Somewhere in your contracts folder is a date. On that date a per-seat AI subscription renews for another year, usually without anyone deciding anything, because the term is annual and auto-renewing by default. That date is the most valuable thing an AI enablement business case has going for it. The spend is already approved, the invoice is already in the plan, and for about six weeks beforehand the person who signs it is willing to reopen the question. Bring a training request in a random month and you are asking for a new budget line. Bring the same request before the renewal and you are reallocating one that already exists. The document is nearly identical. The outcome is not.
Why the renewal date decides whether an AI enablement case lands
Most writing about an AI enablement business case treats it as a document problem. Get the value model right, add a risk register, phase the roadmap, approval follows. In a company of 50 to 500 people it is a calendar problem first. There is no standing AI budget to draw from, no Chief AI Officer whose discretionary line this falls under, and whoever assembles the case is doing it alongside their actual job.
What there is, every year, is a renewal. Microsoft’s published terms for Microsoft 365 Copilot plans are annual and auto-renewing, with the promotional add-on rate applying to the first year only. Anthropic’s Enterprise seats are billed annually, with usage charged on top at API rates. The consequence is the same across vendors: once a year a number you have already paid gets re-presented, and the default answer is yes.
That is the only moment when three things are true. The spend is visible, the person who signs it is paying attention, and nobody has to invent a new line item, because the comparison is against last year’s invoice, not against zero.
What the renewal quote actually commits you to
Read the quote as a buyer, not a renewal. Published US list prices at the time of writing, from each vendor’s page:
| Plan | Per user, per month | Terms |
|---|---|---|
| Microsoft 365 Copilot Business (add-on) | $18.00 yearly, $25.20 monthly | Promotional through 31 December 2026, first year only, from a $21.00 list price. Requires an eligible base plan |
| Microsoft 365 Business Standard with Copilot | $23.50 yearly, $28.20 monthly | Annual, auto-renewing. No separate base license |
| Microsoft 365 Business Premium with Copilot | $32.00 yearly, $38.40 monthly | Annual, auto-renewing |
| Claude Team Standard | $20 billed annually, $25 billed monthly | 2 to 150 seats, mixable with Premium seats |
| Claude Enterprise | $20 billed annually | 20-seat minimum, usage billed at API rates on top |
Two details matter more than the headline numbers. Several of these plans bill advanced features by usage beyond the license fee, so the renewal line is not fixed the way a per-seat software line used to be. And the promotional rate on the Microsoft add-on steps up after the first year, so a quote that looks flat may carry a built-in increase.
Call it $25 per seat per month as a blended figure. A 180-person company with 150 AI seats is renewing $45,000 a year, and that spend is already approved. What the quote does not say is what any of it bought; the arithmetic of a rollout that was never sequenced is a separate and larger bill.
The number no renewal quote contains
A renewal quote has a seat count, a unit price, and a term. It carries no capability number; neither does your admin console. Microsoft’s documentation defines Active Users as licensed people who “tried a user-initiated Microsoft Copilot feature” in the selected window, which can be set to 7, 28, 90, or 180 days. One prompt, one app, inside the window. Useful for license hygiene, silent on whether anyone got better at their job, which is the gap covered in why usage dashboards overstate real capability.
The best substitute is scored testing rather than telemetry. Section’s AI Proficiency Report, published June 2026, assessed 5,026 US knowledge workers with hands-on skill tasks rather than self-report. Workers whose employers provided training specific to the tools being deployed averaged 44 out of 100. Where the same tools went in without it, the average was 25.
Put both halves of that comparison in your document, not just the favorable one. A 1.8x difference is the strongest evidence available that the training lever exists. It is also an association rather than a promise about your company, and 44 out of 100 is a failing grade. A case that quotes the 44 and hides the rest will not survive the first reader who opens the source.
Three questions an AI enablement business case has to survive
Every rejected version of this case dies on one of three questions. Answer them before you write anything else.
“Did we not already pay for this?” The licenses bought access. They did not buy method, and nothing in the subscription delivers it. The renewal funds the tool; the enablement line funds the use of it. Different purchases, different vendors.
“How do we know it worked?” Not from a usage chart. Define two or three recurring deliverables, time them before the session, and retest the same tasks eight to twelve weeks later with no refresher. A number that can come back flat is the only kind worth promising, and the retest date belongs in the approval.
“Who owns it in month four?” The common failure is not a bad session but a good session with nobody holding the calendar afterward. Name the owner, the review cadence, and how much of their time this takes before anyone signs. Treating the work as a change program rather than an event is the argument in enablement as change management rather than instruction.
Fund the AI enablement line out of the renewal itself
Here is the part no vendor template includes: where the money comes from. The renewal is also a seat audit, and seat audits generate cash.
Export the user-level table from your admin console. Microsoft’s usage report carries per-app last-activity-date columns, exports to CSV, and covers everyone licensed “at any point over the past 180 days,” including people who never used the product once. Sort by last activity and count the seats with no activity date at all in the full window.
Those are the safest seats to drop, because the question of whether training would have activated them does not arise: nobody ever opened it. Keep the sparse users. They are your training population.
Zylo’s 2026 SaaS Management Index, drawn from roughly 40 million licenses and $75 billion in spend, puts average unused licenses at 46 percent in a typical 30-day period. Leave that figure out of your case. It is vendor customer data across all software categories, and your tenant has a real number you can produce in ten minutes.
Run the arithmetic on the 180-person company above:
| Line | Before | After |
|---|---|---|
| AI seats renewed | 150 | 130 |
| Annual license spend at $25 per seat per month | $45,000 | $39,000 |
| Enablement line | $0 | $6,000 |
| Total | $45,000 | $45,000 |
Twenty never-touched seats at $25 a month release $6,000 a year, which lands inside the market range for a facilitated workshop. Total AI spend is unchanged, the license line goes down, and no new budget was requested. That is a reallocation memo rather than a funding request, and a much easier document to get signed.
How big the ask should be
Two benchmarks make the number legible to the two people most likely to push back.
For a CFO, express it as a share of the renewal. Facilitated LLM fluency workshops sit in a $2,000 to $15,000 market range depending on headcount, session count, and how much work goes into building the exercises from the team’s own documents. Against a $45,000 license line that is roughly 4 to 33 percent, with the middle of the range under 20 percent. “Another 15 percent on top of the renewal to make the renewal worth paying” is a sentence a finance lead can evaluate.
For a Head of L&D, express it per employee. ATD’s 2026 State of the Industry report puts average direct learning expenditure at $846 per employee against 16.7 formal learning hours, down from $1,254 the previous year, across roughly 340 organizations. An $8,000 workshop for 60 people is $133 per head, set against a line that already carries several hundred dollars per employee.
Resist sizing the request as a percentage of an “AI budget.” At 50 to 500 people that budget usually does not exist as a line, and inventing one invites a governance conversation that takes a quarter.
Which statistics to cite, and which to leave out
Credibility here is asymmetric. One unqualified number loses more than three good ones gain, because the reader who checks is the reader who signs.
| Statistic | Use it? | Why |
|---|---|---|
| Section, June 2026: 44 out of 100 with tool-specific training against 25 without, from hands-on tests of 5,026 US knowledge workers | Yes, with both halves | Scored tasks, not self-report. Say it is an association and that both scores are low |
| ATD 2026: $846 direct learning expenditure per employee | Yes | Independent benchmark your L&D lead already knows |
| Vendor pricing from vendors’ own pages | Yes | Verifiable in one click, which is the point |
| Your own dormant-seat count | Yes, lead with it | The only number in the document nobody can dispute |
| DataCamp and YouGov, early 2026: just over 1 in 5 leaders report significant positive AI ROI, against 42 percent where data and AI literacy upskilling is mature | Only with the caveat | CIO Dive’s coverage notes this compares the whole sample to a subgroup, not two split groups. Directional, not a doubling claim |
| McKinsey’s State of AI: high performers are roughly 6 percent of respondents and far likelier to have redesigned workflows | Sparingly | Enterprise-weighted sample. Supports method over tooling, nothing more specific |
| The widely quoted claim that 95 percent of generative AI pilots show no measurable return | Leave out, prepare for it | One preliminary study with a narrow definition of failure. Someone will raise it, so hold the methodology answer rather than the number |
| Vendor claims of a 353 percent three-year ROI | Never | No published methodology |
The pattern across two years of these conversations, examined in what readiness signals actually predicted, is that cases built on the reader’s own data clear faster than cases built on industry statistics. The statistics are context. Your tenant is evidence.
A six-week sequence working backward from the date
The renewal date is fixed, so plan against it. Six weeks is enough and four is not, because procurement and finance each need a week nobody scheduled.
Week six. Find the actual renewal date and the auto-renewal notice period in the contract. Pull the seat count and per-seat price from the last invoice rather than from memory.
Week five. Export the usage report and count never-activated seats, sparse users, and regular users. Those three numbers are the spine of the case.
Week four. Pick the two or three workflows the enablement line is meant to move and time them as they are done today. A rough baseline takes an afternoon and beats no baseline.
Week three. Get two or three quotes for a facilitated session built on your own documents, so the price in the memo is real.
Week two. Write the memo: current line, revised seat count, released amount, enablement line, retest date, named owner. One page.
Week one. Present it as a renewal decision rather than a training proposal. The signature you need is on the renewal either way.
If this year’s renewal already passed, run the sequence anyway, document the dormant seats now, and present the case in month eleven when the quote arrives. The seat audit only gets easier with two windows of data instead of one.
For the full arc from a license rollout to a capability number you can report, see the AI enablement playbook, and for why a seat invoice and a capability budget are two different problems, the broader argument covers the ground underneath this one.
If you would rather not assemble the numbers alone, we run a free AI-readiness call that works through your seat data, your dormant-seat count, and what a defensible ask looks like at your headcount.
Frequently asked questions
What is an AI enablement business case?
A short document that justifies spending on capability rather than access. It states current license spend, what that spend has and has not produced, a specific training ask, where the money comes from, how success will be measured, and who owns the result. One page is usually enough at 50 to 500 people.
Why tie it to a license renewal?
Because the renewal is the only moment when AI spend is already approved, already visible, and already in front of the person who signs. A request made then is a reallocation; the same request in a random month is a new budget line, and new lines compete against everything else.
How much should we budget for AI enablement?
Facilitated LLM fluency workshops run roughly $2,000 to $15,000 in the market depending on headcount and session count, commonly 4 to 33 percent of a mid-market annual license line. Per employee it usually sits well under the several hundred dollars per head already carried in a typical learning budget.
Where does the money come from if there is no AI budget?
From the renewal. Export your usage report, count the seats never activated in the last 180 days, renew fewer, and move the difference to an enablement line. Twenty never-touched seats at $25 a month release $6,000 a year with total spend flat.
Is it safe to cut seats before training people?
Cut only the seats with no recorded activity at all. Those people never opened the product, so training was never the blocker. Keep every sparse user: they are who the session is for, and dropping them removes the evidence you are about to generate.
What should we promise as the return?
A measurement, not a percentage. Name two or three recurring deliverables, record how long they take now, and commit to retesting the same tasks eight to twelve weeks after the session with no refresher. Put the retest date in the approval; a promise that can fail is the only one worth making.
Who should own the AI enablement line after approval?
Whoever already signs the renewal, usually a COO, CFO, or VP of Operations, with a named day-to-day owner underneath. Both belong in the memo with an honest estimate of how much time the work takes, because an owner without calendar time is the most common reason month four looks like month one.
What if leadership says the tools are intuitive enough?
Point at the scored data rather than arguing. Section’s June 2026 testing of 5,026 US knowledge workers found 82 percent could not correctly identify what an AI agent is on a multiple-choice question, and fewer than one in ten could define one in their own words. Intuitive access is not working knowledge.
Key takeaways
- The renewal date, not the quality of the document, decides whether an AI enablement business case gets funded. Six weeks before it the ask is a reallocation; any other month it is a new budget line.
- Read the renewal quote as a buyer. Published prices run about $18 to $32 per user per month across the mainstream business plans, several bill advanced features by usage on top, and promotional rates step up after year one.
- Neither the quote nor the admin console contains a capability number. Microsoft counts a single prompt in a window as active usage, and the only credible substitute is a scored retest of real work.
- Fund the ask from the renewal itself: drop the seats with no activity at all over 180 days, renew fewer, and move the released amount to an enablement line.
- Cite your own tenant data first and published research second, state the limits of every statistic you use, and leave out the contested failure-rate headline while preparing the answer to it.
Arthur Wandzel
