Understanding SAP's AI for HR

Learn what SAP is saying in its AI marketing and how it compares to its AI products on the market. This post is part of Textio's "AI washing" series.

Cassie Sanchez

Lifecycle Marketing Strategist

September 16, 2024

This post is part of Textio’s “AI washing” series, a review of popular HR AI products and their marketing. This is a collaborative series from Textio Marketing, Engineering, and Product teams. Some vendors are not AI washing. Sign up for updates.

Today’s review: SAP.

What they’re saying

SAP began publishing a consistent drip of AI-related press releases in May 2023 with an announcement that it would embed IBM Watson AI into its products. Since then, SAP has released ten additional AI announcements.

To sum it up: SAP has announced partnerships with IBM, Microsoft, Google Cloud, NVIDIA, AWS, Meta, and Mistral AI for AI projects. As of July 2023, SAP has also invested in three generative AI companies. The SAP AI vision is to embed what it calls “Business AI” throughout its portfolio, providing “rich insights at users’ fingertips, empowering them to deliver better outcomes and elevate the creative problem-solving humans do best.”

In September 2023, SAP announced it was building Joule, a generative AI assistant, and said it would be available in SAP SuccessFactors solutions and the SAP Start site later that year, and within “SAP S/4HANA Cloud, public edition” early the next year. SAP said in June 2024 that Joule was now embedded into additional SAP cloud products as promised. It also reported in June that it “is adding large language models … to [the SAP Business Technology Platform’s] generative AI hub” that will make it “easier to build generative AI use cases for SAP applications.”

At its annual SAP Sapphire event, also in June, SAP announced plans to integrate Joule with Microsoft Copilot for Microsoft 365 later this year, “combining enterprise data residing in SAP with contextual knowledge from Microsoft 365, including Microsoft Teams, Microsoft Outlook, Microsoft Word, and more, delivering richer insights for better decision-making.” SAP gave an example use case of a product manager “coordinat[ing] business activities” for a product launch with a cross-functional team that uses Microsoft Teams and Microsoft 365, with just “a few queries”—because “Joule can connect to the team’s individual skill profiles stored in SAP SuccessFactors software and core business information that resides in SAP S/4HANA Cloud.”

Finding specifics on SAP’s AI functionality outside of example use cases in press releases is a challenge. The website, particularly in terms of info on AI, is a nest of links that often merry-go-rounds you through the same few pages, each promising but often failing to answer your questions. A reasonable place to start is this AI for HR page, though following the threads from hypothetical use case to product you can buy is typically an unsuccessful treasure hunt.

You may find your way to a Joule product tour, which demonstrates how Joule helps a manager give employee feedback. It takes you through a series of question prompts like “What went well?” and “What could I improve on?”

You write the answers, and it continues to ask the same default questions until you’re done answering.

Then it sends your feedback.

You can also request information like “What is my cost center?” or change personal information like your pronouns. Joule can also prompt you through the paperwork of promoting an employee.

Joule product tour: Promoting an employee

SAP also makes general mentions of AI assistance specifically for recruiting on its website: ”Quickly create inclusive job descriptions that are personalized to your business. Prepare for candidate interviews with questions tailored to the specific role. Access relevant information from your company’s guidelines and policies at your fingertips.” But you’re going to need to look through an 11-page solution brief to get any specifics, in the form of (mostly unreadable) screenshots. The words are still marketing-speak.

SAP SuccessFactors Recruiting solution brief

The pertinent AI-related info in the brief is:

Re: careers sites: “flexible rules engine automates posting of external and internal jobs”
Re: candidate relationship management: “automated email campaigns”
Re: talent acquisition workflows: “auto-populated talent pools, and auto-progression of candidates”
Re: job descriptions: “With AI-powered recommendations, recruiters and hiring managers can easily create compelling job descriptions that display relevant skills and competencies needed for a job”

So a lot of automation capabilities, no follow-up on “inclusive” and “personalized” job descriptions, nor “tailored” interview questions.

A blog from October 2023 gives some insight on SAP’s “talent intelligence hub” mentioned in other materials. It’s an “AI-powered skills framework built into SAP SuccessFactors HXM Suite,” and is intended to assess and store individuals’ skills and goals across the org, and identify gaps and opportunities for upskilling and reskilling for employees and teams. Employees would benefit from auto-populated personalized learning recommendations and growth assignments, and HR teams would be able to use the hub to “drive [their] organization’s entire learning and talent strategy” with a skills model that “covers recruiting, onboarding, learning and development, performance, and succession.” The talent intelligence hub was generally available as of the blog’s publishing.

There are nods to “AI-assisted writing” in recent SAP marketing content, which you’ll find the most helpful info on in a Help Portal article. Based on the article, this feature is intended to help employees “with their writing tasks” and should “enhance multiple aspects of their writing, such as clarity, conciseness, and tone.” Beyond standard LLM capabilities like shortening or lengthening text or modifying the tone, SAP’s AI writing assistance is said to “perform a safety scan” on both the text you put into the tool and the text you receive back. This is “to detect bias, and suggest replacement text for any language flagged as potentially biased, discriminatory, or harmful.”

In marketing and interviews, SAP puts a big emphasis on a need for bias protections and data privacy in AI. Daniel Beck, president and Chief Product Officer of SAP SuccessFactors, noted in June 2024 that SAP “has put controls in place through the SAP AI Global Ethics organization and legal experts to evaluate AI use cases – ensuring that they will be used safely, responsibly, and without bias – before they are made available to customers.”

What they have

A whole lot of marketing, for one thing. Thankfully SAP also has release notes, so we can begin to understand functionality as well as what‘s available to buy and use today in terms of HR AI. Though SAP does issue this caution:

SAP cautions against expectations vs. commitments in release notes

Essentially, SAP appears to have some automation and talent intelligence, mostly basic large language model (LLM) functionality, and a budding chatbot.

Basic automation and talent intelligence

It is probably safe to assume the many automation capabilities for SAP SuccessFactors Recruiting highlighted in the solution brief are available, given they are presented within a solution brief. Among the features, auto-progression of candidates seems to be the only novel offering in terms of what’s available in HR tech broadly, though when scaled out it could indeed save HR teams meaningful time.

Within the talent intelligence hub, we see a lot of standard HR AI capabilities around employee and candidate skills. HR teams can use AI to pull skills info from resumes to speed up screening; employees and their managers can access AI-enabled recommendations for growth opportunities and internal jobs and assignments. SAP’s AI functionality for creating a skills library based on job requisitions and existing roles sounds like a unique and useful dashboard for workforce planning that should drive more successful talent strategies. And it is generally available.

Mostly basic LLM functionality

Among the generative AI features, outside of Joule, SAP has launched “person insights”: summaries of “complex information” on compensation and job history to aid managers in comp discussions. Without more details, the benefit of this feature sounds oversold—that information would likely not be so complex, and an AI-generated summary of it would have minimal benefit.

The AI-assisted writing feature appears to have no real added benefit beyond a free LLM like ChatGPT, other than the “safety scan.” There are no details on how SAP identifies unsafe content or examples of how it could help writers use more inclusive and appropriate language, so it’s difficult to determine the true value or even accuracy of this functionality. Previous Textio analyses on bias detection in job descriptions showed SAP struggled here.

Textio analyzed job descriptions written by ChatGPT in six job post tools

SAP ultimately offers no guarantees on the safety of its generative output.

SAP does say it has launched “25 new AI capabilities” with generative AI, including the ability to help employees craft “ready-made goals” for performance and development, and managers to “enhance the quality and tone of performance goals, and [make] feedback more readily actionable.” There is no detail on how or if these models are trained on goals and feedback data specifically, suggesting the guidance is based on general writing and therefore not actually optimized for performance-management content.

One generative AI tool that does sound quite useful is SAP’s use of “retrieval augmented generation (RAG)” which would contextualize general LLM output with “personal or business information” to provide answers and insights to HR teams. There is no further detail on the types of answers or insights, but modifying generic LLM functionality to incorporate organizational context is a significantly beneficial upgrade regardless. It is unclear if this is part of the Joule product or separate.

An evolving chatbot

On Joule, we can see that SAP has delivered its chatbot as promised and continues to upgrade it. SAP says in its release notes that it has “added more than 15 new capabilities to Joule.” Right now it appears you can use it to complete basic administrative tasks like updating your pronouns, submitting paperwork for an employee promotion, clocking in and out, viewing pay statements, requesting and approving time off, and getting answers to policy questions. For any company that requires significant HR involvement for these processes or otherwise makes it difficult for employees to access them, using Joule would be a big time saver for all parties.

Managers can also use Joule to give employee feedback by answering a series of questions Joule prompts you with. There is no assistance here for writing the feedback itself though, and it is unclear if you can use it for performance reviews.

Integrating Joule with Microsoft Copilot and Microsoft products should allow for more useful outputs and functionality from Joule for teams that use Microsoft. This feature has not yet shipped, so functionality and ultimate utility are still to be determined.

How they rank

We put SAP at 60% on the Marketing Honesty spectrum and 50% on the AI Utility spectrum. SAP gets credit for delivering the products and features it has promised. However, its positioning of generative AI capabilities is at times exaggerated or misleading, as in the case of “inclusive” and “personalized” job descriptions, or summaries of "complex" compensation and job history data. SAP’s AI for HR is currently a broad but basic set of AI tools that combine to about the halfway point on the AI utility spectrum. SAP has a basic chatbot, with plans to make it better; mostly basic LLM functionality, with lip service but no specifics on bias protection; and basic automation and talent intelligence that could have been offered years ago when the technology became available.

graphic of AI utility spectrum (from This Is Generic/Free AI to This is Unique and Useful AI) and Marketing Honesty spectrum (from This Product Doesn't Exist to This Is Exactly As Advertised) with a plot point at 50% AI Utility and 60% Marketing Honesty

Next up in the series: Oracle. Subscribe for updates.