A repeatable leadership cycle for building, assessing, and sustaining AI fluency across your data team.
Most leaders try to develop their team's AI capability with one tool: a self-assessment survey or a generic training course. Both fail for the same reason.
They never make contact with the team's actual work. The COACH model is a repeatable leadership cycle that moves from definition, to organizational context, to self-reported capability, to observed reality, to targeted coaching. It deliberately separates what people say they can do from what they demonstrably do — and the gap between those two is where the coaching value lives.
COACH is a loop, not a one-time audit. Roles change, tools change, and people grow. The most fluent leaders re-run it on a cadence — quarterly to biannually — treating AI fluency as an ongoing leadership capability rather than a project with an end date.
What you (the leader) do, in sequence.
What you look for — Understand, Work with, Evaluate, Manage risk.
How you grade what you see — Resistor → Enthusiast → Operator → Multiplier.
Establish a shared, role-specific definition of AI fluency before you assess anyone against it.
You cannot coach toward a target you haven't defined.
Calibration is where you internalize the AI Fluency Framework and translate it into what good looks like for each role and level on your team.
Grasping what AI is, where it helps, and deciding what to delegate.
Directing it clearly — the right tool, the right prompt, the right level of automation.
Judging the output critically — calibrated trust, not passive acceptance.
Owning the outcome — bias, ethics, safety, and governance.
A role-by-component picture of "good" that becomes your scoring rubric for the rest of the cycle.
Separate capability gaps (the person) from enablement gaps (the organization) before you coach.
It's unfair — and ineffective — to coach an individual for low fluency when the organization hasn't enabled them.
Orient yourself to the environment your team is actually operating in across three dimensions.
Which AI tools and models are approved and provisioned? Do people have the paid tiers that actually matter, or are they quietly using personal accounts — Shadow AI?
What frameworks or policies govern how AI should be used? Is there guidance on confidentiality, disclosure, and acceptable use — or a vacuum people are filling on their own?
What learning resources, budget, and protected time does the team have to build fluency? Is there a path, or just an expectation?
A clear read on your org's maturity and the enablement gaps you must either fix or work around before individual coaching can land.
Get a fast, cheap baseline of each person's capability, confidence, and exposure — and surface self-awareness gaps.
Treat self-report as a hypothesis, not a verdict.
Have each direct report complete a short self-assessment. It tells you not just where people are but how accurately they see themselves — the overconfident Enthusiast and the underconfident Operator are both coaching signals.
A self-reported baseline per person, plus an early read on self-awareness — both to be validated in Check.
Replace self-report with evidence by watching people work with AI on tasks that actually matter.
The stage almost everyone skips — and the most important one.
Routine work is the only thing that forces a genuine re-architecture of process and judgment, and it's where fluency — or its absence — becomes visible.
Select a workflow the person owns and runs regularly — chosen by you, tied to their specific role. This is where Calibrate pays off: you know what good looks like for that role.
Break the workflow down. You don't monitor everything — that's noise. Look for the natural seams where AI would or should touch the work.
Evaluate each against the AI Fluency Framework: Does the person know how much of this task to delegate vs. keep human-led? Are they briefing the tool well? Are they verifying the output? Are they handling the risk?
Provide support and explicitly carve out time and capacity. Forcing this on top of a full load guarantees a bad signal — and breeds resentment.
An evidence-based read of where each person genuinely sits, per component and on the fluency spectrum.
Close the specific gaps you've now evidenced — and protect capacity so growth is sustainable.
With three data sources in hand, you can finally coach with precision.
Use the AI Fluency Framework to name the gap exactly — this isn't "get better at AI." It's: "You're strong at directing the tool but you're taking its output at face value. Let's build your evaluation habit."
Click a position to see the recommended coaching approach.
Rejects or avoids AI — skepticism, fear, or a bad early experience.
High usage, low judgment. Fast and confidently wrong. The dangerous middle.
Uses AI well within their lane. Solid, reliable, ready to grow.
Elevates the team's AI use. Sets standards, mentors others.
Address the why first — skepticism, fear, or a bad early experience are all solvable, but only if you surface them. Then start with one high-trust, low-risk use case: something low-stakes where the worst outcome is a mildly unhelpful output. Build confidence from a single win before expanding scope.
The priority is Evaluate and Manage Risk. This is where most hidden risk lives on a team. The Enthusiast moves fast and ships AI outputs without checking them — they're not malicious, they just don't know they should. Build a verification habit: slow them down at the output review step before you worry about anything else.
Stretch them toward broader use cases and workflow automation. The Operator is reliable within their current scope — the coaching goal is to push the edges of that scope. Introduce adjacent use cases, encourage them to experiment, and give them a safe environment to fail fast before making AI-assisted changes production-ready.
Give them scope to set standards and mentor others — and protect them from becoming a bottleneck. The risk with Multipliers is that they become the team's AI resource, fielding questions and reviewing everyone's work. That's not scaling; that's a single point of failure. Help them build systems that teach the team to fish rather than feeding everyone themselves.
Build AI fluency into regular 1:1s rather than an annual review. Fluency is a living capability — it deserves living feedback.
Fluency stalls on friction you can remove — blocked access, missing tooling, unclear policies. Surface these in check-ins and act on them quickly.
Faster output isn't free. Monitor for the cognitive load of constant context-switching and the emerging risk of "AI burnout" from task overload. Protecting your team's capacity is a fluency-leadership behavior — it sits inside the model, not outside it.
Rising, role-appropriate fluency across the team — and a leader who can see, develop, and sustain it.
Five stages. Each one builds on the last. The loop makes AI fluency a leadership habit, not a one-time initiative.
| Stage | The leader's job | Key tool / action | Output |
|---|---|---|---|
|
C
Calibrate
|
Define what fluent looks like, by role & level | The AI Fluency Framework | A scoring rubric |
|
O
Orient
|
Audit org access, governance & enablement | Cisco AI Readiness Assessment | Map of enablement gaps |
|
A
Ask
|
Capture self-reported capability & confidence | Self-assessment questionnaire | Per-person baseline (hypothesis) |
|
C
Check
|
Observe real work; validate the baseline | Rebuild a real workflow; evaluate 3 core tasks | Evidence-based read + spectrum placement |
|
H
Hone
|
Coach the specific gap; sustain capacity | Four-component lens + spectrum-matched coaching | Rising fluency; healthy team |
New tools ship, people move between roles, and the bar for "fluent" keeps rising. Run the COACH cycle on a cadence — quarterly to biannually — and treat AI fluency as an ongoing leadership capability, not a box you check.