Rahul Saluja is a technology and business leader focused on AI-driven enterprise transformation and operating-model innovation.
A successful AI pilot can leave leaders confident about the technology and unclear about the work still required to make it useful.
We encountered that gap during an AI initiative. The use case was strong, and the early results were promising. We kept the problem narrow, involved the business early and put something tangible in front of users. Those choices helped us learn from actual use instead of spending months discussing AI strategy.
The difficulty came when we tried to move beyond the controlled pilot environment. The biggest friction was data: inconsistencies in how information was captured, accessed and governed. The pilot had demonstrated potential, but moving into everyday operations exposed conditions it hadn’t resolved.
The lesson was more specific than “we needed better data.” We needed to treat data readiness as part of the AI initiative rather than as a prerequisite someone else would solve. I would bring that work forward much earlier now.
Examining The Conditions Behind The Result
A pilot necessarily has boundaries. Narrowing the problem and controlling inputs can help a team establish whether an idea deserves further investment. The next step is to examine how those conditions differ from the intended operating environment.
For the workflow being tested, I would ask: How was the pilot’s information prepared? How will that same information be captured during normal operations? Who will need access? What governance decisions remain unresolved?
These questions help explain what the result actually demonstrates. A promising output is evidence about the inputs and conditions tested. It doesn’t establish that the business can consistently provide those inputs under different conditions.
That distinction should shape the next stage of work. Each material difference between the pilot and the operating environment needs to become an explicit part of the plan.
Keeping The Data Work As Narrow As The Use Case
There’s a danger in responding to every data gap with a much larger enterprise transformation program. An initiative can become dependent on solving problems far beyond the workflow it was intended to improve.
I would apply the same discipline to data readiness that helped the pilot: Keep the scope specific. Start with the information the chosen workflow depends on. Establish what must be consistent, who must be able to access it and how its use must be governed. Then identify which gaps prevent the next stage of deployment.
This makes readiness something a team can investigate and act on. “Improve data quality” is difficult to assign or complete. Identifying an inconsistency in information that a particular workflow requires creates a clearer problem to resolve.
The objective is sufficient readiness for the intended use, with the remaining limitations understood.
Turning Gaps Into Decisions And Ownership
Recognizing a data problem doesn’t resolve it. A useful plan also needs to establish who has the authority to act.
For each material gap, I would want three things made explicit: the decision required, the person accountable for resolving it and the evidence that will show it’s been addressed. Some gaps may require changes to how information is captured. Others may involve access or governance. Treating them all as technical cleanup can obscure the business decisions involved.
This is where early business involvement needs to continue beyond demonstrating the use case. The people who understand the workflow must remain involved as the team determines what needs to change. Those decisions belong in the initiative’s scope, resourcing and rollout plan. Leaving them outside the plan makes progress dependent on work for which nobody has taken responsibility.
Giving The Pilot A More Useful Exit Test
I would still move quickly to put something tangible in front of users. Our early results helped establish that the idea was worth pursuing. What I would change is the conversation at the end of that stage.
Alongside reviewing performance, I would ask the team to explain what had to be true for the pilot to work, which conditions would differ in everyday operations and who owned closing each consequential gap. A team doesn’t need every answer before it can learn, but it does need to distinguish what it’s demonstrated from what it’s still assuming.
For leaders, that distinction makes a successful pilot more useful. It provides both evidence of potential and a clearer account of the work required to realize it.
Before approving the next stage, ask: Does the plan include the work needed to make the pilot’s conditions repeatable? If that work sits outside the initiative, part of the AI business case is still unowned.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

