Most of the planning effort on an AI initiative goes into testing assumptions. Is the data good enough? Will the model handle our cases? Can we connect it to the systems we already run? Those questions matter. There is a second set of questions that gets far less attention and, in our experience, causes as much trouble: what the people involved expect to happen.
Expectations are worth reviewing on their own because they behave differently from assumptions. A bad assumption often stays hidden until the initiative is well under way. A bad expectation is usually visible from the first conversation. That makes expectations easier to test and cheaper to fix, as long as someone looks at them before the plan is locked in.
AI makes this more important. Claims about what AI can do move faster than most organizations can absorb them, and executives often arrive at the first planning meeting with expectations shaped by demonstrations, headlines and vendor pitches. The potential is often real. What gets underestimated is the time, effort and conditions needed to realize it.
Consider a common exchange. A sponsor says, "We need the AI assistant live for customers by July 1." The team replies, "OK, we'll plan around July 1." The sponsor leaves believing the date is achievable and starts repeating it to senior executives, customers and partners. Within weeks, the team realizes the date is unlikely. Pressure builds, steering meetings get tense, testing gets cut short and the date slides anyway. The eventual launch often lands later than a realistic plan would have, and the organization has spent goodwill getting there.
A different reply changes the outcome. "I understand why this is urgent, and we will move as fast as we responsibly can. Before we commit to a date, let's make sure our expectations are realistic, so that the launch succeeds." That reply keeps pace and opens the conversation that holds the initiative on track.
In our experience, eight areas of expectation are worth reviewing during planning, ideally in a workshop with the sponsor, the delivery team and the people whose work will change.
Duration. How long will this take, including the time to test output with real users and real data rather than a demonstration set? How long have comparable initiatives taken elsewhere in the organization?
Effort. How many people need to be dedicated to it? Is anyone expected to deliver the initiative on top of a full-time role? Who will review and correct the AI's output once it is live, and has that work been counted?
Risk and complexity. Is the initiative being described as simpler than it is? What happens when the AI produces a wrong answer, and how often would that be acceptable?
Skills. Does the team have people who can judge data quality, evaluate output and redesign the work around the tool, as well as people who can deliver change?
Budget and benefits. Is the budget realistic, including the ongoing cost of running the tool at the volumes planned? Are benefits being promised for a date before people will have adopted the new way of working?
Accountability. Is it clear who owns the outcome, who approves the output and who decides when the initiative is ready for customers?
Disruption. How much will this disrupt the business while it is being introduced, and afterward? How will the organization support the people whose roles change?
Technology and data. Is the technology proven for this use, or still emerging? Is the data ready, or does it need work nobody has planned for? Is anyone assuming integration will be easy because the demonstration was?
Each area has a simple test: would the sponsor, the delivery team and the people affected give the same answer? Where they would not, there is an expectation gap, and it is far cheaper to close it in planning than in delivery.
Realistic expectations are not low expectations. AI can change the economics of how an organization works, and the initiatives that deliver that change are usually the ones planned with a clear view of what it would take.