An AI initiative is exciting right up until someone asks when it will be done. That question arrives early, usually right behind what will it cost, and it is hard to answer well because the thing being asked about is still a concept. And concepts do not have durations. So how do you estimate a realistic timeline?
1. Establish a clear end point
Before you can put a length on a timeline, you have to know what done means. The easy answer is "when the AI tool is implemented," but is that actually complete? Implementation does not account for how many people have been onboarded, what training has been delivered, or which processes have moved over. A tool that is live and used by only four people may not be considered finished or successful. Make sure you understand what everyone's expectations are about the project and identify the end point that aligns with them.
2. Check your readiness to begin
A timeline estimated off the top of someone's head covers the project and nothing else. Start digging into your systems and find the work that has to happen first: data that is inaccurate or scattered, access that has to be provisioned, integrations that were never built, policies that do not exist yet. That preparation can add weeks or months and can ruin your timeline before the project even begins. Look before you commit to a date rather than after.
3. Determine what you own and what gets handed off
You will not be running all of this yourself. An outside vendor may handle execution, support, or advising, and your team will handle the rest. Split the work by owner first, then look honestly at your side of the line. The question is not whether your people can do their part. It is what else is in their queue this quarter, and whether any single person is the only one who can unblock the next step.
4. Estimate in phases rather than in total
Break the initiative into its components and put them in the order they have to happen. Estimate each one and note which can run in parallel. This takes the pressure off having to be an expert over a large project that may consist of many unknowns, to being able to estimate the pieces that teams probably already have experience in. It also shows where the initiative could pause without being abandoned.
5. Establish checkpoints and say what will exist at each one
Set regular points to review progress, and define what will be working by then. Not a status update, something people can see and use. Checkpoints give you a place to correct the timeline while correcting it is still cheap, and they give everyone evidence between the approval and the finish line.
Unrealistic timelines consist of miscalculated unknowns. Every step above allows you to change an unknown to a piece of information that is now accounted for. Which is why the date gets more accurate as the work gets more real rather than the other way around.