Skip to content

Insights / AI Training

Staff AI Training Program: A 30-60-90 Day Plan That Sticks

Build a staff AI training program with role-based tracks, a 30-60-90 day plan, acceptable use rules and measures for adoption and quality.

Private AI

A clearer view of what comes next.

Most organizations buy AI licenses before they train anyone to use them. A few enthusiasts get real value, most staff try it twice, and a quiet minority pastes sensitive material into whatever tool is open. A training program fixes that by making the safe, useful path the easy one.

This guide is for the IT director, operations leader or school or agency administrator who has to turn “we should do something about AI” into a plan with owners, dates and measures.

Start With Roles, Not a Single Course

One all-hands session cannot serve a records clerk, a teacher, a finance analyst and a help desk technician. Group staff by what they do with information, then train each group on its own tasks.

Role group What they need to learn Typical first use
All staff Acceptable use, data handling, checking output Drafting, summarizing approved material
Managers Setting expectations, reviewing AI-assisted work Team workflows, status reporting
Records, finance, HR, legal Handling regulated or sensitive information Only approved tools and inputs
Technical staff Evaluation, connectors, logging, prompt injection Scripts, ticket triage, internal tools
Educators and program staff Student data rules, age-appropriate use Lesson planning, family communication

The NIST AI Risk Management Framework is a voluntary framework aimed at building trustworthiness into how AI systems are designed, used and evaluated. Its core functions (Govern, Measure, Manage, Map) are a useful frame for a training owner: someone must govern the rules, map where AI is used, measure results and manage problems.

Write Acceptable Use Before You Teach It

Training that contradicts policy, or has no policy behind it, teaches staff to ignore both. Keep the first version to one page:

  • Which AI services and account types are approved, by name.
  • Which categories of information may never be entered (credentials, student records, health information, anything under contract restrictions).
  • Which uses need a manager or data owner to approve first.
  • Who must review output before it leaves the organization.
  • How to report a mistake, with no penalty for reporting promptly.

For the data side, our shadow AI data checklist gives staff a decision they can make before they upload anything.

The 30-60-90 Day Plan

Days 1 to 30: Foundation. Name an executive sponsor and one program owner. Publish the acceptable use page. Run a short baseline survey: which tools do people already use, for what, and where do they feel stuck? Train a pilot group of 10 to 25 people across at least three roles. Give each person two or three real tasks to try, and collect examples that worked and examples that failed.

Days 31 to 60: Role-based rollout. Use the pilot results to write role-specific guides: approved prompts, a checklist for verifying output, and examples of inputs that must not be used. Train the remaining staff in role groups, in short sessions with hands-on practice on their own tasks. Appoint a few “champions” per department who answer questions and report patterns back to the program owner.

Days 61 to 90: Embed and measure. Move from classes to routine. Add a short AI item to new-hire onboarding. Review the first measures (below) with leadership, retire guidance that no one uses, and decide where deeper work is justified: a connected assistant, an automation, or a private deployment for sensitive data.

Where staff need structured courses, Z7 Academy offers short, practical courses for government and education teams with knowledge checks and a certificate on completion. A course supports a program; it does not replace your own policy and role guides.

Teach Verification as a Core Skill

The most common training gap is checking output. Staff should leave every session able to do three things: confirm facts against a source, notice when an answer is confident but unsupported, and recognize that text inside a document can try to steer an assistant. The OWASP Top 10 for Large Language Model Applications is a good reference for technical staff on the risks of LLM applications, including prompt injection.

Make verification concrete: for each role, define what “reviewed” means. A grant narrative needs a person to check every figure. A parent letter needs a check of tone and policy. A summary of a long report needs a spot check against the source pages.

Measure Adoption and Quality Separately

High usage with poor quality is a risk, not a win. Track both.

Measure How to collect it What it tells you
Trained by role Training records Coverage gaps
Active use by approved tool Admin reports from the approved service Whether the approved route is actually used
Task time on sampled workflows Before and after sample from pilot staff Whether it helps real work
Review findings Sample of AI-assisted work checked by a reviewer Error types to fix in training
Policy questions and incidents Help desk and reporting channel Where guidance is unclear
Staff confidence Short pulse survey at 30, 60, 90 days Whether people feel able to use it safely

Keep the sample small and the method consistent. Avoid quoting a productivity number you have not measured in your own environment.

Common Mistakes

  • Training on a tool the organization has not approved, or whose settings are unreviewed.
  • Using generic examples instead of tasks from each department.
  • Skipping managers, who set the real expectations.
  • Treating the program as a one-time event instead of a quarterly refresh as tools and rules change.
  • Ignoring the technical staff who will be asked to connect assistants to business systems.

Where Z7 fits

Z7 runs AI training and change management alongside its AI services, including role-based sessions and prompt libraries tuned to a client’s work. If you want a 90-day plan built around your roles and data rules, book a conversation.

Put it into practice

Illustrative planning model

A useful path to private AI

Explore step 01

Choose a workflow

What should become easier?

Start with one repeatable task and the person responsible for its outcome. Define what a useful answer would look like before choosing a model.

Z7 Weekly Intelligence

We aim for one email a week: the threats that matter, new Z7 intelligence briefs and articles, and what to do about them. Unsubscribe anytime.

By subscribing, you agree to our Terms of Use. Our Privacy Policy explains how we use your email.

Put your AI questions on the calendar.

Bring your use case, data concerns and budget questions. We will talk through what fits, what it costs and a practical next step.

Ask Zeus AI assistant