterprise Nerds. Helping subscribers stay on top of SAP, AI, Cloud, and enterprise software news through their hit Boring Enterprise Nerdletter, Jelena and Paul are funny, cynical, and always interesting—a breath of fresh air for the enterprise software landscape. Below, the Boring Enterprise Nerds offer their take on artificial intelligence (AI) in the SAP user ecosystem.
Jelena Perfiljeva: Everybody talks about AI these days. It’s important. But it’s also frankly overwhelming, and frustration about all the sensationalism and hype builds up. How can we better see what’s important and valuable behind smoke and mirrors? Paul, any advice?
Paul Modderman: It’s been such an overwhelming couple of years since generative AI became part of my everyday work process. It’s so hard to keep up with the news, the promises, and the frustrations that have come along with this light-speed pace of change. If I were to summarize my experienced, enthusiastic, and exhausted perspective in one piece of advice, it would be: Don’t expect magic.
Jelena: I agree! SAP Build Code is a great example of that. I've been able to try it out, and I've got to say that if you want to get real value out of it, you need to understand what you’re doing. And remember I showed you my ChatGPT experiment with “give me an ABAP report”? The results were all over the place but to an untrained eye they probably look fantastic.
Paul: Right now, the people finding the most success with this new technology are those using generative AI tools like helpful junior collaborators, rather than effortless order-takers who can complete and submit finished work autonomously. We've spoken to ABAP developers who use GitHub Copilot (I use it all the time), and the AI assisting with small-to-medium-sized chunks of predictive code is a boon to productivity.
Jelena: We mentioned magic already, and I feel that too many times there is an assumption that “we can just give this to AI, and it will figure it out.” But is there even data available to train that AI tool? Is the data reliable and properly structured? This subject is somehow frequently overlooked in AI conversations.
Paul: Data in AI is a subject I get nitpicky about. If you're looking specifically at generative AI, then you'll need data for most enterprise tools. A common use case is augmenting the prompts you send to the generative AI with corporate documentation that contains supporting information. You might've heard this referred to as Retrieval Augmented Generation, or RAG. That information is then semantically unpacked by the AI and used to create meaningful responses.
That is different from the gobs of structured data that you would use to create a traditional machine learning solution, for forecasting sales or predicting machine maintenance needs from sensor data. That usually involves umpteen million data points that can be attacked with specific algorithms to mine for insights. As of right now, you can't stuff millions of records into a prompt to your corporate ChatGPT and ask for a predictive analysis. That will have to be handed off to a system designed for that purpose.
However, I predict that these lines will blur. Generative AI still can't directly ingest those huge tables, but some solutions are getting better at understanding when to hand over tasks to non-AI systems. I think you'll ask generative AI for a deep data analysis, and it will then decide how to parcel that out to structured systems. Those structured systems will still require lots of effort by developers, administrators, and analysts.
Jelena: Sounds like AI is not going to take our developer jobs just yet? That’s a relief! :) So, what would be your practical suggestions for SAP customers who want to “get into AI,” as it were?
Paul: I think there are two paths to dig in: first, enhancing a business process with a delivered AI solution like we see in SAP SuccessFactors, which can create targeted job descriptions or help employees upskill with personalized, AI generated recommendations. That's going to be a process challenge, but I think it's worth the effort. The second is getting business users accustomed to interacting with generative AI, and how they can best collaborate with the tools. I have seen many people start out not knowing what to do with a blank chat prompt. It takes time to get acclimated.
If you're interested in making a difference with generative AI in SAP environments, I recommend getting familiar with SAP Datasphere. SAP data will play a vital role in many AI data initiatives, and Datasphere is a good way for SAP-centric enterprises to stitch together SAP and non-SAP data threads.
Jelena: Turns out “now with 25% more AI” comes with 100% more learning! Take "prompt engineering", for instance. How do you articulate what you want in a way the AI understands? How do you even know the right questions to ask? It’s like learning another language. The lack of resources around this is a real issue for enterprise use cases. I think a “prompt library” for SAP professionals could be helpful.
Paul: SAP Joule is getting lots of love and attention, especially coming out of SAP Sapphire. SAP should maintain this momentum by flooding the internet with examples and videos of Joule usage. The more that people see generative AI in action, the more they will bring their own ideas to the table with regard to its usage.
Jelena: Maybe we can close with a controversy? Some enterprises seem to be missing the forest looking for AI trees. It’s not a coincidence that quite a few solutions that are sold as “AI” hardly include any AI at all. Good old automation and integration might not sound as exciting, but they can create real business value, today. Keep an eye on the AI prize, but don’t forget to collect power-ups along the way.