SAP and Infosys executives say companies scaling AI beyond pilots need a sound data foundation, carefully chosen use cases and clear governance. Their practical advice is to measure whether an AI project improves a real business outcome before letting it loose across the organisation.
SAP Watch analysis
What happened
In a sponsored MIT Technology Review Business Lab episode, SAP executives Thorsten Leiduck and Satinder Pal Singh discussed the shift from isolated AI experiments to systems embedded in business processes. Leiduck said disconnected, inconsistent data foundations can lead to duplicated work, uneven performance and no clear route to scale.
The executives recommend assessing potential uses through impact, feasibility and scalability: will the system move a meaningful business metric, can the organisation support it with its current data and skills, and can a successful deployment be repeated? They suggest setting measures such as cycle time, accuracy, cost per transaction or revenue improvement in advance.
Why it matters
That is a more useful test than asking whether a task can be automated. The interview notes that some routine applications may tolerate an imperfect result, while errors in areas such as supply chains or financial close can be far more serious. The right threshold depends on the work, not the enthusiasm in the boardroom.
As agents take on more responsibilities, the speakers also point to policies, access controls, monitoring and oversight as essentials. Those are recommendations from the executives, not evidence that every organisation has put them in place.
Our read
The strongest advice here is pleasingly unglamorous: choose a problem with a measurable payoff, check that the data and people are ready, then scale only what works. “AI transformation” sounds grand; a defined metric is harder to hide behind.
The episode was produced in partnership with Infosys, whose executive also appears in the discussion. Treat the framework as their advice, not a neutral audit of what enterprises should do.
What to watch
- Whether companies publish results against measures such as accuracy, cycle time or cost per transaction.
- Whether agent deployments gain clear access controls and ongoing monitoring.
- Whether successful pilots can be repeated across teams without losing their value.
Discussion spark: Should companies scale AI only after a pilot proves measurable value, or does that standard risk making them too cautious to learn?
Sources and evidence
- Building AI that can work across the enterprise (16 November 2026, 13:00 UTC)
- AMD Closes World Labs Acquisition and Names Fei-Fei Li Chief Scientist – Unite.AI (16 November 2026, 13:00 UTC)
Independent WittyWires tracker for public updates about SAP. Not affiliated with or endorsed by SAP; this is not an official account.