Two AI enablement proposals land for what sounds like the same work. One quotes $6,800. The other quotes $34,000. Neither vendor is lying. They are quoting different jobs, and the difference hides in eleven line items most buyers skim past on the way to the total.
The procurement templates circulating for AI purchases will not help here. They were written for buying software: model transparency, SOC 2 Type II, data residency, uptime converted into downtime hours. Fine questions for a platform, useless when what you are buying is whether forty people write a better first draft on Monday. The case for treating capability as the purchase sits in our argument that licenses are not capability; the operating sequence is in the enablement playbook. This piece is narrower: how to read the quote in front of you.
What an AI enablement proposal pays for
Three things that have little to do with each other: preparation, time in the room, and what happens after everyone leaves. Responses to an AI enablement RFP arrive in the same shape, so the eleven lines below are what make two documents comparable at all.
Preparation is the vendor reading your work and rebuilding the session around it. Time in the room is facilitation, the one line almost every vendor itemizes honestly because it is easiest to count. What happens after is reinforcement, manager support, and measurement, and it separates the two quotes.
The $6,800 version is one facilitator, one day, a deck adapted the night before, a PDF of prompts. The $34,000 version buys two weeks of document review, four cohorts sized for practice rather than seating, artifacts built from your own files, a reinforcement cycle, and a measurement design you can show a CFO. Both are legitimate products. Only one is the job you described, and the proposal rarely says which.
ATD’s 2026 State of the Industry report, covering 2025, puts average direct learning spend at $846 per employee against 16.7 formal learning hours, down from $1,254 a year earlier. More hours, less money. That squeeze is why the post-session lines get cut first.
The eleven line items in an AI enablement proposal
1. Scope statement and target roles
What it buys: a named audience and a named job. “Operations and finance staff who draft recurring documents” is a scope. “All employees” is a seating chart.
What it excludes: every role left unnamed. Work scoped to individual contributors will not change how directors review output, and the gap resurfaces six weeks later as “nobody is using it.”
Ask: which three tasks look different after this, and for whom?
2. Discovery and curriculum customization
What it buys: the vendor’s prep time, and the widest honest range in the document. One vendor’s “customization” is a 30-minute intake call and a swapped logo. Another’s is reading twenty of your files and rebuilding the exercises around them.
What it excludes: anything you did not send. If the proposal never states how many work samples get reviewed, assume zero.
Ask: how many hours of our material will you read first? A structured readiness assessment can stand in for this line, but only when its output is a curriculum rather than a maturity score.
3. Facilitation hours and session format
What it buys: a facilitator in front of people for a stated number of hours. The honest unit is delivery hours, not “a workshop.”
What it excludes: prep, travel, recording, a repeat run for whoever missed it. “A one-day session” with no hour count means six hours including lunch.
Ask: how many delivery hours, across how many days, and what does one extra session cost?
4. Cohort size and the facilitator ratio
What it buys: the chance for each person to do the work instead of watching it. Hands-on fluency with ChatGPT, Claude, Gemini, or Microsoft Copilot needs every participant to type, fail, and get corrected. In our experience that stops above twenty-five people per facilitator.
What it excludes: the arithmetic. A flat rate for “up to 100 participants” sells a lecture at a practice price. Sessions we run get rebuilt into smaller cohorts more often than any other line changes: the client asked for practice and booked an auditorium.
Ask: what is your cap per facilitator, and what happens to the format above it?
5. Participant materials and practice artifacts
What it buys: what people keep. Weak version: a slide PDF and a generic prompt list. Strong version: artifacts built from your documents, so each person leaves with a working prompt for the report they file monthly.
What it excludes: reuse rights. Decks are often licensed for internal viewing only, which blocks you from running the next cohort yourself.
Ask: do we own the materials, and may we reuse them without you?
6. Work-sample handling and confidentiality
What it buys: the terms under which your documents enter the vendor’s hands. This work only lands on real work, so contracts, financials, and client files move. The line should name where samples sit, who sees them, and when they are deleted.
What it excludes: nothing you should accept. A missing data-handling line stops the purchase, less for the compliance exposure than for what it reveals: a vendor who has not thought about it has not done this work at scale.
Ask: where do our files live, and what is the deletion date?
7. Reinforcement after the session
What it buys: the cycle between a session and a habit. Office hours, a clinic two weeks out, a shared prompt library, a channel where questions get answered. Priced as a block of hours or a short retainer.
What it excludes: itself, usually. The most commonly absent line in the proposals clients send us to review, and the first dropped to win on price.
Ask: what is in place in week three, and who runs it?
8. Manager enablement
What it buys: a shorter session for the people who approve the work. Managers who cannot tell a strong AI-assisted draft from a lazy one quietly penalize both, and the capability dies in review.
Ask: what do managers get, and when, relative to their teams?
9. Measurement design
What it buys: an agreed answer to “did it work,” written before the work starts. A usable version names a baseline taken before the session, two or three task-level measures, and a date for the second reading.
What it excludes: usage dashboards. Seat activity tells you licenses are open, not that capability moved. LinkedIn’s Workplace Learning Report found only 8% of L&D professionals are highly confident they can measure learning’s business impact, so few buyers can supply the rigor a vendor omits.
Ask: what baseline do you capture, and what number appears in your closing report?
10. Named delivery team and substitution terms
What it buys: the person who shows up. Delivery quality here is the practitioner, not the brand. A proposal naming a senior facilitator in the pitch and “a certified trainer” in the scope quotes two different people.
Ask: who leads the room, what have they shipped with these tools, and what are our rights if that changes?
11. Exclusions, change orders, and rescheduling
What it buys: clarity about the edges: what sits out of scope, the rate for added work, the cancellation window in days.
What it excludes: by design. Read this line first, not last. A full exclusions list marks a vendor burned into honesty; an empty one means the surprises are still coming. What AI enablement services include and quietly leave out repeats closely enough across vendors that you can predict the gaps.
Ask: what is the rescheduling fee, and how many days of notice trigger it?
Where corporate AI enablement proposals hide the real cost
Three places, in order of how often they bite.
Your own people’s time. Forty people at six hours is 240 hours of payroll, which at mid-market loaded rates often exceeds the vendor’s whole fee. Participant hours appear in almost no quote, and a vendor who raises them unasked is telling you something.
License and seat cost. The work assumes a working seat on at least one assistant for everyone present. If half the room is on a free tier, the exercises break in the first hour.
The second cohort. Session four is invisible in a quote for sessions one through three, and the per-session rate quietly doubles once the block discount expires.
Three ways the number gets built
Three structures for the same work, and the structure decides who carries the risk.
| Structure | What you pay for | Favors you when |
|---|---|---|
| Per seat | A headcount price, like a catalog subscription. Udemy Business publishes a Team plan list price of $30 per user per month billed annually, with enterprise pricing on request | Many people, low customization needs |
| Per cohort | A fixed fee per session or group, whoever attends | A defined team that needs real practice |
| Retainer | Monthly access to facilitation, clinics, reinforcement | Capability must hold across quarters, not weeks |
Hands-on work for a mid-market team lands in the $2,000 to $15,000 band per engagement across the market, with follow-on custom automation at $25,000 to $150,000. A per-seat quote far below that buys a course catalog, a reasonable product that does not change behavior on its own.
“Training” means two unrelated things in an AI contract
One meaning is curriculum: what you are buying. The other is model training: whether the vendor’s tooling feeds your documents into someone’s training set. Both hide under headings that read “training” and “data,” and a proposal can carry a strong curriculum section with a silent model-training clause, or the reverse. Check both.
Which missing lines are fatal and which are negotiable
Not every gap is a reason to walk.
Fatal. Missing data handling (item 6), missing scope and target roles (item 1), or a facilitation line with no hour count (item 3). Each means the vendor has not run this work, or is leaving room to under-deliver.
Negotiable, and cheap to add. Reinforcement (item 7), manager sessions (item 8), substitution terms (item 10). These go missing because nobody asked. A vendor who adds them in revision without inflating the total is showing you their cost base.
Worth paying more for. Measurement design (item 9) and real customization (item 2). These decide whether anything survives to the next quarter, and they are the first two cut under price pressure.
A 30-minute method for comparing two proposals
Rebuild both quotes on one page, in the order above, blanks included.
- Write the eleven items down the left column, the two vendors across the top.
- Fill in only what each proposal states. No charitable reading.
- Divide each total by stated delivery hours for a comparable hourly rate.
- Count the blanks. The vendor with fewer usually costs more and is usually quoting the job you described.
- Send both the same three questions from your largest blanks, and compare the replies rather than the documents.
Step five does most of the work. A vendor who answers a scope question with a calendar invite is showing you how the engagement will run. Still shortlisting? Our view on which enablement firms fit mid-market teams covers that stage.
Frequently asked questions
What should an AI enablement proposal include?
Eleven lines: scope and target roles, discovery and customization, facilitation hours, cohort size, participant materials, work-sample handling, reinforcement, manager enablement, measurement design, named delivery team, exclusions with change-order rates. A document missing the first, the sixth, or an hour count on the third is not comparable. Four to eight pages is the normal length; over fifteen and you are reading a platform RFP response.
What is a fair price range for a hands-on session?
Roughly $2,000 to $15,000 for a mid-market engagement, driven by prep depth, cohort count, and whether reinforcement is included. Below that band you are buying open-enrollment seats. Above it, ask which of the eleven lines carries the extra cost.
Per seat or per cohort?
Per cohort when you want people practicing on their own work: the price holds while the format stays small enough to teach. Per seat when the goal is broad catalog access and completion rates are the metric you report.
Who owns the materials afterward?
Whatever the proposal says, and silence favors the vendor. Ask for internal reuse rights on decks, exercises, and prompt libraries built from your documents. Most grant it before signing and charge for it after.
What if the facilitator we met is not the one who shows up?
Add a substitution clause: named facilitator, your written approval for any change, a right to reschedule without penalty if the replacement is not comparable. The cheapest protective term in the document, and almost never offered unasked.
Is a paid pilot cohort worth running first?
For teams above about 100 people, run one. A single cohort at full price, with the measurement design in place, tells you more than any reference call. Ask that its per-cohort rate carry into the larger block.
What belongs in the data-handling section?
Where work samples sit, who accesses them, whether a third-party tool processes them, the retention period, and the deletion date. Two sentences will do if they answer all five. A blank here should stop the purchase.
How do I compare two proposals with completely different structures?
Normalize to delivery hours and count the blanks. Divide each total by stated facilitation hours, then list which of the eleven lines each leaves empty. Structural differences stop mattering once both sit on one grid.
Where to start
Pull the quote you have and read item 11 first. The exclusions list tells you more than the cover letter does, and the blanks point at the three questions worth sending back.
Want a second read before you sign? Book a free AI-readiness call. Thirty minutes, no pitch, and you keep the notes.
Arthur Wandzel
