What Will Generative AI Really Mean For SAP Customers?

Danielle Chung, ANZSAP Magazine & Pavlos Panagiotidis, SAP

As SAP’s BTP Innovation Days showcase rolls around Australia, ANZSAP Magazine grabbed a chat with SAP’s Global Head of BTP AI and Automation, Pavlos Panagiotidis, and SAP Technology Ambassador and Solutions Advisor, Murali Shanmugham, to find out how the technology that’s the biggest buzzword since the arrival of the Internet is going to impact you and your enterprise.

To begin with, Pavlos points, out, the impact will depend on who you ask. For teachers marking homework, Generative AI already represents a clear and present to threat to traditional learning systems, while workers in process-driven occupations also look on in trepidation for the future of their livelihood.

In the business of creating next-gen enterprise-level process and business solutions, there’s plenty of bullishness about the many possibilities of Gen AI. Yet within SAP, embracing that opportunity is not so much about pressing the accelerator to the floor as it is about gently tapping the brakes.

“(SAP) are not the earliest technology adopters, because we have responsibility for very important business systems,” Pavlos says. “Therefore, there is a small lag between the new technology coming to the market, and us making sure we can find the right use for it within corporate systems that can be trusted and can scale.”

It seems that SAP’s caution is warranted and necessary. Already, instances have emerged of chatbot applications going rogue – or “hallucinating”. This refers to the innate need for AI to provide an answer to any given question, without having the proper preparation, training and data to arrive at the right answer.

“So generative AI can indeed mimic a lot of human activities based on the data available, but it needs a lot of grounding,” Pavlos says. “If your chatbot hallucinates a little bit and it's affecting a business system, you cannot really trust it. So that's where we are investing a lot with SAP, and we are still at the beginning.”

Part of his role is to work with customers to help them understand SAP’s offerings and listen to their feedback, while also running teams of experts around AI development. “This customer feedback is a big part of our job – it's actually the most important, in my point of view,” he says.

For Murali, whose job is focused on driving adoption of BTP, customer feedback is also king. “Our goal is to work with customers to see what their business challenges are and then see how technology could be used to help solve those challenges,” he says.

“What we're seeing with AI is a lot of customers embracing it. They're looking for opportunities to be able to use it - a competitive advantage. It could be around improving operational efficiency, or delivering exceptional customer experience.

“The key challenge is, how do we operationalise AI and deliver business value? And I think that's where customers are looking for a good technology partner to work alongside them to help realise the benefits.”

It is inevitable that Gen AI will transform the business systems landscape. An early indicator of its progress can already be seen via the rollout of SAP’s new AI-driven chatbot assistant, Joule, which presenters have been recommending to attendees at SAP’s BTP Innovation Days series in Brisbane, Sydney, Melbourne and Auckland, with further events scheduled in May for Perth and Wellington.

Demonstrations showed Joule being able to perform intra-database research in seconds that would take a human significantly longer, and to instantly aggregate data to generate reports or recommendations.

“An area where we see a lot of investment is user productivity. Whether an end user or a casual user of a solution, these users could definitely benefit from using a digital assistant, instead of them having to manually go and find the right applications to use or to conduct transactions themselves,” Pavlos says.

But he maintains that SAP is firmly committed to utilising the tech solely to develop user-focused platforms and processes that directly address their customers’ requirements.

“The biggest issue is actually finding the right use case, one where it can improve something, versus just following the bus,” Pavlos says.

“We’re looking for things we can do better, focusing on things that grow the business, freeing up our customers to interact with the public instead of just doing administrative tasks.

“So I think over the course of next year, we're going to see the first really enterprise level solutions that make use of it. But still we are in the very beginning and that’s where we are investing a lot.”

Pavlos also cautions customers not to rush too quickly into building Gen AI-based custom solutions. “Many customers are excited and try to build something that meets their needs. But actually if they wait for a year or so, the software vendors that they use will probably deliver those out of the box,” he says.

“If it's not a market-maker for them, like embedding AI into their own products or whatever, if it is what we call a situational build, that is difficult. Most probably, one of their software renderers will deliver it out of the box, so it's good to assess and wait.”

While some occupations will inevitably be made redundant by Gen AI, talented developers and software engineers may see their value soar, Murali predicts. “I remember a tweet by (American software engineer) Kent Beck who said, ‘the value of 90% of my skills just dropped to zero, and the leverage for the remaining 10% went up 1000 times’. It's time to recalibrate and see again where we need to hone our skills, and be able to focus more in those aspects.”

As the technology evolves, so will the ways in which we interact with it. Pavlos expects that the user interface as we know it may soon cease to exist. “Especially for a casual user of a system, they will interact with natural language,” he predicts.

“But at the same time, if you want to go beyond just having a conversation or some summarisation of information, and achieve really large-scale business automation, you stumble upon the limitations of AI. These are not easy to fix in one year, or even two.”

A key to overcoming these roadblocks – and avoiding Gen AI “hallucinations” – will be ensuring that AI actions are grounded in facts. “This is where SAP is going, we’re introducing our own knowledge graph, auto-populated by all the SAP entities, so you can ask a question that is sitting four or five hops away from the answer, and it is able to go back and find the right answer,” Pavlos says.

“We are also trying to build this out of the box, to allow our customers to enhance those graphs so they can get their own business-specific information in there.”

SAP is also researching its own “foundation model” for Gen AI based on business knowledge embedded in its own applications. “We also have access to 25,000 tenants of SAP software on the Cloud through our customers, who’ve allowed us, based on our Cloud agreement, to have anonymised access to their structured data,” Pavlos says.

“We want to ground AI into business facts and relationships around the SAP systems, so that it gives a credible answer and not one that hallucinates.”

SAP further differentiates itself from other software vendors, Pavlos says, by configuring its GenAI tools to reach solutions by referencing business relationships in addition to available data. This measure can further mitigate the potential for data-based ‘hallucination’ to occur.

Pavlos gives the example of a user asking Joule if an order will be delivered on time. Joule responds that the order has been delayed by 72 hours, but also recommends that because the order is a critical component of another product being delivered in 48 hours, sales plans should be adjusted accordingly.

“It’s obvious such an answer cannot be given by any GenAI-based assistant, even if it has access to all the data, if it doesn’t have access to the relationships behind it that SAP has,” Pavlos says.

SAP’s need to ensure its Gen AI-based systems are robust and reliable means the rollout of products will take a little longer than many customers may envisage. But the mark of success will be a seamless end user experience, Pavlos predicts.

“The way to productise it is to make it part of a business process so that the end user in some cases does not even know that it is AI-based. That is the SAP strategy, to try to make it simpler to productise and operationalise AI, so that the end user actually does not even know or care.”

Pavlos and Murali’s five top tips for getting started in business AI:

1. Build awareness – ensure that experts within your business understand the capabilities and not just the hype around AI, so you can identify good use cases for your specific needs.

2. Collaborate – create a team combining people who understand AI with those who understand the business’s challenges, to identify valuable use cases.

3. Explore – go to the SAP Business AI web page to learn about applications and road maps, to help choose whether to use an out-of-the-box solution for instant results, or to implement a customised application for a specific use case.

4. Prepare – ensure any AI-based plan of action to address specific use cases can be supported by high-quality data.

5. Implement – start small by implementing simple out-of-the-box AI solutions in areas such as customer service or productivity enhancement, before taking on larger systemic changes that carry a higher level of risk.