Your clients' AI bill belongs to nobody. You sell the review, your logo is on it, and the recurring revenue is yours to set.
The gap your competitors haven't priced. Almost every MSP now sells "AI enablement and governance." Almost none of them sells the thing the client's CFO actually feels: the AI line on their bill was set on prices that keep moving. I keep the daily log of those moves — 467 models tracked — and I package it into a monthly review you deliver under your own name.
"Not necessarily being just an IT provider, but how are you transforming how this business works, how it performs, how it increases its revenues, how it reduces their expenses, how it mitigates risk utilizing AI?" Danny Perry, managing partner, ITCubed (Houston) — CRN XChange NexGen panel, reported 23 Oct 2026
"But training, for sure, we need to be collecting on stuff like that." Brian Ruschman, president, C-Forward (Covington, KY) — same panel
Paraphrased from the same panel: Ruschman said the data he has seen points to more than half of employees using six or more AI tools a day, and that those tools are "not being governed for the most part." Same panel, CRN's report on XChange NexGen 2026
Two MSP 500 honorees on one stage, both saying the same thing: the consulting and the reporting are where the margin is, and clients will pay for it. The audit is the part with a number attached — which makes it the easiest renewal conversation you will have all year. Six tools per employee, ungoverned, is also the reason a client's AI spend is usually bigger and messier than their own IT budget line says.
Deloitte's 3Q 2026 CFO Signals survey (200 North American CFOs, all at $1B+ revenue, published 21 Oct 2026): 74% need several business days to several weeks to pause an AI cost overrun, 53% only ever review AI spend retrospectively off vendor bills, and 56% have nothing beyond basic guardrails. Most of them cannot tell you what a token costs. That is the gap you are selling into — and the audit is the only deliverable that closes it without buying a platform.
An eight-line worksheet that takes a client's AI and software spend and returns three numbers: spend today, spend after the cuts, and what has to be re-priced because the vendor changed the price. Runs on a CSV export — no client credentials, no screen sharing.
A one-page template in your branding: the cuts, the dated evidence for each, and the decision the client has to make. Written to be forwarded to a CFO, not read aloud in a meeting.
Every model price change, logged daily, with an embeddable chart you can put on your own site or in your monthly newsletter. Each entry is dated — so a saving you claim is checkable, which is the difference between a report and an opinion.
When a client's bill is big enough to be worth the deep work, I do the audit behind you. You keep the relationship, the account and your margin; I stay invisible.
| Mode | Price to you | What you sell it as |
|---|---|---|
| Monthly review retainer | $300 per client account / month | Your own line item — most MSPs price it $750–1,500 per account |
| One-off client audit | $500 fixed, or 30% of verified savings | A project fee, or a "we only get paid if we find money" offer |
Two MSPs running the retainer across ten client accounts is $3,000/month of recurring work I do not have to sell again. That is the whole business model, stated plainly.
I ran this on my own stack first and cancelled what didn't earn its keep: $5,988 a year off four line items, each one itemised with the date it was cut. Read the receipts.
Today's log, for instance, retired three GPT-3.5 Turbo variants outright — a client still invoicing against those numbers is budgeting for a model that no longer exists.
Ask for the worksheet and the report templateWhat this is not. It is not a monitoring platform, a licence to resell software, or a promise that every client is overpaying — some are billed correctly and the review will say so. It is not an AI-governance or security assessment. The worksheet runs on exports the client hands over; I don't need access to their systems, and no client data leaves the machine it's analysed on.
Prices quoted are proposed rates, not a signed rate card. Figures on this page come from the public log at /data/changes.json, refreshed daily.