Most advice on AI enablement partner selection is a list of virtues, and virtues cannot separate two bidders. Every proposal on your desk claims domain expertise, transparency, a focus on people, and a long-term mindset. All of them do. The rules below are built the other way around: each one names the thing to ask for and the answer that should end the meeting. They assume you have already bought ChatGPT, Claude, Gemini, or Microsoft Copilot seats for a company of 50 to 500 people, that usage looks fine on the dashboard, and that someone has asked what the licenses bought. That gap, and why spend alone does not close it, is the argument in the case for building capability rather than buying seats.
What You Are Buying When You Buy AI Enablement
Three unlike products are sold as AI enablement services, and buyers compare them on price per head as though they were substitutes. What you want from any of them is LLM fluency: practical skill with the large language models behind ChatGPT, Claude, Gemini, and Microsoft Copilot.
Catalog subscriptions (Coursera, Udemy Business, LinkedIn Learning, Pluralsight, Skillsoft) sell access to a library. You pay per seat per year, your people watch at their desks, and completion rates become the metric because they are the easiest thing to report. Live cohort programs (General Assembly, BrainStation, Section, Maven, NobleProg, Certstaffix, and the independent practitioners) sell instructor-led hours. Build-and-integrate consultancies sell software, with a training day near the end so the thing they shipped gets used.
Only the second category reliably changes what a person can do on Monday, and only when it is built on the team’s own work. The first is a reference shelf. The third is a software project wearing a training label. Where that line falls is the subject of our guide to what these engagements include and quietly leave out.
Hold two numbers. McKinsey’s State of AI found that only 21% of organizations using generative AI had “fundamentally redesigned” any workflow, and that redesign moved earnings more than any other practice. A LinkedIn survey reported by Fortune in August 2025 found 51% of professionals calling the intensity and frequency of AI training demands excessive, and 35% nervous about discussing AI at work at all. Your people are not short of modules. They are short of a reason to work differently, and they are already tired.
Rules 1–3: Scope and Fit
These three decide whether the engagement is aimed at your company or at a template.
1. Make Them Name the Capability, Not the Curriculum
Ask what a named role will be able to do afterward that it cannot do now, in one sentence, with a verb. A good answer sounds like: “Your proposal writers will turn a 40-page RFP into a scoped first draft in 90 minutes instead of two days, and will know which parts to never trust.”
End the conversation if you get a module list back. “Module 1: Introduction to Large Language Models. Module 2: Prompt Engineering Fundamentals.” A module list is a table of contents, not an outcome, and it is what a vendor sends when the program predates you.
2. Insist the Exercises Run on Your Documents
Fluency is a transfer problem. Few people carry a skill from a sample dataset into their own inbox, because the hard part was always the messy input they already have.
So ask a blunt question: which of our files will people work on in hour two? The answer should name a real artifact type, and their next sentence should be about redaction and handling, because that is what a partner who has done this before worries about.
End the conversation if the exercises use the vendor’s case study, a fictional company, or a public dataset. The reservoir of generic examples is the cheapest content a training business owns.
3. Check That the Curriculum Splits by Role
A finance analyst, a sales rep, and a recruiting coordinator have three different jobs, three different risk profiles, and three different reasons to distrust the tool. One curriculum for all three teaches the intersection, which is close to nothing.
End the conversation if one agenda covers the whole company and the only variation offered is “we tailor the examples.” That is a slide swap. Role separation shows up in the schedule as separate sessions, not different anecdotes.
Rules 4–6: How the Work Is Delivered
Corporate AI enablement fails in delivery more often than in design, and the mechanics below are missing from most buyers’ checklists.
4. Find Out Who Stands in the Room
The person selling is often not the person facilitating. That is fine, as long as you meet the facilitator before signing and can judge whether they have done the work they will teach.
Ask for a name, what they built with these tools, and 20 minutes with them. In the sessions we run, the question that decides the room is an unscripted one from a skeptical senior person, and a practitioner is the only person who answers it well.
End the conversation if the facilitator is unnamed, “assigned closer to the date,” or described as a certified trainer with no delivery history of their own.
5. Cap the Cohort and Make Them Say the Number
A workshop is a working session. Above roughly 20 people, hands-on stops being possible and the event becomes a lecture with a chat window, whatever the invoice calls it.
End the conversation if the answer is “we can accommodate your whole team at once.” A vendor willing to put 80 people in one webinar is optimizing its own margin, and that is the easiest way to spend a budget on nothing.
6. Ask What Happens in the Following 30 Days
Capability built on a Tuesday tends to decay within weeks unless something holds it in place. Reinforcement is not a nice extra; it is most of the result.
Press for specifics: who checks in, what the second touchpoint is, whether there is an office-hours slot for a real stuck task, and who inside your company owns the habit after the partner leaves.
End the conversation if the follow-through is a session recording and a Slack channel. Both are storage, not reinforcement, and almost nobody rewatches a four-hour recording.
Rules 7–9: Proof, Not Promises
Every bidder promises results. These three make the promise checkable.
7. Demand a Baseline Before the Session, Not After
A partner who measures only after delivery has made its own result unprovable, and that is usually the point. You need a before: how long named tasks take now, what people do with the tool today, and where they distrust it. Fifteen minutes per role is enough. What that measurement can and cannot claim is the subject of our look at what readiness assessments measure and what they miss.
End the conversation if measurement is a post-session satisfaction survey. Smile sheets measure the catering.
8. Ask for a Reference Who Is Six Months Past Delivery
Mid-engagement references are warm by design. The useful call is with a buyer whose program ended two or three quarters ago, because only they know what stuck. Ask them one question: what are people doing differently now, and what reverted?
End the conversation if every reference is inside a live engagement, or if every name is withheld for confidentiality while the vendor’s own site lists clients.
9. Ask What They Will Not Teach
A partner with a point of view has boundaries. Ours is narrow on purpose: fluency with the tools your team already pays for, applied to the work it already does. We are not the right call for a model build or a policy drafting exercise, and saying so is what honest scoping sounds like.
Boundaries also show whether a partner knows the compliance floor at all. Article 4 of the EU AI Act, applicable since 2 February 2025, requires providers and deployers to “take measures to support the development of AI literacy of their staff.” If your company has staff or customers in the EU, a partner should say plainly whether its program speaks to that duty or sits outside it.
End the conversation if the answer is a version of “whatever you need.” A vendor with no edges has no expertise, only capacity.
Rules 10–12: Price, Contract, and Exit
These are the three a buyer regrets skipping, and all of them are negotiable before signature.
10. Price the Structure, Not the Sticker
Three pricing models dominate, and each pays the vendor to behave differently. Per-seat rewards headcount, so the vendor wants the biggest room. Per-cohort rewards sessions delivered, so it wants more dates. A retainer rewards the relationship lasting, the only structure that pays for the 30-day follow-through in rule 6 and the only one that can quietly bill for months of nothing.
Market ranges for a hands-on LLM fluency program land around $2,000 to $15,000 depending on cohort count and depth, and the automation work that sometimes follows sits in a different band, roughly $25,000 to $150,000. A quote far below the floor is a webinar. Far above it, without a build attached, you are paying brand.
End the conversation if a vendor will not say which structure it is quoting or how the number changes when headcount does.
11. Separate the Enablement From the Build
Watch for training priced as a loss-leader attached to a software project scoped after you have committed. It inverts the decision: you chose a teacher and got a contractor.
End the conversation if the proposal bundles workshops with an unscoped integration phase and the two cannot be bought apart. The comparison that matters here, between buying capability and building it in-house, is laid out in our enablement playbook.
12. Buy the Exit at the Same Time as the Work
The test for capability transfer is contractual, not rhetorical. Every proposal claims to build internal capability rather than dependency. Three clauses prove it.
- Who owns the materials and the prompts your team wrote during the sessions.
- Whether an internal person can be certified to run the program later, and what that costs.
- How long the agreement runs and what notice ends it.
End the conversation if materials are licensed back to you for the term, or if an internal-facilitator path does not exist at any price. A partner whose model requires you to keep buying has told you what it sells.
Scoring Two AI Enablement Proposals in One Afternoon
Score each bidder 0, 1, or 2 on all twelve rules: 0 is a disqualifying answer, 1 is vague, 2 is specific and verifiable. Do it in writing, before the next demo.
The arithmetic matters less than one pattern. A proposal scoring well on rules 1 through 3 and badly on 4 through 6 is a good program you will not receive. One that scores well on 7 through 9 and badly on 10 through 12 is a real practitioner with a contract built to keep you. Both are fixable by asking. Neither is fixable after signature.
Frequently Asked Questions
What is an AI enablement partner?
A firm or practitioner you hire to raise what your team can do with the AI tools it already has, not to build software. The work is facilitated sessions, role-specific practice on your own documents, and reinforcement afterward. The output is measured capability, not a deployed system or a course catalog.
How is this different from buying a training subscription?
A subscription sells access and measures completion; an enablement engagement sells capability change and should measure it against a baseline. Both have a place. If you want a reference shelf for self-directed learners, buy seats. If you want proposal writers working differently next month, a library will not get you there.
What should an AI enablement program cost?
Hands-on LLM fluency programs for a mid-market company usually quote in the $2,000 to $15,000 range, set by cohort count and depth of role tailoring rather than headcount alone. Automation projects that sometimes follow sit in a different band, roughly $25,000 to $150,000. Treat any quote well under the low end as a webinar in disguise.
How do I know the trainer has used these tools themselves?
Ask what they shipped. A practitioner answers with a specific artifact and the thing that went wrong in it. A certified trainer with no delivery history answers with a methodology. You can also hand them a real document in the sales call and watch them work it live, the fastest honest signal available.
Should we run a pilot before committing to the whole company?
One cohort from one role is the right first purchase. It buys a real baseline, an honest read on the facilitator, and 30 days of reinforcement data before you commit the rest of the budget. Resist a pilot spread thin across six departments: a nice photo, no measurable change.
Who should own AI enablement internally?
Someone with authority over how the work is done, usually an operations or functional leader, not whoever has spare time. Enablement that reports to a volunteer committee usually loses to quarter-end. A good partner asks who that owner is before you ask them anything.
Is a bigger consultancy a safer choice?
Not for a company of this size. Large firms often staff mid-market engagements with their most junior people, and price in overhead you do not use. Rule 4 is the real safety check, whatever the logo on the deck.
What does our team need to prepare before the first session?
Documents and permission. Pick two real artifacts per role, decide what gets redacted, and confirm which tools people may paste into. The partner should drive that list. If they ask for nothing before day one, they plan to teach from their own examples.
Where to Start
Score the proposals you already have against the twelve rules this week, before the next call. One bidder usually collapses on rules 4 through 6, and it was visible in the proposal all along.
For the market view, our survey of enablement firms serving mid-market companies covers who does what, and the enablement playbook walks the rollout from licenses to measured capability.
When you want a second read on a specific proposal, book a free AI-readiness call. We will score your bidders against these rules, tell you which rooms are worth buying, and say so if you do not need a partner yet.
Arthur Wandzel
