The Agency AI Savings Checklist

01 Why you have to ask

AI didn't kill the agencies and IT firms. It became part of what they sell you. Their own numbers show it, and they also show where your savings are.

The gap. HCLTech, one of India's biggest IT firms, reports something it calls advanced AI revenue, and it decides what counts as AI. That grew 78% in a year. The whole company grew about 2%. So part of what clients used to pay for is now sold as AI, or the same work is done in fewer hours. Buyers who ask are collecting. According to the Financial Times, KPMG pushed its auditor, Grant Thornton, to share the savings from AI and reportedly warned it could switch. The fee went from $416,000 to $357,000, about 14% less. Neither firm will discuss the terms. The big clients aren't waiting. In June, HFS Research reported that companies are reopening IT contracts within two years of signing them, instead of waiting for renewal.

The AI part is growing fast. The whole bill isn't. That gap is where your savings are hiding.

Nobody will call you. Those renegotiations are huge contracts run by procurement teams. If you pay a monthly retainer, you have to ask.

02 Question 1: where does AI do the work?

You want three answers: which tasks, which AI model, and who checks the work before it goes out. A vendor that uses AI well can answer all three in one email. HCLTech, for example, says its Claude-powered workflows keep a person in charge of high-impact changes.

Email to your agency or IT vendor

Hi [NAME],

Before our next review, could you show me where AI does the work on our account?

1. Which tasks on our account does AI now do, fully or in part?
2. Which AI model or tool does each one use?
3. Who on your team checks that work before it goes out under our name?

If something that used to take a day now takes an hour, I'd like to know which tasks those are.

Thanks,
[YOUR NAME]

What a good answer looks like. A short list of named tasks, a named model or tool for each, and a named person who reviews. "We use AI across everything" answers none of the three.

03 Question 2: hours or results?

If your contract still counts hours, and AI cut the hours, one of these should change. Pick the one you want and get it in writing.

The price comes down. Same work, fewer hours, a lower bill.

You get more work for the same money. Name the extra deliverables in the contract.

A price tied to a result. You pay for an outcome, not for the time it took.

A share of the savings. What the big clients ask for: when AI cuts the cost of delivery, part of it comes back to you.

This is where the big firms are heading. At Accenture, fixed-price work, which includes deals priced on outcomes, is over 65% of new bookings. WPP's CEO, Cindy Rose, said in August that billing by time probably isn't sustainable, and also that only one WPP client pays on outcomes so far. So don't wait for your vendor to offer it.

The line to send

Since AI now does part of the work on our account, I'd like to review how we're billed. Either the price reflects the hours AI saved, or we agree on [MORE DELIVERABLES / A PRICE TIED TO A RESULT / A SHARE OF THE SAVINGS] for the same fee. Can we put that in writing before [DATE]?

04 Question 3, and what to keep in house

Whatever they build on your data, you own. The prompts, the workflows, the agents. If your agency builds a customer service agent trained on your emails and you leave, you shouldn't have to start again from zero. Get it agreed before you sign. It's much harder after.

Ownership line for your contract

All prompts, workflows, AI agents and other work built on [COMPANY]'s data under this agreement belong to [COMPANY], and will be handed over in a usable form if the agreement ends.

You don't need to drop your agency. Wiring AI into your systems, testing it and fixing it when it breaks is real work, and it's worth paying for. But keep these three things on your side:

Your customer knowledge. Your data, your notes, what your customers actually say. Every AI tool runs on it, so it lives with you, not in a vendor's account.

The final say. On anything that goes out under your name.

One person who can try it. Someone on your team who can open the same AI your vendor uses and try a task themselves. That's how you know what a fair price looks like.

Sources

HCLTech, Q2 FY27 results (October 2026) CRN Asia, HCLTech sets up an Anthropic business unit TechSpot, KPMG asked its own auditor for a discount, citing AI (reporting the Financial Times) HFS Research, AI drives IT contract renegotiations within 24 months Accenture Q4 FY26 earnings call transcript Digiday, WPP CEO Cindy Rose on outcome-based pay

Mathew's LibraryLibrary card

Free access

Keep reading free with your library card

One email opens this checklist and all 3 resources in Mathew's library. No account, no password.