Many organizations have tested AI in isolated projects, but those experiments can be difficult to turn into dependable business workflows. As AI agents begin taking on tasks across applications, companies face a broader challenge: connecting data, processes, governance, and people so AI can operate as part of the enterprise.
Build the foundation before expanding AI
Thorsten Leiduck, SVP and global head of SAP Business AI Platform Ecosystem, argues that AI depends on a sound data foundation. Agents need connected, semantically rich information that reflects how a company works, rather than relying only on generic patterns.
When projects begin with disconnected technologies and fragmented data, organizations can end up duplicating work and seeing inconsistent performance. A shared platform and data layer can help connect information from different sources with the context of business processes.
That context matters because accuracy expectations vary by task. An AI-generated presentation or email draft can be reviewed by a person, so imperfect output may still be useful. In a supply chain process or financial close, however, the consequences of errors are greater. Leiduck says that while “80% might be good enough” for some everyday applications, it is “a very terrible result” for mission-critical work.
Choose opportunities with a path to scale
Satinder Pal Singh, associate VP North America for SAP BAIP and SAP AI Strategy at Infosys, describes AI agents as “digital coworkers.” His advice is to evaluate potential uses through three lenses: impact, feasibility, and scalability.
- Impact: Identify whether an agent can improve meaningful business measures.
- Feasibility: Check whether the company has the data foundation and talent to deliver the use case.
- Scalability: Decide whether an effective application can be replicated beyond its initial deployment.
The sequence matters. A high-impact idea may not be viable if the data foundation is weak. Once an organization proves a use case with clear measures, it also needs an architecture that supports applying the approach elsewhere. Singh puts it simply: “A pilot that can't scale is just an experiment.”
This approach also helps leaders avoid treating AI as the default answer to every task. Some work may remain more economical or effective when done by people. Companies can begin with a defined set of valuable use cases, establish success measures in advance, and expand the ones that deliver results.
Make governance part of the operating model
Moving agents into business workflows raises questions about security, reliability, accountability, and oversight. Clear policies, access controls, monitoring, and governance help keep deployments aligned with business requirements as their responsibilities grow.
Governance can be treated as part of the work of scaling, rather than a later review. Singh warns that its absence does not make adoption faster: “It creates a risk backlog” that may eventually lead to a more disruptive stop and audit.
The broader shift is organizational as well as technical. AI cannot remain a set of experiments isolated from the teams and workflows it is meant to support. Leaders need to connect business priorities with the data, people, and oversight required to put agents to work responsibly.
From productivity aid to decision partner
Both executives anticipate agentic systems coordinating work across applications. They also see customer-specific knowledge graphs and governed data becoming important enterprise assets as AI takes on more responsibilities.
Singh describes the longer-term possibility as moving from AI as a productivity tool to “AI as a decision-making partner.” In that role, AI could help leaders understand current conditions, anticipate what may come next, and consider how to respond. Reaching that point depends on the groundwork: useful business goals, data that reflects the organization, and governance designed to keep systems trustworthy.