SAP’s €1 billion bet: the AI running your ledger will not be a language model

SAP completed its acquisition of Prior Labs on 17 July and committed more than €1 billion over the next four years to scale it into a frontier AI lab for structured data. Rachel Hunter, Head of AI at SAP ANZ, says what comes out of it will belong to a different category from the models most organisations have spent the past two years learning to use.

“Tabular models are specialised for looking at numbers in the structures of data,” Hunter told SAPinsider at SAP NOW AI Tour Australia and New Zealand in Sydney. “The LLMs are typically trained to look at sentence structures and how to put paragraphs together. Tabular models are looking at lines and rows in a different way.”

According to Hunter, “this is going to be a game changer” for SAP customers.

Prior Labs will continue to operate as an independent entity under SAP ownership. Its TabPFN model series predicts outcomes directly from tabular business data, including payment delays, supplier risk, upsell opportunities and customer churn, and does so without training or fine-tuning on the customer’s own data. SAP’s stated reason for buying the company is that large language models have only a rudimentary understanding of tables, numbers and statistics, which is why the technology sits outside the language model line that started with SAP-RPT-1.

Noting that a tabular model is not just a spreadsheet-shaped version of an LLM, Hunter advises customers to “watch this space” for what comes after the Prior Labs acquisition.

The tolerance question

The case for a separate class of model rests on how much error a process can absorb. Hunter drew that line directly at a customer briefing in Sydney the same morning, on the subject of business data: “We don’t need any creative inflection. We need accuracy.”

That standard sits awkwardly against how most organisations have been using AI. Hunter says the tasks people have found most successful, narrative and conversational work, tolerate far more variance than transactional work does, and that the gap between the two is where a lot of current disappointment is being generated.

“What is the quality of the question?”

SAP’s research puts 73% of companies reporting data quality challenges. SAPinsider’s benchmark research finds 88% of enterprises using AI somewhere, but just 12% saying their data is genuinely AI-ready. Hunter, who spent more than 25 years in data engineering before joining SAP at the end of May, treats the underlying problem as a constant rather than a new development.

“We will always find challenges with data,” she said, while describing the foundation of data as incredibly important to the quality and the trust organisations see in AI outputs. What has changed is the speed at which data can be managed, and how much of it organisations can now see. Customers on the clean core journey have a trusted layer underneath them that they did not have before.

According to Hunter, evaluation of AI-data readiness might also lie in the process as much as the data quality. Organisations reach for a generative tool, expect a rules-based answer, and blame their data when they do not get one. “Maybe the answer is, what is the quality of the question?”

“Did we decide the data quality is poor because we used the wrong tool to solve?” she said.

Where to find the value

While much of the current commentary states that AI compresses the value of experience, Hunter sees experience as a vital part of an AI-enabled organisation.

“I would almost argue that AI is most empowering in the hands of people who have a lot of career experience,” she said, “because we’re able to look at something pretty quickly and determine if the output is the quality that we would normally produce ourselves in a much more significant amount of time.” AI can accelerate the work, she said, but “it doesn’t ultimately change what good looks like.”

That is where she put her expectations for the coming year.

“What really excites me about the next 12 months is seeing people, particularly in the SAP landscape, who are able to pick up the trusted AI from SAP and turn their years of experience into powerful outcomes,” she said. “They have so much knowledge and experience, and they will now be able to harness AI with their trusted business data to really get cool things so much more quickly than they could have in the past.”