The AI Marketing Maturity Model: 4 Stages from Aware to Compounding
A four-stage diagnostic for marketing teams adopting AI. What each stage actually looks like, why most teams stall, and the single move that gets you to the next one.
Every marketing leader has been asked the same question: where are we on AI?
Almost nobody has a real answer. The honest answers all sound either falsely confident (“we're ahead”) or falsely modest (“we're behind”) — both are vanity moves dressed up as humility, because neither one names the thing that actually matters.
Tool count is a vanity metric. License count is a vanity metric. AI adoption percentage is a vanity metric. What matters is where AI sits in your operating model — and that requires a diagnostic, not a count.
This piece is that diagnostic. Four stages, named honestly, with a single behavioral marker for each that tells you which one you're actually in.
A note on the framework before we go further. I wrote about 4 stages of AI enablement earlier this year on Substack. The model has held up — the four stages are still the right structure. What's changed is what the destination looks like. The Compounding stage in 2026 includes a dimension that didn't exist when I wrote the original: AI running workflows autonomously, not just being used inside them. The framework hasn't expanded; the bar at the top has risen. This is the canonical version.
The 4 Stages of AI Marketing Maturity
The stages are cumulative. You don't earn Stage 3 by buying new tools; you earn it by passing Stage 2's hidden test — turning individual usage into team capability. The same is true of Stage 4. The shift from Embedded to Compounding isn't about doing more workflows; it's about building infrastructure that makes each new workflow cheaper than the last and letting AI run the existing ones without human triggers.
The rest of this guide goes deep on each stage: what you'd actually see, what's hidden, why teams get stuck, and the single move that gets you to the next one.
What this is, and isn't
This framework is a diagnostic for honest leadership conversations. It's a way to place your team on the spectrum and figure out what's actually holding you back.
It's not a scorecard you can game. It's not a quiz with a marketing payoff at the end. It's not a sales tool for someone trying to sell you an AI transformation program.
Most “AI readiness” assessments fail because they measure inputs — tools, licenses, training hours, percent of employees who've tried ChatGPT. Those are easy to count and easy to grow. They tell you nothing about whether AI is actually changing how work gets done. A team with 800 enterprise licenses and zero shared workflows is at Stage 1 with extra spending. A team of four people running three documented AI workflows that move output metrics is at Stage 3.
The behavior is the test, not the inventory.
01Aware
Leadership has talked about AI in offsites and all-hands. Someone has been "looking into" AI tools. There may be an AI subscription paid for by the company that two people use. Individuals on the team may have their own ChatGPT or Claude subscriptions and bring them up in conversation. Slack channels mention AI articles, podcasts, what competitors are doing. Nobody has been asked to own AI as a function.
The team has talked itself into thinking the discussions count as progress. They don’t. Aware is a starting line, not a stage of motion. The most reliable way to recognize Aware is by what’s NOT there: no shared inventory of who’s using what, no measurement of any time savings, no documentation of even one AI workflow.
Nobody owns AI as a function. It’s everybody’s "side interest," which means it’s nobody’s actual job. There’s no friction forcing the team to convert talk into practice, and no slot on the org chart that would be empty if AI work didn’t happen.
Name one person whose job description explicitly includes AI — even at 10% of their role. The percentage doesn’t matter; the public ownership does. The moment someone is on the hook for AI adoption, Aware starts converting into Experimenting by structural pressure, not by inspiration.
A team that mistakes "we’ve been discussing AI for six months" for being further along than Aware. If you don’t have a shared inventory of current usage, you’re at Aware regardless of how long the conversations have lasted.
02Experimenting
Three to five people on the team use AI heavily for their own work. Their output is faster, cleaner, or more interesting than it was a year ago, and people notice. A Slack channel exists for prompt-sharing. The CMO points to specific wins ("look what Marie did with Claude last week"). The team has at least one or two paid subscriptions, sometimes more than they need.
The team’s capability hasn’t changed — only a few individuals’ capabilities have. When the power users go on vacation, AI usage on the team drops. When a new hire joins, they don’t inherit AI workflows; they have to either become a power user themselves or work the old way. The work product looks different in places, but the operating model hasn’t moved.
Knowledge stays in individuals. The power users are doing too much value-creation to stop and document. The rest of the team feels behind and either avoids AI entirely or asks the power users for one-off help. Nobody has the time or the explicit job of turning personal practice into team practice — which is the standardization work that gets you to Stage 3.
Take ONE workflow that a power user is doing successfully and make it the team’s. Document the prompt, name the workflow, assign it an owner, and have a second person run it end-to-end. The goal isn’t perfection; it’s transferability. Once one workflow is team property, the path to two and three is clear.
A team that mistakes high-volume individual usage for team capability. Marie shipping twice as much content as anyone else doesn’t mean the team has Embedded AI in its content workflow. It means Marie has. That’s a Stage 2 marker, not a Stage 3 one.
03Embedded
Specific workflows include AI steps as a documented part of the process. The team has a shared prompt library — in Notion, in a Custom GPT, in a shared Claude Projects setup. New hires are onboarded onto these workflows. There’s measurement: time-to-publish has dropped, content output has grown, reporting cadence has tightened. AI usage isn’t dependent on which individual is in the room. The team can describe its AI workflows the way it describes any other operational process.
A human still triggers every workflow. The AI is the accelerant, but the system depends on someone deciding to run it. There are no agents running on a schedule, no automations that fire without human input, no monitoring loops that catch exceptions while everyone is asleep. The team has integrated AI into the way work flows; it hasn’t yet built infrastructure that RUNS without continuous human attention.
Embedded is a comfortable plateau. The wins are real, the case is defensible, the next stage requires a structurally different kind of work: infrastructure, ownership, and an investment that pays off slowly. Most teams stop here because the marginal next workflow is faster to build than the first agent.
Build ONE always-on workflow — an agent that monitors a channel, summarizes inputs, runs research, or flags exceptions without being triggered by a person. The first one is the hardest. The second is half as hard. The fifth is where the infrastructure starts compounding — which is the destination.
A team that confuses "we use AI in our workflows" with "AI runs our workflows." Both can be true at the same time; only one is Stage 4. The test is whether anything happens overnight without a person making a decision.
04Compounding
Agents running 24/7 — research, monitoring, reporting, exception-flagging. Each workflow has a named owner accountable for what it produces and how it improves. The team’s institutional context — brand voice, customer data, past campaign performance, decisioning history — lives in systems that every new agent inherits, not in human heads that have to onboard new people from scratch. When a new AI capability or model becomes available, the team has a playbook for evaluating and integrating it within weeks, not quarters.
Compounding has two halves. The first is what you can see — autonomous workflows, named owners, infrastructure. The second is more subtle: each new workflow built on this stack is cheaper than the last, because the context, tooling, and integration patterns are already there. The team’s meta-capability — its ability to absorb new AI capabilities — is what compounds. Teams that have the autonomous half but not the meta-capability half are at the edge of Stage 4 but not solidly in it.
Stage 4 isn’t a stable equilibrium. It requires continuous reinvestment: in infrastructure that ages, in playbooks that need updating, in leadership behavior that keeps modeling the work. Teams that reach Stage 4 and then assume they’re done tend to drift back toward Stage 3 over twelve to eighteen months as their infrastructure decays and their playbook becomes outdated.
Maintain the meta-capability. Assume the current playbook will be obsolete in twelve months. Hire and structure for ABSORPTION of new tools and models, not just optimization of current ones. The compounding only continues if the work to compound it continues.
A team that confuses "we have agents" with "we have a compounding capability." Agents are necessary but not sufficient. If your team can run autonomous workflows but takes six months to integrate a new model or workflow type, you’ve built the first half of Stage 4 but not the second.
The Distance Between 2 and 3
Of the three gaps in this framework, the one between Experimenting and Embedded is the largest, the hardest, and the one most teams never cross. It's also the only one nobody wants to do.
The other transitions — Aware to Experimenting, Embedded to Compounding — are exciting. They involve new tools, new agents, new capabilities. They feel like progress. The transition from Experimenting to Embedded is the opposite. It's slow, unglamorous, and most of the work is making things that already work transferable. Documentation. Standardization. Ownership. Training. Workflow inventory. The boring middle work.
This is the gap. Teams that take it seriously reach Stage 3 in three to six months. Teams that don't can sit at Stage 2 indefinitely — for years, with a growing pile of AI tools and no team-level capability to show for it.
Why teams stall in the gap
The power users don’t want to do it. They’re shipping faster than anyone else; documentation slows them down. Asking your best AI users to stop and write SOPs feels like punishing performance.
Leadership can’t measure it. The work doesn’t produce visible output for weeks. It’s hard to defend in a status update. "We documented two workflows this quarter" doesn’t read like progress next to "we shipped 30% more campaigns."
The team thinks they’re past it. When 30% of your people use AI heavily, it can feel like adoption. What it actually is, is visible variance. Adoption is a team property, not an individual property.
What it actually takes to cross
You need someone whose job is specifically to turn individual practice into team practice. Call them an AI workflow architect, an AI ops lead, or just “the person responsible for the team's AI capability.” The title doesn't matter; the explicit ownership does. Without that role on the org chart, the standardization work gets queued behind every other request and never happens.
You also need to accept the tradeoff. Documenting a workflow makes the workflow itself slightly slower. The power user who could ship three blog drafts a day now ships two and a half — because half a draft's worth of time goes into the SOP. That's the price of moving from Stage 2 to Stage 3. Teams that aren't willing to pay it stay at Stage 2.
The fastest way through
Pick one workflow that's already working at the individual level. Make that one workflow team property — fully documented, owned, transferred, measured. Don't try to standardize all of AI at once; that's how teams paralyze themselves. One workflow, in one quarter, with a named owner. That's the entrance to Stage 3.
Hiring, budget, and leadership per stage
The headcount, budget allocation, and ownership pattern that works for your team depends entirely on which stage you're in. Here's what I've seen actually work.
The pattern across all four stages: the further along you are, the less you need an AI-specialist hire and the more you need infrastructure ownership. Teams that try to hire their way to Stage 4 by importing AI specialists tend to stall earlier than teams that grow it from within.
The honest diagnostic
Naming your stage in your head is easy. Naming it correctly is harder, because every team's instinct is to round up.
The shortest path to an honest read is the AI Readiness Scorecardon this site — a three-minute, eleven-question diagnostic that scores your team across five dimensions and places you on the maturity model. It's deliberately built to make rounding up difficult: the questions are falsifying, and the result is structured by dimension so you can see where you're actually strong versus where you've been flattering yourself.
Most leadership teams that take it together get the same surprise: they're a stage earlier than they thought, and the reason they're stuck is one dimension they hadn't been paying attention to.
Place your team on the model → Take the AI Readiness Scorecard — three minutes, eleven questions, instant results.
Where to start, by stage
You don't move to the next stage by trying to be at the next stage. You move by doing one specific thing well. Here's the single move per stage:
Name one person whose job description explicitly includes AI, even at 10% of their role. Public ownership is the catalyst.
Take one workflow that a power user is already doing successfully, document it, name an owner, and have a second person run it end-to-end. Transferability over perfection.
Build one always-on workflow — an agent that runs without human triggers. The first one is the hardest; the system compounds from there.
Treat the meta-capability as a recurring investment. Assume your current playbook will be obsolete in twelve months. Hire and structure for absorption, not just optimization.
If you can name your team's next move from this list, you have a real plan. If you can't, the work is figuring out which stage you're actually in — which is what the scorecard above is for.
Where this fits
This piece is the diagnostic — where am I? Two companion pieces describe the rest of the journey:
How to measure and defend the value as you move between stages.
What the destination looks like at full Compounding — structurally.
Together they answer the three questions every leader running an AI adoption program actually has:
Where am I? (this piece)
How do I prove progress? (ROI Playbook)
Where am I going? (AI-First Org)
Frequently asked questions
What's the difference between Embedded and Compounding?
Embedded uses AI inside workflows that humans trigger. Compounding has AI running workflows autonomously AND has the meta-capability to absorb new AI tools quickly. The hidden difference: Compounding has institutional context living in systems, not in individual heads. If a new hire onboards by reading documentation and inheriting agents, you’re at Compounding; if onboarding requires shadowing power users, you’re at Embedded.
How do I know which stage my team is in?
Take the AI Readiness Scorecard at /tools/scorecard. The instinct to round up is universal — every leadership team thinks they’re a stage further along than they are. The scorecard’s diagnostic questions are designed to make rounding up difficult.
How long does it take to move between stages?
Aware to Experimenting: weeks, if you name the right person. Experimenting to Embedded: three to six months of standardization work — this is the gap most teams stall in. Embedded to Compounding: twelve to eighteen months, because it requires building infrastructure, not just running more workflows.
Can a team skip a stage?
Rarely, and usually not for long. Teams that try to skip Stage 3 by going from "we have a few power users" to "we have agents" tend to build fragile systems that can’t be maintained. The standardization work at Stage 3 isn’t optional — it’s what makes Stage 4 possible.
What's the biggest mistake teams make assessing their stage?
Rounding up. The second biggest: confusing individual usage with team capability. A team where 40% of people use AI heavily but no workflows are documented is at Stage 2, not Stage 3. Adoption is a team property.
How does this relate to your original 4 Stages of AI Enablement post?
The four stages are the same. What’s changed is what Compounding looks like in 2026: the autonomous-operations dimension didn’t exist when I wrote the original Substack version. This is the canonical, expanded definition of the framework.
Do small teams move through stages faster than big ones?
Sometimes. Small teams have less bureaucracy and faster decisions, but they also have less capacity to do the standardization work at Stage 2–3. The teams I’ve seen move fastest are 8–30 people: big enough to have a workflow architect, small enough to standardize without committee.
Where to find me
The framework above is what I use when I'm helping marketing teams figure out their AI adoption stage and what to do next. If you're working through any of it and want a second set of eyes, the easiest place to find me is LinkedIn ↗.
Last updated: May 2026.
