Main content

Sapphire Barcelona – SAP customers share challenges with rolling out AI with older platforms and siloed data

Madeline Bennett Profile picture for user Madeline Bennett June 13, 2024
Customers are always the best advocates for technology. Gen AI's potential is real, but it's going to take time to catch up with the hype cycle.

Customers have their say...

While SAP was busy at its Sapphire event in Barcelona this week sharing more detail on the new features of its AI strategy, customers at the show opened up about the challenges they’re facing around using AI, with many only at the very earliest stages of their journey.

At Sapphire Orlando earlier this month, SAP announced the integration of its generative AI technology into more of its applications, along with additional capabilities for its copilot Joule, including the ability to communicate in languages other than English like Spanish and German, starting from July.

At the event, SAP also announced the integration of Joule with Microsoft Copilot for Microsoft 365; new or expanded AI and LLM (Large Language Model) partnerships with industry players like Nvidia, AWS and Meta; and the $1.5 billion acquisition of digital adoption platform WalkMe, which again SAP plans to integrate with Joule.

During a panel in Barcelona at this week’s European event, SAP CEO Christian Klein highlighted Joule’s integration within SAP core systems as a key differentiator between its own and rival copilots. It's not only a new interface, but it's deeply integrated into SAP products at the back end, avoiding the need for IT teams to integrate it within their systems, he explained:

When you are working with an SAP system in HR and procurement, your IT department doesn't need to integrate the copilot to do a job posting or to do sourcing. When you are using SAP, it comes out of the box, you're using Joule, you activate it and here we go. You can create job postings, you can do sourcing, you can do financial accounting postings in the general ledger, out of the box.

Klein also outlined the advantage of having bidirectional integration between Joule and Microsoft 365. If you ask Joule an analytical question about workforce data, for example what kind of profiles the company has, some of that data might be sitting in Microsoft 365 in an email that adds some context to those workforce profiles. He added:

The good piece about building this deep integration bidirectional, you can ask Joule and you get access to this Microsoft 365 data so that you get all your employee skill data and not only the one sitting maybe in SuccessFactors.

Muhammad Alam, who leads SAP's product engineering, took the opportunity to promote SAP’s approach to LLMs, and its partnerships with other tech providers in this area. As part of its underlying AI tech, Alam said SAP continuously tests and validates which large language model is performing best and automatically selects the right one for customers, ensuring they get the most reliable result. There is an option for customers to override this and make this own selection, but all within the SAP system. He added:

Where the choices come into play is, if a customer's also building alongside those some custom scenarios, then they can pick which they think is best for them without having to go deal with multiple companies out there as well.


While SAP is keen to pitch the potential for its embedded AI capabilities, customers aren’t necessarily ready for it just yet. Speaking at the show, Gordon Smit, CIO at lingerie brand Hunkemöller, acknowledged that the future is definitely AI, but added:

It is a long way for us to get there.

To fully benefit from AI, businesses first need to have their data in order, something Hunkemöller is still working on. Smit explained:

Running on an installation that was installed in 2009 and software that was developed in 2004, when the Nokia 3310 was the hottest phone you could have - and we're still running on that - it was very difficult for us to get a start with AI.

Hunkemöller started with aspects like personalization and segmentation, an area where they had the necessary data as 75% of its customers are members.

Today, if you talk about personalization, that's already a commodity for us, it's a day-to-day business having AI implemented there.

When it comes to where generative AI could help the business, Hunkemöller is interested in product design.

We are a brand, so we design our products ourselves. How can generative AI help us with that, for example, creating the mood boards with gen AI instead of people scraping the internet or going through magazines to find the right pictures and put them on a big board? If we can automate our design process with generative AI and it's not by just scraping the internet, it really generates new pictures, creates a mood board, immediately creates a pattern from the products in the picture, and at that moment you also have all your pictures ready immediately for your website for all your sales channels, our time to market will speed up by at least 75%.

Hunkemöller is also keen to explore how AI can help with pricing, determining the right price and when to mark down. Smit explained:

That’s very important in retail. How can we make sure we sell at the highest price and get the highest margin? These are interesting steps for us in the coming year.

UK supermarket chain Co-op also has work to do before it can take full advantage of AI. Ian Cox, the firm’s Director of Technology-SAP, said:

As a retailer, we're probably not where we would like to be with AI. It’s a fast-moving environment, we know we need to get there.

While Co-op is heavily focused on the technology, at the moment it’s more around concepts and where it might be able to support the retailer – and some heavy data work is required first. Cox added:

We have a big focus on data, structured data or unstructured data, cleaning that up, getting that into the place where it can help us

The business has recently signed up to work with SAP on a major retail transformation project, deploying Datasphere to clean up its data. Cox said:

From a retail perspective, that’s our first step. We're also learning about AI and the data around it with some of our other smaller businesses. For example, in our legal services, we've got chat bots running and case management through that. We're learning a bit about responses and data that way.

Once Co-op is underway with its Datasphere retail transformation, Cox sees a big future for the retail business using data to exploit sales forecasting, better inventory management, more predictive forecasting, and more accurate reporting and results.

Pricing and promotions is an area where AI could really help the Co-op, Cox said:

We tend to run quite a lot of promotions that often overlap and our pricing strategy needs quite a bit of work. Getting the right promotion at the right time and the right pricing strategy timed with things like the weather and other events that are going on, that's an area where we will really focus. That's going to be the big focus over the next 12 months.

BayWa is further ahead in its AI journey. The agriculture, construction and renewable energy supplier has already implemented a use case in its machinery department, developing its own machine learning algorithm to calculate how much to pay to farmers for agriculture equipment based on information, pictures and the market situation.

BayWa now wants to move away from developing its own AI models to using and consuming them out the box. It already has some pilots using SAP AI technology, for example, selling second-hand equipment. Lars Pischke, Global Enterprise Architect at BayWa, said:

The configuration of used machines is just plain text. It’s complicated for sales staff to sell these machines because we have hundreds in stock. To find the right one for the requirement of our customers, it's very difficult and time consuming. That's why we have a big stock of machines.

Pischke welcomed SAP’s approach to integrate AI models inside of the application as a way to help solve this challenge. The company is now using SAP gen AI to analyze the text to find the right machine for a customer’s requirement. Pischke added:

That means it's much faster and reduces the stock of our machines, so we can have a quicker cash flow to get the money in which we paid for the producer of the machine. We have use cases and in the future, we don't want to develop that ourselves. We want to consume the algorithm of the applications.


A grey colored placeholder image