# Training Module: Leadership Education

By Cody Saunders, LMSW · October 11, 2026

## Lead from the work your team needs

A useful AI decision begins with a real work problem. A tool may help a team finish a task with less effort. It may also shift effort into review, training, and correction. Leaders need a way to see both sides before expanding use.

This free training helps behavioral health leaders plan a small, accountable trial. Participants define the purpose, assign roles, choose measures, and decide what evidence is needed for the next step. The focus is leadership judgment: where to invest time, how to protect care, and how to make change manageable for staff.

## Audience and learning goals

This 75-minute module is for practice owners, supervisors, operations leads, and clinical leaders. Invite privacy, IT, finance, and frontline staff when possible. Their views help reveal effort and risks that a product demo may miss.

By the end, participants should be able to describe one work problem without jumping to a product; assign the people who own the trial and its final decisions; compare total effort and quality; and explain a reason to hold, limit, expand, or pause use.

This lesson does not approve a vendor, provide legal advice, or certify a leader. It does not promise continuing education credit. The trial steps are teaching proposals that need adaptation to the host’s setting.

## Facilitator preparation

Bring the fictional trial cases below, paper for a one-page plan, and the host’s current approval process. Confirm the contact for data and incident questions. Do not use client records or a vendor account that has not been approved for the session. Live AI use is optional; the lesson works with printed material.

Ask the host to choose a goal such as less after-hours paperwork or clearer staff instructions. Do not collect private staff or client details. When participants discuss their own work, keep the discussion at the process level. Make room for staff to explain what extra review feels like in a busy day.

## Session timing

**0–10 minutes: Define the problem.** Compare a vague goal with a task-specific one. Ask participants what evidence would show that the problem improved.

**10–25 minutes: Name roles and limits.** Teach the trial plan below. Have the group assign a clinical owner, data reviewers, a workflow owner, and a final decision-maker to the first fictional case.

**25–40 minutes: Evaluate total value.** Use Activity 1. Ask participants to calculate full effort and identify missing measures.

**40–55 minutes: Make a bounded decision.** Use Activity 2. Work through the five domains of the existing framework without turning them into a score or a seal of approval.

**55–65 minutes: Plan for a change.** Use Activity 3. Practice what happens when a feature, recipient, or reviewer changes.

**65–75 minutes: Write a plan and teach it back.** Each group drafts a short trial plan and explains its limits. Use the completion check below. End by naming a real next step within the host’s authority.

## Define success before choosing a tool

“Use more AI” is not a useful outcome by itself. “Reduce time spent rewriting approved public staff guides while keeping every required step” is easier to test. It names the task and the quality that must stay intact.

For care work, include the patient relationship. A visit-note tool could make room for attention during a session, or it could make a client uncomfortable. Ask how patient experience will be heard. Do not infer better care from shorter notes or a lower time count alone.

For billing work, accuracy matters as well as speed. A draft that adds unsupported facts can create more work and a poor record. Any trial must preserve accurate documentation and the right review process. Faster drafting is not evidence of proper billing or higher payment.

## Build a small trial plan

Name the purpose, exact task, tool, account, feature, and allowed information. State who can use it and what they must do before releasing a result. A trial should have a start, a review date, and a way to end.

Assign a workflow owner who knows the daily task. For care uses, name the qualified clinical reviewer and final record owner. Privacy, security, legal, and other staff should review the parts within their roles. Name the leader who can approve the next stage and the person who can stop the trial when a limit is crossed.

Agree on a fair comparison. Use similar tasks and count the whole workflow: drafting, checking, fixing, and filing. Include training, access, and support costs where relevant. Ask staff whether the work is easier, harder, or merely different. Do not use speed alone as a reason to expand.

Set stop conditions before starting. Examples include an unapproved data request, a serious error, loss of the assigned reviewer, or a changed data recipient. Explain the approved alternative so staff can keep working during a pause. These are proposed trial controls, not a universal legal checklist.

[NIST’s AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) offers voluntary guidance for managing AI risks through defined roles and ongoing review. It is not law or product approval. The task-specific plan here is a teaching method; it does not claim to validate a tool.

## Activity 1: Compare the whole effort

A fictional clinic tests a tool for public staff guides. In ten similar tasks, the usual process takes 300 minutes in total. The AI process takes 80 minutes to draft and 170 minutes to review and correct. Staff spend another 60 minutes setting up the trial. Both groups meet the same final quality check.

Ask participants to calculate the direct task time and the first-trial total. Should the leader say the tool has already saved time? What else should they ask?

**Answer and debrief:** AI task work takes 250 minutes. Including setup, this trial takes 310 minutes, compared with 300 minutes for the usual task work. Setup may be spread across later tasks, but the leader should report it honestly. This short fictional trial does not establish long-term savings. Ask about ongoing costs, error types, staff effort, and whether the tasks were truly comparable. If quality differs, the time comparison alone is not enough.

A plan could track the next group of approved tasks before deciding. It should not hide review work or use these made-up numbers in a public benefit claim.

## Activity 2: Use the existing five-domain framework

A fictional clinic wants to trial AI note drafts. The vendor describes its data practices, but the clinic has not finished reviewing the contract. A clinician is willing to review each draft, though no protected review time has been planned. The only demo used one simple made-up case.

Ask each group to name one missing piece in each domain of the [Behavioral Health Responsible AI Framework](/resources/behavioral-health-responsible-ai-framework/): Clinical Appropriateness, Privacy & Security, Reliability & Safety, Human Oversight, and Governance & Accountability.

**Answer and debrief:** Clinical fit needs a defined task and limits. Privacy and security need the actual service and data path reviewed, including required agreements. Reliability needs broader task-relevant tests and a way to capture failures. Human oversight needs time, authority, and a backup reviewer. Governance needs owners, a review date, and pause rules. The case supports holding the proposed real-data trial until these gaps are resolved. A demo is not evidence of care benefit, and a willing reviewer does not establish a workable review process.

HHS explains obligations for covered organizations and cloud services handling protected health information in its [cloud computing guidance](https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html). A business associate agreement may be required along with other safeguards. The responsible team must determine what applies. Completing this exercise does not establish compliance.

## Activity 3: The workflow changes

A fictional team completes an approved public-text trial. Later, the vendor adds a visit-recording feature. The trial lead also leaves the team. Staff want to continue because the product is already on the approved list.

Ask groups to decide which use can continue, what needs a new review, and who should receive the concern.

**Answer and debrief:** The new recording use needs review before it begins. Product approval does not automatically cover a new feature or private information. The departure also leaves a gap in ownership. The leader should assign a new qualified owner and confirm the existing controls before continuing the trial. If those controls cannot be maintained, pause under the agreed plan. Use the normal approved process while the gap is resolved.

A feature change is a chance to revisit purpose and data, not merely a technical update. Give staff a clear message about the current allowed use and a way to ask questions.

## A one-page planning exercise

Have each group write a short plan with these headings: problem and goal; approved task and information; owners and reviewers; quality checks; full effort and costs; staff and patient feedback where relevant; stop conditions; and review date.

Ask another group to read it and explain the next action. If they cannot tell who checks the final work or what data is allowed, revise the plan. This peer check tests clarity, not legal sufficiency or tool safety.

A strong plan keeps the first step small enough to learn from. It states what remains unknown and who will resolve it. Leaders do not need to promise a large change before they know whether the tool helps.

## Completion check and follow-up

Ask participants to explain why a faster draft may not lower total costs, what a reviewer needs to do the job, and which changes trigger another review. Listen for full effort, time and authority, and the exact task, data, feature, and people.

For follow-up, bring one proposed plan to the actual review team. Use the [purchasing checklist](/resources/before-you-buy-ai-tool/) and [data-handling worksheet](/resources/ai-data-flow-map/) rather than creating duplicate tools. Choose a review date before any approved trial begins. The goal is evidence that supports better work and care, with clear responsibility for the next decision.

## Sources and scope

Sources checked October 11, 2026. All cases and figures are fictional. Suggested trial steps are not a validated assessment. No savings, billing, care, certification, or continuing education outcome is promised.

[NIST AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework): voluntary guidance first released in January 2023. NIST notes that version 1.0 is being revised.

[HHS guidance on HIPAA and cloud computing](https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html): explains duties for covered organizations and business associates; it does not approve a specific product.

For a fuller team process, read the [governance article](/resources/responsible-ai-governance-behavioral-health/). To discuss free leadership training, visit [Work With Me](/work-with-me/).