Generic AI training gets 23% sustained adoption. Role-specific training gets 67%. That's industry research and it is the single most under-circulated stat in corporate AI right now. The number basically tells you whether to write the check.
The interesting question is what counts as "role-specific" and what is actually generic dressed up in a logo swap. The line is not where most teams think it is, and the difference shows up in the budget review six months later when somebody asks why the team that got training looks identical to the team that didn't.
Here's the anatomy of why most AI training fails, and what the buyers who get it right are doing instead.
The failure mode looks the same every time
A team buys Claude Teams or Enterprise. Someone — usually a VP of Ops, a COO, or a department head who championed the purchase — gets handed the job of rolling it out. They bring in a trainer. The trainer runs a half-day session. The team leaves enthusiastic. Two weeks later, two or three people are using Claude heavily and the rest of the team is back to the way they worked before.
The trainer was not bad. The team was not lazy. The training itself was the failure point, and almost always for the same four reasons:
- No intake. The trainer never asked about the team's tools, workflows, or current AI maturity. They showed up and delivered the same deck they deliver everywhere.
- No workflow context. The examples were generic — "imagine you're writing a marketing email" — instead of "let's build the actual board pack template you build every Tuesday."
- No Skills built around real work. The training taught prompting. It didn't install anything. Nothing got deployed to the team's AI Portal that survived the session.
- No follow-up measurement. No baseline before, no re-measure after. The training existed in a vacuum and there was no mechanism to detect the fade.
Any one of those is enough to cap the engagement at 23%. All four together is the default. And the default is what most "AI training" still is in 2026.
What "role-specific" actually means
There is a lot of training in the market labeled "customized" or "role-specific" that fails the test in practice. Customization means three concrete things:
An intake before anyone shows up. A real conversation with the buyer (and ideally one or two department heads) about the team's actual tools — Salesforce or HubSpot, Slack or Teams, Notion or Confluence, what the daily and weekly workflows actually look like, where the time sinks live, what the team has already tried with AI and given up on. The intake is outcome-based, not time-bound. The output is a custom curriculum, not the same deck with a different cover slide.
Workflow examples pulled from the team's real work. If the team writes IC memos, the training builds an IC memo Skill in the room. If the team prepares board packs, the training works through a board pack template. The examples are recognizable to the people in the room because they are pulled from the room.
Something gets built and deployed during the engagement. Not as homework. Not as "we'll get to it later." During. A Project loaded with the company's real context. A Skill the team can invoke from any conversation. A validator that runs over output and catches anything unsourced. Without an artifact deployed to the team's AI Portal, the training is a memory by Wednesday.
The 67% number is what you get when those three things happen. The 23% number is what you get when one or two of them are present and the rest are decorative.
The three tiers and what each one is for
If you've looked at the claudetraining.com offer ladder, there's a logic to the tiers that maps to this question. Each tier solves for a different version of the role-specific problem.
Foundations Workshop — $2,500 (90 min, up to 25 people)
This is generic Claude and AI fundamentals taught well. No intake. No customization. The team gets a solid baseline — what Claude is, how prompting works, how Projects and Skills function. It is honest about what it is: the floor-raiser, not the operating layer.
The Foundations Workshop is what you buy when the team's starting point is so wide-ranging that customization wouldn't land — the senior analyst is at maturity level 4 and the junior PM has never opened Claude. In that case, getting everyone to level 2 is the right play and a generic curriculum is the right tool. It is not what you buy when the team is past basics and the rollout needs to produce sustained behavior change. Foundations Workshop details.
Custom Workshop — $4,500 (intake + 90 min, up to 25 people)
This is where role-specificity kicks in. The intake conversation produces a curriculum built around the team's actual tools and workflows. The examples in the room are pulled from the team's work. There is no blueprint deliverable and no ongoing support, but the workshop itself is engineered around the buyer's reality.
The Custom Workshop is "more than basics, less than a full engagement." It is the right call for a team that wants the role-specific bump without committing to a multi-month rollout. Custom Workshop details.
The 60-Day Claude Rollout™ — from $22,500 (60-day productized engagement)
This is the full operating-layer install. Foundation training across every department, role-specific tracks, three workflow builds delivered during the engagement, a written data security protocol, an admin and governance layer, and a baseline-to-delta dashboard with platform telemetry, enablement forms, and ROI math.
The Rollout is what you buy when the rollout needs to produce a measurable adoption story by a named board, LP, or leadership date. Section sells you a cohort. McKinsey sells you a deck. The Rollout installs the operating layer with numbers your board can read. The 60-Day Claude Rollout™.
Why the gap between tiers is bigger than the price suggests
A common reaction to seeing the ladder is "why is the gap between $4,500 and $22,500 so big?" The answer is that everything past the Custom Workshop is no longer training. It is rollout infrastructure that happens to include training.
The Custom Workshop teaches role-specific Claude well, in 90 minutes, with no deployed artifact and no measurement loop. That is the right product for a lot of teams.
The Rollout is foundation training across every department, role-specific tracks, three Claude-native workflow builds deployed into the team's AI Portal, an admin and governance layer, a written data security protocol, and a measurement instrument that produces a board-ready dashboard at Day 60. Two of the three named workflows alone usually save more annual hours than the entire engagement costs.
The gap is not in the training. The gap is in everything that wraps the training to make it stick. That wrapping is the 23%-to-67% delta.
What this means for the budget conversation
The default budget pattern for AI training is to set it at "low" because the alternative — "actually significant" — feels like overspending. The data argues the opposite. The competitor benchmarks for a 25-person team:
- Section School's generic AI cohort: about $50,000, no customization.
- General Assembly's custom corporate workshop: $25,000 to $50,000, light customization.
- Reforge AI cohorts: $37,500 to $50,000, self-paced plus live.
- McKinsey, BCG, Deloitte AI training: $75,000 to $150,000-plus, custom enterprise.
The 60-Day Claude Rollout at $22,500 is 5% to 30% of the equivalent custom-consulting engagement. The Custom Workshop at $4,500 is roughly $180 a seat for a 25-person team — Section's generic cohort is $1,995 a seat. The pricing math is not the problem. The problem is that "AI training" became a category buyers anchor low on because most of the market is selling 23% engagements at $5,000.
The reframe is: 93% of enterprise tech spend goes to tools, 7% goes to training. The Deloitte 2026 number. If your team has 25 Claude Teams seats at $30 a month, that's $9,000 a year in tool spend. The research benchmark says you should be spending $18,000 to $27,000 on training to realize the value. Most companies spend close to zero on the training and wonder why the tool isn't sticking.
How to tell good training from theater
If you're evaluating trainers and want to filter out the 23% engagements, ask four questions:
- What does the intake look like? If the answer is "we'll send you a form to fill out" instead of "we'll run a 30 to 60 minute conversation with you and your department heads," it's not custom.
- What gets built during the engagement? If the answer is "we'll teach the team how to build their own," nothing is going to get built. Builds happen in the room or they don't happen at all.
- What does the measurement look like at end? If the answer is "we'll send a satisfaction survey," it's not measurement. The right answer is platform telemetry, an enablement baseline-to-delta, and ROI math.
- What's the deliverable you walk away with? If it's a recording and a PDF, you bought a workshop. If it's a recording, a PDF, three deployed Skills, a written data security protocol, and a dashboard, you bought a rollout.
Those four questions filter the category faster than any sales call.
Where to start
If your team needs more than training — the full installation with three workflow builds and a measurement dashboard: The 60-Day Claude Rollout™. If you're not ready for that yet, start with a Custom Workshop — the intake conversation alone is usually worth the $4,500 because it surfaces which workflows are highest-leverage to build later.
Related reading: The Three Things Every AI Rollout Misses, and How to Measure AI Adoption Without Making It Up for the methodology.
Nicole Patten is the founder of Elevate Online and runs a Claude-specific training practice. She spent 7 years at Google as a Senior UX Engineer before dedicating her career to helping teams use AI responsibly and effectively. 100% of her business runs on Claude.