Halfway through the year is the right time to ask the question most managers avoid: does my team actually know how to use AI tools — or do they just know enough to seem like they do?

The gap matters more than it used to. If your H2 plans assume your team will use AI to do more with the same headcount, you need to know what their actual skill level is before July, not after.

Here's how to run a meaningful mid-year AI skills audit in about 2–3 hours of total effort.

What a Skills Audit Actually Is (and Isn't)

A skills audit is not a survey. Surveys tell you what employees think their skill level is. Those answers are almost always optimistic — people overestimate their proficiency at tasks they use occasionally, and underestimate their proficiency at tasks they use constantly without realizing they're doing them.

A skills audit uses structured assessment to identify what employees can actually do, role by role, skill category by skill category.

The skills you're assessing aren't abstract. They're practical: - Can they construct a prompt that gets useful output for their specific job function? - Do they know when to verify AI output before using it — and how? - Do they know what data is safe to include in a prompt and what isn't? - Can they apply AI to the 3–5 highest-leverage tasks in their role? - Do they understand the limitations of the tools they're using?

These are the skills that determine whether AI helps your team or creates more cleanup work than it saves.

Step 1 — Define What "Good" Looks Like for Each Role

Before you audit, you need a baseline. For each role in your team, write down:

  • The 3 tasks most likely to be done faster or better with AI
  • The primary risk area for that role (data handling, output verification, client-facing quality)
  • What "competent AI use" looks like for that role — specifically, not generally

This takes 20–30 minutes for most SMBs. You don't need a comprehensive job analysis. You need to be clear enough that an assessment result is meaningful when you see it.

If you're formalizing those role expectations for the first time, how to build an AI learning path for each role is the practical companion.

Example for a customer service role: - Tasks: Ticket drafting, escalation triage, knowledge base lookup - Primary risk: Customer data in prompts; AI-generated responses sent without verification - Competent use: Drafts responses that require only minor edits; knows what can go in a prompt; verifies product details before sending

Example for a finance/ops analyst: - Tasks: Report summarization, data interpretation, internal documentation - Primary risk: Client data in prompts; unverified AI-generated figures in client materials - Competent use: Uses AI for drafts, not final outputs; always verifies numbers; never pastes client financials into consumer AI tools

Step 2 — Run the Assessments

The assessment for each employee should take 15 minutes. It should cover:

  1. Practical AI knowledge — What does this employee know about using AI tools in their workflow?
  2. Data handling — Do they know what should and shouldn't go into a prompt?
  3. Output verification — Do they know when and how to check AI-generated content?
  4. Role-specific application — Can they apply AI to the tasks most relevant to their function?

If you don't have a structured assessment tool, a simple skills inventory works: give each employee 10–12 scenario questions relevant to their role and have them respond. You'll learn more from a focused scenario set than from a generic "rate yourself 1–5" survey.

OpenSkills AI runs this assessment automatically by role — the employee completes it in 15 minutes, and you see results immediately in the admin dashboard.

Step 3 — Read the Results at Two Levels

Once assessments are complete, look at the data at two levels:

Individual level: - Who are your strongest AI users? These are candidates for peer coaching or expanded responsibility. - Who has the most critical gaps? Prioritize based on risk exposure (a billing staff member with a data-handling gap is a higher priority than a warehouse coordinator with a prompt-quality gap). - Who has gaps that will block H2 objectives? If you're planning to use AI for a specific workflow in Q3, who specifically isn't ready to execute it?

Team level: - What's the most common gap across the team? This likely points to a training topic that wasn't covered in onboarding. - Are there role-specific patterns? If all your customer service staff score low on output verification, that's a workflow problem as much as a training problem. - Is there a gap between stated proficiency (from past surveys or self-reports) and actual assessment scores? A large gap means your team doesn't know what they don't know — which means mistakes are happening without visibility.

Step 4 — Build H2 Learning Priorities From the Data

The output of an audit shouldn't be a report that gets filed. It should translate directly into two or three H2 priorities:

Priority 1: The highest-risk gap. Whatever gap creates the most liability or customer-facing risk, address it first. This is usually data handling or output verification for compliance-sensitive roles.

Priority 2: The highest-leverage gap. Whatever skill improvement would deliver the most productivity gain. This is usually role-specific AI application — the difference between using AI generically and using it well for the 3 tasks that dominate that role's workday.

Priority 3 (optional): The onboarding gap. If your newest hires are consistently weaker than your experienced staff in ways that don't reflect the experience difference, you have a structured onboarding gap. Fix the onboarding, and every subsequent hire lands better.

Step 5 — Set a 30-Day Checkpoint

An audit is a snapshot. Set a 30-day checkpoint to measure whether the learning priorities are moving.

At 30 days, re-run abbreviated versions of the assessment questions that targeted the highest-priority gaps. You don't need a full re-assessment — you need enough data to know whether the gaps are closing.

If they are: continue. If they aren't: the training approach isn't working, and you need to understand why before putting more time into it.

For the scorecard side of that checkpoint, how to measure learning ROI without a data analyst shows what to track without building a full analytics stack.

What It Costs to Skip This

Most teams skip the mid-year audit because it feels like a nice-to-have — an investment in preparation before the "real work" of H2 starts.

The cost shows up in two ways:

H2 plans built on false assumptions. If you plan to use AI to speed up a process and the person executing that process has fundamental gaps in how they use AI tools, your plan is wrong. You'll find out when the process fails, not when you're planning.

Skill gaps that compound. An employee who uses AI incorrectly for 6 months develops confident bad habits that are harder to correct than gaps caught at 3 months. The longer the audit is deferred, the more entrenched the incorrect patterns become.

A 2–3 hour mid-year audit is the cheapest way to make sure your H2 assumptions are built on something real. If your bigger concern is the hidden spread between strong and weak adopters, the learning gap between your best and worst AI user is the next read.

OpenSkills AI runs role-specific AI skills assessments automatically. 15 minutes per employee, results in your admin dashboard the same day. Start your 14-day free trial →

Related reading: - How to Build an AI Learning Path for Each Role - How to Measure Learning ROI Without a Data Analyst - The Learning Gap Between Your Best and Worst AI User