Most AI projects die before week three.

The pattern is consistent enough to predict from the first conversation.

A business decides it should be doing something with AI. Someone finds a tool. The tool gets bought, usually on a monthly plan, usually after a demo that looked impressive. It gets introduced at a meeting. Two or three people try it that week. By week three nobody has opened it. Four months later it shows up on a card statement and gets cancelled quietly.

Nobody calls this a failure. It gets filed as "we tried AI and it was overhyped." That conclusion is wrong, and it is expensive, because it inoculates the business against trying again for about two years.

The actual cause

The project did not fail at the tool. It failed before the tool, at a step that was skipped.

Nobody wrote down where the hours were going.

Every one of these purchases was made against an idea of the business rather than the business. Somebody imagined that writing marketing copy was the bottleneck, because that is the thing AI is loudest about. Meanwhile the actual bottleneck was that quotes took three days to go out, or that calls after four o'clock went to a voicemail nobody checked, or that the same six questions arrived by email every day and got answered by hand every time.

A tool bought against an imagined problem will be abandoned, no matter how good the tool is.

The two hours that prevent it

Before anything gets purchased, someone needs to sit with the people doing the work and watch. Not a survey. Not a meeting about it. Watching.

What comes out of two hours of that is usually surprising to the owner, and it is almost never what they expected:

• The tasks that eat the most time are boring and invisible, which is exactly why nobody flagged them.
• At least one of them is being done twice by two different people who do not know about each other.
• One of them can be eliminated entirely without any AI at all, by changing a form or a handoff.
• The one that would benefit most from AI is rarely the one anybody named at the start.

Then rank it honestly

Once you know where the hours go, rank the opportunities by hours saved against cost and setup effort, and be willing to conclude that most of them are not worth doing yet. A list of ten ideas where seven get crossed out is a good outcome. It means the three that survive are real.

This is unglamorous and it is the entire difference between a project that sticks and one that gets cancelled in month four. It also costs less than the tool would have.