Large organisations around the world have overcome some crucial hurdles to adopting AI and are moving from proof of concept (POC) tests to more widespread adoption this year, according to software firm SAP.
The company, which supplies much of the software needed for payroll, procurement and other corporate functions for large businesses, is seeing AI efforts accelerate despite early stumbling blocks.
“In the last two years, it was POCs, but now, we are seeing adoption,” said Philipp Herzig, SAP’s chief technology officer and chief AI officer.
The conversation has moved from discussing graphics processing units (GPUs), models and platforms to actual production to deliver real results, he told reporters last week in Singapore.
“To build out takes time,” he noted, but increasingly, organisations are learning to take AI “out of the box” from SAP products instead of using “precious IT people or data scientists” to tap on AI to improve human resource (HR) functions, for example.
These experts, he argued, should be used for creating AI use cases to improve an organisation’s core products and make a bigger impact for businesses.
In September 2023, SAP showcased its Joule generative AI copilot that was embedded in its cloud-based enterprise resource planning (ERP) products. It promised to quickly sort through and contextualise data from multiple systems to surface smarter insights.
Since then, the German company has seen adoption by customers in manufacturing, retail and professional services. These are typically large organisations running various corporate IT functions with SAP software.
As AI has become easier to adopt with many features built into their SAP software, organisations have also learnt to overcome issues with data privacy and security, for example, by limiting the data shared and used externally, said Herzig.
AI has also become more scalable through the use of ready-made solutions that cater to local markets, such as SAP products that provide for Asian languages and policies, he added.
“When the user is logged in, we know the user already… and their privileges, and we can pull the respective context,” said Herzig.
Like all AI efforts, however, data quality is crucial to success. Without bringing in the right data onboard, businesses will find it difficult to get accurate or useful results from AI. Rubbish in, rubbish out, after all.
SAP is keen to say that its customers already have a lot of great data on hand, thanks to years of running, say, payroll, procurement and corporate travel systems on its ERP software.
For example, its “first of its kind” AI agent for accounts payables and receivables draws on reasoning models that can deeply analysing the data stored in complex tables, thanks to SAP’s structured systems built over the years.
However, while some have solved the data problem, say, with “beautiful data schema” on SAP’s most recent S4 Hana systems, Herzig acknowledged that others will have issues with data quality.
If they have changed all the columns in their data structure, or if they are using a 20- or 30-year-old data schema, then it won’t be so easy to roll out AI out of the box, he noted.
Plus, there is also non-SAP data and other data that are unstructured, which may complicate things, he added, though more signal in future can still help improve AI success.
