What a license count can't tell you, why most teams find out the hard way, and a free template that shows you instead of making you guess.
An AI skills matrix maps your people against the AI skills that actually matter, everyday tool use, prompting for the role, judgment and verification, role-specific workflows, data and privacy judgment, and teaching others, scored on a simple 4-level scale (Not Yet, Getting By, Confident, Go-To). Build one before a rollout, six-plus months after one if usage looks uneven, or before you set a training budget. Skip it for a tiny team you already know well. A free, filled-in-example template (CSV and PDF) is linked in the template section below, no sign-up required.
Picture a mid-size professional services firm about six months after it rolled out Copilot licenses to everyone on staff. The launch email went out, the training webinar ran once, and IT can tell leadership exactly how many people logged in last week. What nobody can answer, when the CFO asks where the training budget went, is which three people on her own team could use the thing for anything beyond drafting an email.
That’s the part a license count can’t see. A handful of people, mostly in marketing, are using it daily and getting real value. Most of finance and operations opened it once, stared at a blank chat box, and went back to doing things the old way. And somewhere in the building, one person has quietly become the unofficial AI help desk, the one colleagues Slack with “can you check this prompt for me,” without anyone in leadership noticing or naming it.
Handing out licenses is an access strategy, not a skills strategy, and the two get treated as the same thing constantly. Access is not capability. The gap I see most often when a rollout goes company-wide is this: everyone gets the same tool on the same day, and six months later the team is more unevenly capable than before the rollout started, not less, because the confident got more confident and the hesitant got left further behind with nobody checking in. It’s part of why upskilling employees on AI isn’t optional anymore, but upskilling only works if you know who actually needs it.
Skill gaps are the single most-cited barrier employers name when asked what’s stopping workforce transformation, with 63% calling it a major barrier through 2030, according to the World Economic Forum’s Future of Jobs Report 2025.[1] That’s the global workforce, not your building specifically, but it implies the same thing at any scale: an organization that thinks it has an AI adoption problem usually has a visibility problem first. Nobody closes a gap they can’t see.
A license tells you who logged in. It never tells you who can actually do the work.
An AI skills matrix is the tool that closes that visibility gap. It’s not complicated, and it’s not a certification program. It’s a structured way to answer the question every leader ends up guessing at six months into a rollout: who can actually do what, with AI, right now, and what do we do about the gap.
An AI skills matrix is a grid. Your people, or the roles on your team, run down one side. The AI-specific skills that actually matter for the work run across the top. Every cell gets a proficiency score. That’s the entire structure, and it’s deliberately unglamorous, because the value isn’t in the format, it’s in what filling it in honestly forces you to see.
General skills matrices have been a workforce-planning staple for years, mapping a team’s competencies, project management, negotiation, a specific software platform, against what a role requires, usually to spot training needs or plan who covers what.[2] An AI skills matrix borrows that same grid, then narrows and speeds it up in two ways that matter.
A general matrix might have one row for “software proficiency” covering everything from Excel to Salesforce. An AI matrix needs to go narrower: everyday tool use is a different skill from prompting well for your specific role, different again from knowing when to double-check output before it goes out under your name. Lump them together and you lose the distinction you built the matrix to find.
Someone’s “advanced” Excel rating is a safe bet for two or three years. Someone’s AI proficiency can shift within a single quarter: a new feature ships, a workflow gets built around it, and what counted as “confident” in January looks thin by summer. A general skills matrix reviewed annually is usually fine. One reviewed only annually for AI is already stale by the time anyone reads it, which is why the last section of this guide is about keeping it alive, not just building it once.
There’s a role-specific piece worth naming here too. “The team needs AI training” is too broad to act on. A content marketer, a finance analyst, and a customer support lead hit completely different walls with AI: brand-voice guardrails for one, what’s safe to paste into a chat tool for another, a fast drafting workflow without losing accuracy for the third. A matrix that scores everyone against the same generic bar misses all of that.
Not every team needs a formal matrix, and building one where it’s not needed just adds a spreadsheet nobody updates. Four situations genuinely call for it, and they share a common thread: someone is about to make a decision and doesn’t actually have the information to make it well.
It’s overkill in the reverse: a team of three or four people you already know well doesn’t need a formal instrument to confirm what you can see, and neither does a single narrow tool used by one person for one task. The test isn’t team size, it’s whether a real decision is currently being made on a guess. If you’re earlier than this, still working through general AI hesitation, taking a team from AI awareness to AI fluency usually comes before a matrix makes sense at all.
| Situation | Build one now | Skip it for now |
|---|---|---|
| Rolling out a new AI tool company-wide | ✓ Baseline before licenses go out | |
| 6+ months post-rollout, usage looks uneven | ✓ Turns a vague impression into a plan | |
| About to set a training budget | ✓ Prevents spending on the wrong people | |
| Heading into a corporate AI training engagement | ✓ Lets the training fit your real gaps | |
| Team of 3-4 people you already know well | ✓ You can already see the gap without a form | |
| One tool, one person, one narrow task | ✓ Overhead outweighs the insight |
The test is whether a real decision, a rollout plan, a budget, a training brief, is currently being made on a guess.
A generic 1-to-5 scale is the most common first-draft mistake, tempting because it’s fast to build. The problem shows the moment two managers score the same person differently: what does a “3” actually mean? Nobody agrees, because the number carries no description of behavior, just a position on a line.
A couple of public frameworks already do this well. Sprad publishes a 5-level, 6-domain matrix tied to the AI-literacy obligation in Article 4 of the EU AI Act, useful if that’s your specific compliance need.[3] Rework goes shorter: three levels (Aware, Practitioner, Expert) plus a gap score, on the reasoning that more levels just create disagreement.[4] Both are worth a look if your situation matches what they’re built for. Neither is what this guide uses, but both converge on the lesson this one follows: shorter beats granular, and each level should describe a behavior, not a number.
Future Factors’ own scale keeps it to four levels, the ceiling this guide is working to on purpose, each one written as something you could observe, not infer:
| Level | What it looks like in practice |
|---|---|
| Not Yet | Hasn’t used AI tools for this task yet, or actively avoids them. Not a judgment, just a starting point. |
| Getting By | Uses it occasionally, often through heavy hand-holding or pure copy-paste habits. Output needs heavy checking before it’s trusted. |
| Confident | Uses it independently for real work. Knows when to trust the output and when to check it. Gets consistent value, task after task. |
| Go-To | Other people ask this person for help. Can teach or troubleshoot for someone else, and understands where the tool tends to break. |
Four levels, each one a description of what someone actually does, not a number on a scale nobody has agreed the meaning of.
Notice what “Go-To” describes: it’s the definition of an internal champion, written into the scale itself instead of left as an afterthought. That’s deliberate, and the section on using your results leans on exactly that distinction. For the fuller picture of what separates Confident from Go-To, it overlaps with what makes an AI-powered professional different: the tool matters less than which workflows someone’s built it into.
The columns are where most first attempts go wrong: too broad (“AI skills,” one column, scored once) or too granular (forty rows of tool features nobody has the patience to score honestly). Six domains cover the ground without either failure mode, broad enough to apply whether your team runs on Copilot, ChatGPT, Claude, Gemini, or some mix of all four:
Six is a ceiling, not a floor. Add a seventh domain only for a specific reason, not because more columns feel more thorough. A matrix that takes forty minutes per person to score honestly gets filled in once and abandoned, which defeats the point of building one.
An empty template teaches nothing. Here’s what five rows and six columns look like scored, so you can see the shape before building your own. These are illustrative roles, not real people, scored the way a manager would after a short conversation with each.
| Role | Everyday tool use | Prompting for the role | Judgment & verification | Role tool workflows | Data & privacy judgment | Teaching others |
|---|---|---|---|---|---|---|
| Content Marketing Manager | Confident | Confident | Confident | Go-To | Confident | Go-To |
| Sales Ops Analyst | Getting By | Getting By | Not Yet | Getting By | Confident | Not Yet |
| Customer Support Lead | Confident | Getting By | Getting By | Getting By | Getting By | Not Yet |
| Finance Business Partner | Not Yet | Not Yet | Not Yet | Not Yet | Confident | Not Yet |
| HR Generalist | Getting By | Confident | Confident | Getting By | Confident | Getting By |
Two rows worth a second look: the Content Marketing Manager (bold) is Go-To in two domains, a clear champion candidate. The Finance Business Partner (bold) is Not Yet in five of six, the clearest training case on the sheet, offset only by a Confident data-privacy score likely built from separate compliance training, not AI use. A completed matrix should read like this on sight, not need a workshop to interpret.
Building this honestly takes real time. AI can speed up the mechanical parts without replacing the judgment call at the center of it.
| What AI does | What you still own | How it gets checked |
|---|---|---|
| Drafts a short self-assessment survey from your six domains | Deciding whether a self-score matches what the manager observes | Manager reviews and adjusts every score before it’s final, no exceptions |
| Summarizes survey responses into a first-pass matrix | Deciding who’s a champion and who needs training | The final matrix is signed off by a person, not auto-published |
The free, blank version of this template is ready to download, with the same six domains and four-level scale already built in:
Both are free, no sign-up required, ready to copy into your own sheet and start scoring today.
A filled-in matrix is only useful if it changes what happens next. The three patterns from the interpretation table above map onto three different actions, and mixing them up is the fastest way to waste a training budget you’ve just gone to the trouble of protecting.
| What the matrix shows | What to actually do | What to avoid |
|---|---|---|
| Someone low across the board | Enroll in foundational training, paired with a specific, low-stakes task to practice on | Don’t put them in an advanced workshop hoping it “clicks” |
| A clear champion (Go-To in 2+ domains) | Name the role formally: office hours, first point of contact for their team, a line in their review | Don’t leave it informal. Unrecognized champions burn out or quietly stop helping |
| A team-wide gap in one domain | Run one shared session for the whole team on that specific domain | Don’t schedule five separate 1:1s for a problem everyone has |
This is also where role specificity earns its place again. A CMO reading this table isn’t the same audience as a content marketer or an ops lead, and the same result means something different for each: a budget decision for the CMO, an afternoon of guided practice for the content marketer, a judgment-and-verification gap for the ops lead whose work touches numbers that can’t be wrong. If your gaps cluster on one marketing team, practical ways to drive AI adoption on a marketing team picks up where a result like this one leaves off.
One honest caution: the matrix identifies where training should go. It doesn’t guarantee the training works. Training builds capability, someone can now do the thing in principle. That capability needs somewhere real to land, a recurring workflow not a one-off exercise, and it needs reinforcement, a manager who checks in, before it turns into adoption: the new way of working actually sticking. A matrix without a follow-up plan becomes a well-organized list of people who attended a session and went back to old habits.
When you check whether this worked, resist stopping at usage numbers. Are people still using it a month later without a reminder (persistence), and is the work better or faster because of it (impact)? “Logins were up” measures access again, the exact thing this guide started by arguing isn’t enough. A score moving from Getting By to Confident, still holding three months later, is the signal worth tracking, not who opened the tool this week.
A champion is someone you name formally. Left informal, they become free labor nobody thanked.
Where not to spend budget: on people the matrix already shows are Confident or Go-To in a domain. Obvious written down, and exactly what happens when training gets bought as a blanket rollout instead of scored first. A team can genuinely start this internally with nothing but the free template above and an afternoon. Where a corporate AI training engagement earns its cost is turning a matrix full of role-mapped gaps into a curriculum built for your team’s actual starting point, and building shared capability across the team instead of leaving it concentrated in one unofficial champion.
The most common failure mode isn’t building a skills matrix wrong, it’s building it once, filing it, and not opening it again for a year. Six months later the Sales Ops Analyst who was Getting By in prompting has quietly become Confident, nobody updated the sheet, and a manager makes a decision off data that’s already out of date.
A review cadence is the fix, and it doesn’t need to be elaborate:
| Check | Why it matters |
|---|---|
| Has the owner actually re-scored every existing row, not just added new hires? | Scores move in both directions; an unchanged row is a stale row |
| Is the Go-To column still accurate? | Losing your one named champion is the highest-priority gap to catch early |
| Has a new tool or workflow launched since the last review? | A new rollout resets several columns at once, not just for new hires |
| Did last quarter’s training actually move anyone’s score? | If nothing moved, the training wasn’t the right fit for the gap it targeted |
This is the same discipline behind building an AI upskilling program that doesn’t go stale, worth reading alongside this guide if you’re setting up the ongoing side of this rather than just the initial assessment. And if the matrix surfaces a genuine skills-versus-succession question, where someone’s AI proficiency changes who’s ready for a bigger role, that’s worth folding into how your team already uses AI for succession planning, not treated as a separate process running in parallel.
Pick one team and build the six-column version this week. Score it honestly, including the domain hardest to admit is weak. Then put a date on your calendar three months out to do it again. That second review is the one that tells you whether anything changed.
It’s a grid that maps your team members or roles against the specific AI skills that matter for their work, with a proficiency score in every cell. Unlike a login count or a license report, it shows you who can actually do what with AI right now, not just who has access to a tool.
A general skills matrix covers broad competencies like project management or a specific software platform, reviewed annually. An AI skills matrix narrows the columns to AI-specific behaviors, like prompting for a role or knowing when to verify output, and needs a faster review cycle, since AI tools change within a single quarter, not a couple of years.
Keep it to four levels or fewer, each one describing an actual behavior rather than a number on a scale. Future Factors uses Not Yet, Getting By, Confident, and Go-To. A generic 1-to-5 scale sounds more precise but usually means less, because two managers will disagree on what a bare number like ‘3’ actually represents.
Quarterly, with one named owner responsible for re-scoring every existing row, not just adding new hires. AI proficiency shifts faster than most workforce skills, so an annual review is typically already out of date by the time anyone reads it.
Yes. This guide includes a free, blank six-domain, four-level AI skills matrix template as both a CSV (opens directly in Excel or Google Sheets) and a one-page printable PDF, both linked in the template section of this guide, no sign-up required.
This guide draws on the World Economic Forum’s Future of Jobs Report 2025 for skills-gap context, AIHR’s definition of a general skills matrix for the section 2 comparison, and Harvard Business Review’s May 2025 piece on assessing talent needs in the age of AI for broader framing. Sprad and Rework, two public AI-specific matrix frameworks, are named for contrast in the proficiency-levels section; Future Factors’ own scale and domain structure are built independently of both. Two statistics that surfaced in AI-generated search summaries, attributed to HBR and SHRM, could not be confirmed against a live primary source and do not appear anywhere in this article.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.