Start with the operational tasks where a wrong answer stays inside your own office, and climb to riskier ones only when you can name the check that catches the error and the person who runs it. That is the whole argument of this article, and the rest is how to apply it.
An Education Week Research Center survey reported on October 30, 2026 that nearly one-third of district and school leaders use AI for no operational purpose at all. Most coverage of AI in schools is about classrooms. The office where a leader schedules, communicates, evaluates and plans gets far less attention, and it is where your own judgment is most exposed, because you cannot hand the blame to a teacher or a vendor.
What are school leaders actually using AI for in operations?
Among leaders who do use it, the most common operational uses are school or district communications, foreign-language translation, strategic planning, employee evaluation and class scheduling, according to the same EdWeek report. The survey covered 852 educators nationally, 262 of them district and school leaders, and ran from August 26 to October 6, 2026. The article gives the ranking in a chart I could not read, so I am not quoting per-item percentages.
Ronn Nozoe, CEO of the National Principals Association, told EdWeek that student privacy is the big issue and that he was not surprised so much of the use is about communications. He also warned that otherwise "everybody's iterating on their own, and that's probably not the best use of everyone's time."
Look at that list again. Proofreading a newsletter and evaluating an employee sit side by side as "operational use," but a wrong answer in one is a typo and in the other is a decision about a person's career. Treating them as one category is how a sensible leader ends up either banning everything or approving everything.
Why does the same tool carry different risk in different tasks?
Because risk follows what the output touches, not which tool produced it. A chatbot summarizing a policy you wrote and a chatbot ranking your staff are the same software doing very different jobs.
I suggest a simple way to sort the work, which I call the Consequence Ladder. This is my own way of thinking, not a validated instrument. The Consequence Ladder has three rungs, defined by where the output goes and what happens if it is wrong:
| Rung | Where the output goes | Examples | The check that must exist |
|---|---|---|---|
| 1. Stays in your office | Nowhere. You read it and decide what to keep. | Proofreading a policy, summarizing a long document you have read, a first-draft agenda, a thinking partner for your own planning notes | You compare it with the source you already know |
| 2. Leaves the building | Parents, staff, the community | Newsletters, parent letters, translations of notices | A named person reads the final text before it is sent |
| 3. Shapes a decision about a person | A file, a timetable, a ranking, a flag | Employee evaluation, scheduling that places specific students or staff, attendance or absence flags, hiring screens | A person who could make the same call without the tool, a written record of why, and advice from your HR or legal advisers |
The rule is one sentence. Climb a rung only when you can say, for the rung you are on, what the check is and who does it. If you cannot name the person, you are not on that rung yet; you are guessing.
Strategic planning is a good test of the ladder, because it moves. Asking AI to challenge a draft plan built from non-sensitive material is rung 1. Asking it which staff roles to cut or which students to move is rung 3, whatever the tool's interface calls it.
What does Rung 2 need that leaders skip?
It needs a human who reads the final words, and for translation, someone who can read the target language. Leaders reach for AI translation because the need is real and the alternative is often nothing. The risk is that a fluent-looking translation of a safeguarding or attendance notice can be wrong in a way no one in the building can see.
Two habits keep Rung 2 honest. First, keep student names and identifying details out of the prompt unless the tool has been through your school's approval process; the posts on approving AI tools for schools and on whether staff can put student data into ChatGPT set out how. Second, send routine notices through AI translation if you must, but have a fluent speaker check anything a family has to act on.
Who is responsible when AI gets an operational decision wrong?
The school is, and the English inspectorate says so plainly. In its guidance on how it looks at AI, Ofsted states that "any evaluation that Ofsted makes is about the provider's decision-making," not about the tool itself. The same document says Ofsted does not directly evaluate the use of any AI tool, does not require inspectors to look for AI use, and does not expect providers to use AI in a particular way or at all. Ofsted's page was last updated on October 21, 2025, and applies to England; I re-read it on November 20, 2026.
Its examples are operational: a school using AI to look at the causes of absence, and support staff using AI to help take case notes. In both, inspectors may ask how the provider made sensible decisions. I read that as the Rung 3 test arriving from the outside: can the leader explain the decision, the check and the owner? The principle travels beyond England. A board member or a parent who asks about a decision will not accept "the tool did it" as an answer.
For related questions of accountability, see who is accountable when AI gets it wrong and which decisions AI should never make.
What does the evidence say about starting small?
The EdWeek survey suggests AI is already in the office before the office has a shared rule for it. That is a common shape in schools: the classroom gets the conversation first, and leaders' own use of AI for letters, summaries and planning notes goes unspoken. The gap matters because a reader of an operations report, whether a governor, trustee or board member, does not need to know which tool produced it. They need to know who checked it. A leader who can say "I used it to draft, and I read every line" is in a very different position from one who cannot say how a number got into the report.
The Ofsted study of early adopters, commissioned by the Department for Education and published on June 27, 2025, interviewed leaders of 21 early-adopter schools and colleges in England. It found that they tended to take small, cautious first steps, such as internal trials and staff working groups, before adopting AI more widely, and that very few had a longer-term strategy beyond the piloting stage. Many had not defined what success would look like, and the report notes there is little reliable evidence on AI's impact on outcomes. It is a small qualitative study of leaders who had already chosen to adopt, so it describes careful early users, not schools in general. Even so, the habit it describes, starting small and checking, is the Consequence Ladder in practice.
What does a first month look like? (a hypothetical)
Here is a hypothetical principal, invented for illustration, who wants to use AI for operations without creating a problem.
In week one she lists every operational task she already does with AI or would like to, and places each on a rung. Proofreading and agenda drafts land on Rung 1. The weekly parent letter and the translated attendance notice land on Rung 2. A request from a colleague to use a tool to help rank applicants lands on Rung 3, and she says not yet.
In week two she names the check for each Rung 2 task and who owns it: the office manager reads every outgoing letter, and a bilingual staff member checks translations. In week three she writes one paragraph for staff saying what is fine, what needs a named reader, and what is not for tools at all. In week four she asks a different question: where did a check catch something, and what did it catch? That number, errors caught, tells her more about whether the process works than hours saved.
Nothing in that month required a new policy document, a procurement round or a system-wide rollout. It required a list, a name beside each check, and a paragraph.
Where should you begin on Monday?
Begin by writing down what you and your office already do with AI, because you are almost certainly using it somewhere. Then:
- List every operational task and place it on a rung of the Consequence Ladder.
- For each Rung 2 task, name the person who reads the final text before it leaves the building.
- For each Rung 3 task, decide whether it stays off the list for now, and tell your team in writing.
- Check what your jurisdiction and your HR advisers say about automated tools in decisions about employees before any Rung 3 use.
- Pick one error-catching measure to track for a term.
If your leadership team is working through these questions, this is the kind of work I support through AI strategy sessions for schools and advisory work.
Sources and further reading
- Langreo, L. "Nearly 1 in 3 District and School Leaders Skip AI for Operations. Here's What the Rest Do." Education Week, Education Week Research Center, October 30, 2026. https://www.edweek.org/technology/nearly-1-in-3-district-and-school-leaders-skip-ai-for-operations-heres-what-the-rest-do/2026/10
- Ofsted. "How Ofsted looks at AI during inspection and regulation," part of "Ofsted's approach to AI." GOV.UK, last updated October 21, 2025 (England). https://www.gov.uk/government/publications/ofsteds-approach-to-ai/how-ofsted-looks-at-ai-during-inspection-and-regulation
- Ofsted, commissioned by the Department for Education. "'The biggest risk is doing nothing': insights from early adopters of artificial intelligence in schools and further education colleges." GOV.UK, June 27, 2025 (England). https://www.gov.uk/government/publications/ai-in-schools-and-further-education-findings-from-early-adopters/the-biggest-risk-is-doing-nothing-insights-from-early-adopters-of-artificial-intelligence-in-schools-and-further-education-colleges
Dan Fitzpatrick is the founder of The AI Educator, a former assistant headteacher and a school trustee, and has trained more than 150,000 educators across 30+ countries.


