AI in Business: SAP Recommends Targeting Use Cases

Certain tasks remain less costly or more efficient when entrusted to humans, believes Thorsten Leiduck, head of SAP. Together with Satinder Pal Singh from Infosys, he advocates for a selective generalization of agents, supported by business indicators, coherent data, and ongoing governance.
SAP recommends measuring before generalizing
Human intervention retains a cost or efficiency advantage for certain tasks: this is the caveat expressed by Thorsten Leiduck, SVP in charge of the global SAP Business AI ecosystem. In his view, an agent performs a specific mission within a business process, without every task necessarily calling for automation by AI. In his discussions with clients and their leaders, economic and governance considerations are gaining more prominence as usage progresses. The cost of tokens, model performance, security, and accountability are becoming more significant criteria.
His recommendation is to delineate a portfolio of high-value applications, possibly limited, and then establish success criteria before deployment. He suggests tracking cycle duration, accuracy achieved, cost per transaction, or revenue growth. The extension should then focus on applications that deliver satisfaction.
Satinder Pal Singh, Associate VP for North America at SAP BAIP and SAP AI Strategy at Infosys, proposes a complementary framework for these agents, which he likens to digital collaborators. He distinguishes the impact on significant business indicators, feasibility concerning available data and skills, and the possibility of replicating a successful application. Therefore, the success of a pilot does not constitute the sole selection criterion in this framework.
Surveys indicate limited deployments
According to Infosys and HFS Research, 14% of over 500 global companies surveyed from the Global 2000 have successfully generalized AI. The remaining 86% are still in the initial exploration phase, proof of concepts, or scattered pilots. Another measure cited by Thorsten Leiduck: BCG estimates that 74% of organizations are stuck at the pilot stage, without a generalization plan. These results come from separate surveys.
As agents begin to enter operations, Leiduck describes clients struggling to convert their interest into enterprise-wide value. He attributes this to years of separate experiments: different technologies, fragmented and inconsistent data would produce duplicates, variable results, and a poorly defined deployment trajectory. Singh points to the isolation of AI teams or centers of excellence, capable of technically succeeding in a pilot without gaining recognition from the business. He advocates for involving all functions and their workflows in an operational model transformation from the outset, rather than reserving the topic for IT.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Singh emphasizes that AI has been a priority for boards of directors for the past two years. Infosys' Knowledge Institute indicates that 8% of companies claim to be AI-native and attributes to them a contribution to revenue four times greater than that of laggards. According to Singh, buying a similar platform later does not allow one to catch up on eighteen months of behavioral data from agents, a calibrated knowledge graph, and employees already familiar with these tools. He believes that delaying this learning exacerbates the lag.
For Leiduck, adoption is as much about change management as it is about technical installation. He recommends exposing numerous teams to use cases to generate ideas and make benefits perceptible. Assistance with travel, meeting notes with Copilot, and processes using AI improve, in his view, daily productivity. He also attributes an effect on creativity and innovation stemming from teams.
SAP connects its autonomous framework to business data
According to Thorsten Leiduck, 42% of respondents to a survey conducted by Anthropic report difficulties accessing the required data and quality to expand AI within their organization. The SAP leader advocates for a common platform rather than a juxtaposition of projects. He associates it with a connected data foundation, rich in semantic relationships and representative of activity, linked to the context of application processes.
SAP has launched a framework for the autonomous enterprise, aimed at supporting the transition from experimentation to large-scale results. Leiduck, however, emphasizes different reliability requirements depending on the use case. An accuracy of 80% may seem acceptable for some common applications; he considers this level very poor when it concerns procurement or financial closing.
Infosys advocates for continuous monitoring of agents
Security, observability, reliability of agents, and the ethical implications of their deployment concern the leaders met by Thorsten Leiduck. He recommends framing AI capabilities throughout their lifecycle. Satinder Pal Singh details the instruments for this control: explicit rules, access rights, monitoring, and tracking. In his view, the absence of governance accumulates risks that can later necessitate a more disruptive interruption and audit.
The two leaders envision, in the long term, agent systems tasked with coordinating tasks across multiple applications. In this scenario, governed data and client-specific knowledge graphs would become critical assets. Singh also envisions decision support assisting leaders in understanding events, anticipating their evolution, and determining their response.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.