At Rise of AI 2022 in Berlin, Dr. Feiyu Xu, Senior Vice President and Global Head of Artificial Intelligence at SAP, argues that after decades of progress, from machine translation once taking minutes per sentence to now running in milliseconds, AI should be used to remove boring, repetitive work from corporate life the way it already removes friction from consumer life. She presents three pillars of SAP's enterprise AI strategy: intelligent applications customers can use out of the box with no configuration, adaptability so customers can extend applications to their own industry needs, and trustworthy, responsible AI given that SAP's applications run mission-critical processes in HR, finance and supply chain. She walks through four end-to-end business processes SAP targets with AI, lead to cash, design to operate, source to pay and recruit to retire, and gives concrete results, including a supply chain forecast that dropped from 160 hours to 4 and now considers 66 million data points instead of 22,000, and a Concur travel-expense product used by 26,000 customers that now auto-audits 70% of expense reports. She closes by describing SAP's AI ethics governance, built since 2018 around guiding principles, an internal steering committee and an external expert panel.
Key takeaways
- Xu contrasts 1990s parsing that took minutes per sentence with today's millisecond machine translation to frame three decades of AI progress.
- SAP's enterprise AI strategy rests on three pillars: out-of-the-box usability, adaptability to industry needs, and trustworthy, responsible AI for mission-critical processes.
- She cites a company where AI cut long-term supply chain forecasting from 160 hours to 4 hours while expanding data points considered from 22,000 to 66 million.
- SAP's Concur expense app processes about $1.6 billion in expenses yearly for 26,000 customers and now auto-audits 70% of travel expense reports.
- SAP has run an AI ethics governance program since 2018, including a steering committee, an external expert panel and a mandatory employee AI ethics policy since this year.
Detailed notes
From John and Mary to real-time translation: why enterprise AI matters now
Xu opens by contrasting a 1990s AI system that took several minutes to parse the simple sentence "John loves Mary" with today's machine translation systems that translate thousands of sentences in milliseconds, crediting decades of progress in computing hardware, big data, deep learning and knowledge graphs. She notes how AI already removes friction from consumers' private lives through face unlock, predictive text and smart speaker recommendations, and argues the same comfort should extend to corporate life, for employees, CTOs, CFOs and account managers, citing tedious travel expense reports, manufacturing spare-parts ordering, and unread conference emails as examples of boring work AI applications can take over.
Three pillars of SAP's enterprise AI strategy
She describes SAP's approach around three pillars: first, infusing AI into applications so customers, who are not data scientists, can use them out of the box with low or no configuration; second, adaptability, since each industry and company has different needs, so customers must be able to extend standard applications to their own use cases; third, trustworthy and responsible AI, which she says is critical because SAP's applications run mission-critical processes for human resources, finance and supply chains.
Four end-to-end business processes AI can transform
Xu outlines four processes SAP targets: lead to cash, where AI helps identify potential customers, close contracts compliantly, and process invoices and purchase orders through document extraction and object recognition; design to operate, where AI reads social media and market feedback to inform product design, uses computer vision for manufacturing inspection, and optimizes assembly-line scheduling, which she says one company cut from two engineers spending two hours each morning to a two-second automated calculation; source to pay, covering supply chain resilience; and recruit to retire, covering resume screening, onboarding, learning recommendations and career development.
Deep dive: source to pay
She illustrates source to pay through four fictional roles. Edward, who sources suppliers, is helped by AI that recommends suppliers and monitors risk. Lisa, who receives materials, needs AI that monitors real-world events like natural disasters and predicts delivery delays that could halt production. A third role, matching purchase orders to invoices, uses AI to catch mismatches that otherwise cause payment errors or supplier complaints. Bob, who manages ongoing supplier relationships, uses AI to monitor events like bankruptcy or management changes at suppliers.
Concrete results: expense auditing and supply chain forecasting
Xu gives measured outcomes from SAP Concur and machine learning-based forecasting. Concur's mobile expense app processes about $1.6 billion in expenses annually for 26,000 customers, using computer vision and OCR to read receipts and natural language processing to extract information, and now auto-audits 70% of travel expense reports built on a decade of human audit experience. For supply chain and sales demand forecasting, she describes a customer for whom long-term (three-year) forecasting dropped from 160 hours to 4 hours with AI, while the system can factor in 66 million data points compared with about 22,000 a human forecaster could consider, enabling forecasts roughly every half day. In manufacturing quality inspection, she says computer vision can now replace about 80% of the human effort needed to spot defects like a misplaced or missing sticker.
Trustworthy AI governance and Q&A highlights
Xu describes SAP's AI ethics program: guiding principles released in 2018, aligned with UN human rights principles and compatible with the EU's AI ethics guidelines; an internal steering committee spanning HR, legal, technology, engineering and marketing; an external expert panel; and, from January of this year, a mandatory AI ethics policy for all employees covering each stage from ideation to productization. Asked how SAP reconciles its own AI ethics policy with differing customer frameworks, she says SAP's principles sit under the shared UN and EU umbrella, which she expects gives substantial overlap, though she acknowledges the policy is currently mainly internal and that reconciling differences in ongoing co-innovation with customers remains an open challenge. Asked what excites her most, she points to enterprise AI making business processes, and therefore companies, more intelligent, and to combining data-driven machine learning decisions with explicit business knowledge graphs to make applications explainable, comparing this to how humans combine intuition with learned knowledge to justify their decisions.
Quotes
The difference between enterprise AI and business AI is that there's a high standard of security and compliance, we call it enterprise readiness.
Feiyu Xu
AI ethics, trustworthy AI, often is only very theoretic.
Feiyu Xu
Make the business process intelligent means make every company intelligent.
Feiyu Xu
Questions answered in this talk
What are the three pillars of SAP's enterprise AI strategy?
According to Dr. Feiyu Xu at Rise of AI 2022, SAP's strategy rests on out-of-the-box applications requiring little or no configuration, adaptability so customers can extend applications to their own industry needs, and trustworthy, responsible AI given that its applications run mission-critical HR, finance and supply chain processes.
How much did AI improve supply chain forecasting for one of SAP's customers?
According to Dr. Feiyu Xu at Rise of AI 2022, one company's long-term supply chain and sales forecasting dropped from 160 hours to 4 hours with AI, while the system could factor in 66 million data points versus about 22,000 for a human forecaster.
How does SAP govern trustworthy and responsible AI?
According to Dr. Feiyu Xu at Rise of AI 2022, SAP has published AI guiding principles since 2018 aligned with UN and EU AI ethics frameworks, runs an internal steering committee and an external expert panel, and introduced a mandatory AI ethics policy for all employees this year.