AI Work Manual · AI Usage SOP
Departmental AI Work Manual and Standard Operating Procedure
What is an AI work manual?
- Guidelines set the boundary; the SOP sets each department’s procedure
- An SOP without a review step is just an automation sign-off
- Written by the department, rolled out through hands-on practice, reviewed every six months
An AI usage SOP changes AI use from an individual skill into a departmental procedure. In a company with only guidelines, some people on a team use AI every day while others never touch it, and when a skilled user moves on, their know-how leaves with them. Once “use this tool this way at this step of this task” is written into a procedure, even a new hire can work at the same level from week one, and training becomes simply teaching the procedure. At next year’s planning season, the gap in adoption between departments with an SOP and those without shows which training tasks to prioritize.
A single SOP contains five things: the task and the steps it covers, the approved tools and accounts, the data that may and may not be entered, the standard prompt or template, and the person accountable for reviewing the output (accountability) along with the review method. An SOP that omits the output review step is not a procedure but an automation sign-off. Set the level by task risk. For tasks that are easy to undo if wrong, such as meeting-minutes summaries, a self-check is enough. For outputs that leave the company, such as customer-facing documents or contract reviews, a second person’s check and a comparison against the original must be built into the procedure. The department itself, not the training team, should write it. The structure that lasts longest has the training team supply the template and review criteria while the department’s AI champion writes the first draft.
An SOP sits at a different level from the internal AI usage guidelines, a prompt library, or a case collection. Guidelines set the boundaries the whole company follows, a case collection offers stories for reference, and an SOP is the procedure a department must follow. An SOP is what a department builds to fit its own work, within the range the guidelines allow, so when the guidelines change, the SOP has to change with them. Because tool features change quickly, the document should include a review schedule: every six months, run the procedure for real and fix any step that no longer works.
The most common misstep in practice is making the SOP a single company-wide document. Drafting a sales proposal and summarizing a production work order involve different tools and different risks, and when both sit in one document, no department treats it as its own. Another is writing an SOP and not training people on it. A procedure doesn't work just by being posted on the intranet; without a hands-on run-through, it stays on paper. It is also common to build in a review step only to have the reviewer approve AI output as is. That is a reviewer literacy problem, not a procedure problem, so schedule reviewer training together with the SOP rollout. If procedures are too granular, employees drift back to shadow AI, using personal accounts outside the SOP.
John KimInstructor John Kim, author of B2B Marketing Design and CEO of Nanum Consulting
Seolyeong ParkGenerative AI and AX training specialist with 17 years of hands-on experience and 200+ lectures, helping non-technical staff put AI to work right awayThis is the standard operating procedure (SOP) from manufacturing and quality, adapted for generative AI use in office work. Experience showed that company-wide AI usage guidelines alone don’t lead to real adoption in each department, so putting department-level procedure documents beneath the guidelines became standard practice.
Consider a hypothetical case. Company A, a medical device distributor with 560 employees, had rolled out company-wide AI guidelines a year earlier. Even so, some people on the sales team used AI and others didn’t, and when a senior manager who used it well left, the team’s proposal-writing time went back up. In October, the training team started drafting SOPs for each department.
- Ms. Kang, a deputy manager on the training team, created a form with five fields: “target task, permitted tools, data that must not be entered, standard prompt, and review owner.” She handed the first drafts to the AI champions in three departments: sales, purchasing, and customer support.
- The sales team’s draft had a procedure for drafting proposals but left the review step empty. Ms. Kang, the training team’s deputy manager, said, “As it stands, pricing terms written by AI go straight to the customer,” and had them add a step to check against the team lead’s original.
- The customer support team split its SOP into two levels by risk: a self-check for complaint reply drafts and a second person’s check for refund notices. All three department SOPs wrote a semiannual review schedule into the document itself.
- In the same week the SOP was released, each department held a 2-hour hands-on session where everyone ran through the procedure once. The 9 team leaders who act as reviewers also got a separate 1-hour training on how to review AI outputs.
- Three months later, the average time to draft a sales proposal fell from 3.2 hours to 1.4 hours, and weekly usage across the three departments rose from 31% to 74%. Next year, the company will extend SOP coverage to the remaining 7 departments.