A strategy consultancy sells a decision. An AI practitioner sells a changed Tuesday. If your people already hold licenses and are not using them well, you have a behavior problem rather than a strategy problem, and the second purchase is the one that moves it. The reverse holds too: if nobody has agreed what AI is for at your company, no amount of hands-on AI enablement will settle it, and a practitioner will take your money anyway.
One disclosure first. SF AI Labs runs applied AI enablement workshops, which makes us one of the two options on this page. So every test below is built to survive striking our name out. Each is answerable from a vendor’s own public documents, and the structural claim rests on published data on how consulting firms staff themselves.
What each one is selling
Search this question and you will mostly find pages comparing vendors who build AI systems. Not your question. You are buying whether 40 people can get usable work out of ChatGPT, Claude, Gemini, or Microsoft Copilot by the end of next quarter.
A strategy consultancy sells alignment and a defensible decision. The deliverable is a diagnosis: where AI fits your operating model, which processes to touch first, what the governance posture should be, what the board should be told. Training appears inside it as an executive briefing or awareness module.
An AI practitioner — a specialist enablement firm, a boutique, an independent facilitator — sells applied skill in named hands. The deliverable is that your finance team can get a usable first draft of a board memo out of an assistant, on your template, with your numbers. The scope is narrow on purpose.
General Assembly, itself a training provider and open about it, frames the choice as a test of what is missing: infrastructure, or internal capability. Worth sharpening.
The question is what has to be different, and by when. Infrastructure gaps take quarters. Behavior gaps take a facilitated session plus about six weeks of follow-through.
Which capabilities to rent and which to own is worked through asset by asset in our playbook on buying versus building AI enablement capability. This page covers which kind of outsider to put in the room once you have chosen to rent.
Who is in the room on day one
Here is the part no comparison page states, and it is predictable from public data rather than gossip.
Large professional-services firms run on staffing ratios: a senior person sells and supervises, a larger group of junior people delivers. Inside Public Accounting’s March 2026 analysis of accounting-firm economics put the average staff-to-equity-partner ratio at 11.8 to 1, rising to roughly 16 to 1 at firms above $150 million in revenue. Advisory practices are not audit practices, but they are staffed on the same principle. That is the business model, not a flaw, and it funds the bench, the methodology, and the risk coverage no boutique can match.
Now run the arithmetic on a two-day training room. A partner’s hours are the firm’s scarcest and highest-billing asset. Sixteen of them spent teaching 20 employees to prompt an assistant against their own documents is the worst use the model permits, so it does not permit it.
The partner scopes, frames, and appears for the executive session. A consultant two to four years out of school runs the hands-on work.
That person may be excellent, and plenty are. But they are teaching applied judgment about tools they have used in a consulting context rather than in yours, with a methodology to follow in place of a practice to draw on.
At a specialist firm the economics invert. There is no pyramid to feed, so whoever sold the session usually delivers it. The honest trade: less institutional depth behind the engagement, more of the seller’s own hands in the room.
Either answer can be right. Not knowing which you bought is the unacceptable outcome.
Ask for delivery-team names and in-room hours in writing before signing. Both kinds of vendor can answer in a sentence, and the quality of that answer is diagnostic.
The big firms built their academies for themselves first
The large firms have built serious AI fluency machinery over the past two years. Who it was built for is worth stating precisely, using their own published numbers.
Deloitte’s AI Academy page describes 30-plus hands-on courses and reports that 100% of its US professionals completed at least one AI training in the past year, across roughly 300,000 completed trainings. The page addresses Deloitte’s own people, with an invitation to clients further down. Accenture committed $1 billion over three years to its LearnVantage platform and bought Udacity to stock it. In December 2025 it announced with Anthropic that about 30,000 Accenture professionals would be trained on Claude, one of the largest groups of Claude practitioners anywhere. CNBC reported in February 2026 that the firm had told senior staff regular use of its AI tools would count toward leadership promotion.
Read that for what it is. The most committed corporate AI enablement programs in professional services are internal. Firms whose product is billable human judgment met an existential version of your problem and spent accordingly.
The client-facing academy is that curriculum, productized and resold. Material tested across 300,000 completions is a real asset no boutique can match. It was also built for consultants doing consulting work, so adapting it to your work is a scoping exercise you pay for rather than a property of the product.
Neither firm publishes a training price; every inquiry routes to a conversation. That published-versus-quoted divide runs through the whole market, mapped in our comparison of workshop providers for mid-market companies.
The shape of the price tells you more than the number
Market ranges first. A private applied workshop sits roughly between $2,000 and $15,000 depending on cohort size, days, and preparation against your materials. A custom automation build is a different purchase, roughly $25,000 to $150,000. Strategy engagements begin where the workshop range ends and rise without a ceiling. Confirm any figure against a current quote; this market reprices quarterly.
The number matters less than the structure. Three price shapes, each revealing what the seller thinks it sells.
| Price shape | What is being sold | What it means for you |
|---|---|---|
| Day rate or fixed session fee | Delivery time with a defined end | Exposure capped by scope; you own follow-through |
| Monthly retainer | Continuing access and advice | Renews without anyone re-asking whether it worked |
| Program fee with phases | A change effort, training as one workstream | Priced together, so not accountable separately |
The third shape is where mid-market buyers lose money, rarely through anyone’s bad faith. Training inside a larger program has no line of its own, so no acceptance criterion attaches to it. The program reports green on its strategy milestones and nobody asks whether anyone’s Tuesday changed. A fixed session fee is small, visible, and easy to judge. What makes it feel smaller is what makes it accountable.
What happens when the engagement ends
Both options would rather you ask this late, and it separates them more cleanly than anything else.
A consultancy’s output is a document and a set of decisions. Its durability depends on whether your organization can execute without the authors present.
Forrester’s Ted Schadler wrote in July 2026 about the firm’s evaluation of ten major AI consulting providers. One lesson: center design thinking, user journeys, and behavior change, because unused AI tools have no value. That is an analyst telling enterprise buyers the strategy artifact is not the point. It bites harder at 200 employees than at 20,000, with no transformation office to carry it.
A practitioner’s output is skill in specific people, which decays differently. It does not sit on a shelf; it fades, and it resigns.
The counter is cheap and almost always skipped. Name an internal owner. Keep a short library of the prompts the team built during the session on their own documents. Set one follow-up date before anyone leaves the room.
Name the asymmetry: a strategy document nobody acts on leaves nothing behind, while a workshop nobody reinforces still leaves a dozen people who can do something they could not before. Know which failure you can afford.
Whatever a vendor claims about durability, check it by talking to a buyer two quarters past delivery. How to do that without being managed is in our guide to checking an AI training vendor’s case study.
When the strategy firm is the right call
Four situations where the consultancy is the correct purchase and a practitioner would be malpractice.
Nobody has decided what AI is for. If two executives would answer “what are we trying to achieve here” differently, training 40 people is premature. You will teach them to use tools well in service of nothing agreed.
The constraint is regulatory or contractual. If client agreements, a security review, or sector rules govern what your people may put into a model, settle that before fluency helps. Specialist firms rarely carry the liability posture for this work, and the good ones say so.
The board needs an institution’s name on the recommendation. Sometimes the deliverable is credibility, and a specialist cannot supply it.
The scope is an operating-model redesign. If roles, handoffs, and headcount are genuinely changing, you are buying change management, with training correctly placed inside it. BCG’s Applied AI Index, published September 30 2026 from a survey of 1,330 senior leaders, found nearly half of companies now generating real value from AI and seven in ten of that group already retraining staff. Reskilling is part of that redesign, not a replacement for it.
When the practitioner is the right call
Four situations where the specialist wins, and they describe most 50-to-500-person companies.
Licenses are bought and underused. The common case. Someone approved seats, adoption dashboards read thin, and the ask is a return on money already gone. A facilitated session against real work closes more of that gap per dollar than anything else available. Why license spend is not capability is argued in our case that buying licenses does not build capability.
The timeline is this quarter. A strategy engagement concluding in March does not change February. A workshop happens on a date.
The budget is five figures, not six. Below roughly $25,000 you are not a viable strategy client. You will be served by the most junior available team against a methodology, the worst of both options.
The work is specific and documentable. If the team’s day is proposals, client emails, research summaries, and reporting, those are the artifacts a hands-on session is built around. A catalog course cannot be, which is the real line between catalog and custom delivery, drawn in our comparison of catalog courses against custom-built training.
Putting part of the fee at risk on a small purchase
Forrester’s sharpest advice to enterprise AI buyers is to put part of the provider’s fee at risk through results-based pricing. Written for seven-figure engagements, it scales down intact, and almost nobody applies it to a $9,000 purchase order.
The small-purchase version: hold back 15 to 25% of the fee against one observable outcome measured four to six weeks after delivery, defined before signing in terms nobody can argue about later.
- A named artifact produced with assistant help and sent to a real recipient, from at least 70% of participants
- Three workflows the team now runs differently, written up by the team rather than the vendor
- A named internal owner who can run the next session unaided
Then watch the reaction, the cheapest due diligence available. A practitioner who works this way negotiates the threshold and signs. A vendor selling a training module inside a larger program will explain that its deliverables are not measured that way, which is true, and tells you the training is not separately accountable.
Add one term while drafting: you own the prompts, patterns, and materials the team produces in the session, with no limit on internal reuse. Vendors who decline are selling continued dependence.
For most companies at this size the sequence beats the either-or. Buy strategy once, narrowly, in weeks rather than quarters. Buy applied enablement repeatedly and close to the work, per team, with a measurement step after each. Avoid the bundled program where the strategy work is good, the training is the module attached to it, and neither is priced alone.
Frequently asked questions
What is the difference between an AI strategy consultancy and an AI practitioner?
A consultancy sells a diagnosis and a decision: where AI fits, what to do first, how to govern it. A practitioner sells applied skill in named people, through hands-on sessions against your own work. One changes what leadership has agreed; the other changes what employees do Monday.
Do McKinsey, Deloitte, Accenture, PwC, or EY train client employees on AI tools?
Several do. Deloitte’s AI Academy and Accenture’s LearnVantage are both offered to client organizations, and both were built for the firms’ own workforces first. Deloitte reports roughly 300,000 AI trainings for its US professionals in a year; Accenture committed $1 billion over three years to its platform. Neither publishes a client price.
How much should a 100-person company budget for AI enablement?
Private applied workshops run roughly $2,000 to $15,000 per engagement at market rates, depending on cohort size and preparation. For 100 people you are buying several cohorts, so model it per team. A six-figure quote is a change effort with training inside it: possibly right, definitely a different purchase.
Who delivers the session when a large firm sells training?
Usually a consultant a few years into their career, following the firm’s methodology. Published ratios in public accounting average around 11.8 staff per equity partner, reaching about 16 to 1 at larger firms, and advisory practices are staffed on the same principle. Senior hours in a two-day training room are not something that model sustains. Ask for names and in-room hours in writing.
Can we hire a consultancy for strategy and a practitioner for delivery?
That is the better structure for most mid-market companies. Buy a short, narrow strategy engagement once to settle purpose, policy, and ownership, then buy applied enablement repeatedly, per team. Avoid a single bundled program where training is an unpriced workstream.
What should be in the contract so the capability stays after the vendor leaves?
Three terms. You own all prompts, patterns, and materials the team produces, with unrestricted internal reuse. A named internal owner is identified before delivery. A follow-up checkpoint has a date on it before the session ends. Each costs the vendor nothing and roughly doubles what survives.
How do we tell whether the training worked?
Measure artifacts, not satisfaction. Four to six weeks out, count how many participants shipped real work produced with assistant help, and how many workflows the team now runs differently. Smile sheets measure whether the day was pleasant, which is unrelated to whether anything moved.
Does a specialist firm have the governance credibility a board will accept?
Often not, and that is the honest answer. If the deliverable is an institution’s name on a recommendation, buy the institution. If it is 40 people working differently by March, the institution is an expensive route to it.
What single question separates the two options fastest?
“What will be observably different, for whom, and on what date?” A practitioner answers with people and a calendar. A strategy firm answers with decisions and a document. Both are legitimate. Only one is what you need right now.
Key takeaways
- A consultancy sells a decision; a practitioner sells changed behavior. Underused licenses are a behavior problem, the common situation at 50 to 500 employees.
- Professional-services staffing ratios, averaging 11.8 staff per equity partner in public accounting, predict that senior people will not teach your two-day session. Ask for names and in-room hours in writing.
- The large firms’ AI academies are strong and were built internally first: Deloitte reports about 300,000 AI trainings for its own US professionals, Accenture $1 billion into its own platform.
- Price shape is diagnostic. Training bundled inside a program fee carries no acceptance criterion.
- Hold back 15 to 25% of a workshop fee against one observable outcome measured four to six weeks later, then read the vendor’s reaction.
If you hold licenses already and want a straight read on whether a facilitated session or a strategy engagement is the right next purchase, book a free AI-readiness call. We will say so when it is the latter.
Arthur Wandzel
