AI Implementation Checklist: The Items Nobody Writes Down.

Before the first row: what is true right now.

Every plan rests on a picture of how work gets done. Usually that picture comes from a process map drawn during a systems project some years back, updated once, and then abandoned by the people it describes.

Ground truth is the other version. It is how the work actually moves. The spreadsheet a team lead rebuilds every Monday because the reporting tool never quite fit. The approval that officially takes two days and really takes ten minutes over a message to someone you trust. The client brief that gets rewritten three times because the intake form asks the wrong questions.

You cannot write a useful list against the map. You can only write it against the ground truth, and getting to it takes conversations rather than a form.

The five items below are the ones we almost never see written down. Answer them and the technical rows finally have something to aim at.

Item one: whose Tuesday changes.

Name the people. Not the department and not the function. The roles, and ideally the actual humans. The claims handler on the west coast team. The three brand managers who own the seasonal campaign. The two people in finance who close the month.

When a leader cannot do this, the program is still an idea. Naming forces the useful questions straight away: what does this person do at nine in the morning now, what will they do instead, who has told them, and when. Plenty of programs discover in month three that the answer to the last two questions is nobody and never.

Item two: who is already using AI without telling you.

Somebody in your organization has been using AI for months. Personal account, free tier, a browser tab that gets closed when a manager walks past. They were not waiting for the program to start.

The instinct is to treat that as a policy breach. It is more useful as information: shadow use tells you where the friction actually sits, which tasks people find intolerable, and which of your official use cases nobody cares about. It is the cheapest research you will ever get.

Put it on the list. Find the quiet users, ask them what they are doing and why, and decide what to do about the risk they have been carrying on their own.

Item three: what happens to the work that used to teach people.

The tasks first in line for automation are often the ones that used to train people. Summarizing the call. Drafting the first version. Checking the numbers. Sitting in the meeting and writing it up afterward.

That work was never efficient. It was how a junior person learned what good looks like and which client says one thing and means another. Remove it without replacing it and you have taken out a rung of the ladder.

The cost does not show up this quarter. It shows up in about three years, when the people who should be ready for the next level have never made a hard judgment call with something at stake.

The item itself is short. For each task you automate, write down where that learning now happens instead. If the line stays blank, you are still making a decision, and it is better made deliberately.

Item four: who is allowed to say no.

Ask any program lead whether people can push back and the answer is yes, of course, we want the feedback. Then ask what happened the last time somebody actually did.

Trust debt builds when the invitation is real and the response never comes. People stop raising things, the honest conversation moves to the corridor and the private message, and the official channel starts returning good news that nobody in the building believes. What gets read as resistance to the technology is usually much older than the technology.

So this item is about a mechanism, and a working one. Who can stop this, on what grounds, where an objection goes, and how quickly the person hears back.

Item five: what the organization stops doing.

Almost every plan adds. New tool, new training, new reporting, a champions network, a community of practice, a monthly showcase.

Then people are asked to learn a new way of working in the same week they deliver everything they were already delivering, and the new way becomes a second job done at the end of the day, badly. Adoption slows, and the usual response is more training.

Write down what stops. A report nobody reads. A meeting that exists to relay information. A manual step the tool now covers. A review layer that was there because the first draft was always rough. If you cannot name a single thing, that is worth knowing before launch rather than after.

Running the list without building a program around it.

None of this needs a workstream. It needs one page and roughly three conversations: the leader accountable for the outcome, a manager in the middle of the affected work, and two or three of the people whose Tuesday changes. An hour each. Write the answers where the steering group can see them next to the technical rows.

Cisco's 2025 AI Readiness Index found that only 13% of organizations are fully ready to use AI, a figure that has stayed flat for three years. Three years of better models, cheaper access and enormous spending have not shifted it. That is a strong hint about what is actually being measured.

The technical checklist still stands. Security review, data work, licenses, integration, training, all of it has to happen and none of it is optional. The five items just have to come first.

A reasonable next step.

If you are looking at a plan right now and cannot answer those five questions, you are in ordinary company. The answers take about a week to gather, and they usually change what the plan should contain.

The AI Profit Readiness Assessment covers the same ground in about two minutes and gives you a clear read on where your organization stands before you commit another quarter to a plan. If you would rather work through it with someone, book a discovery call here.