How Should Agencies Price Their Services in the AI Era? Own the Margin or Donate It
The short answer: agencies should price the outcome and the judgment in the AI era — never the hours, and never your costs. AI just collapsed the cost of delivery, and any pricing model anchored to cost passes that entire windfall to your client as a discount, while outcome-based pricing keeps it as margin. The margin AI creates belongs to whoever owns the machine that produces the outcome — so your pricing model isn't a billing preference, it's a bet on what you own.
You've re-quoted the same scope three times this year and eaten the difference every time. The work got faster. The invoice got awkward. And somewhere in the back of your head, a voice keeps asking whether you're supposed to be charging less now.
Say the quiet part out loud: you charge clients $5K a month so they can own a working marketing system — and every time AI makes your delivery cheaper and you trim the invoice to "stay fair," you're donating your own AI margin back across the table. The single biggest windfall this industry has produced in a decade, handed over as a courtesy.
AI was supposed to replace guys like me. Instead I own the machine it threatened me with — and watching it run every morning is what flipped how I think about price. So when someone asks how agencies should price in the AI era, my answer isn't a rate card. It's a question: what do you own?
Why is cost-plus pricing a trap now that AI cut delivery costs?
Because cost-plus pricing has exactly one job — track your costs and add a markup — and when your costs collapse, it tracks the collapse straight into your revenue.
Hourly is the pure form of the trap. But most agency pricing is cost-plus wearing better clothes: the retainer that's secretly an hour budget, the project fee built bottom-up from "how long will this take." All of it makes the same promise to the client — you pay for our effort — and AI just made effort the cheapest ingredient in the building.
Here's the part the pricing guides skip. A price is supposed to split the value you create between you and the client. When one of our partners came to us spending $1,500 a month boosting posts and hit a $105K gross month in January 2026, the value created had nothing to do with how many hours anything took. Cost-based pricing was never logical — it was a truce. Effort was just the only thing both sides could count. AI ended the truce, because now nobody can pretend the hours mean anything.
And notice what's actually deflating versus what's appreciating. Production collapsed — anyone can generate fifty ad variations before lunch. Judgment didn't. Earlier this year we killed an ad after $3.02 of spend — 40 impressions, zero clicks, dead by the math, budget rerouted to the winner the same day. The generation cost nothing. Knowing to pull the trigger at three dollars instead of three hundred is the product. That's live-fire knowledge, paid for with our own money — and it doesn't show up on a timesheet.
You can't bill hours for a decision that takes four seconds and saves four figures. So stop pricing the part of your business that's deflating and start pricing the part that's compounding.
What agency pricing models actually work in 2026?
Two of the four standard models survive contact with AI — and only one of them builds margin.
Hourly: dead. Not declining — dead. It ties your revenue to the exact input AI deflates, which means every model release is a pay cut you scheduled for yourself. If a client insists on hourly, what they're really telling you is they intend to audit your effort instead of your results. Walk.
The flat retainer: survives, conditionally. Retainers are still the industry's default model, and they're not going anywhere. But a retainer only survives when it's anchored to a named outcome and backed by a system the client can see running. A retainer that's secretly an hour budget is just a discount negotiation on a delay timer. Anchored to an outcome — "we operate the machine that produces your booked calls, and here's this morning's numbers" — it becomes the hardest thing in your business to cancel.
Performance and hybrid (base + upside): where the margin lives. This is the only structure where AI's cost collapse works for you. You price the outcome; your cost to produce the outcome drops; the difference is yours — permanently, and growing with every model improvement. We run what we sell, so our contracts carry it in writing: net profit equal to or greater than total amount paid, or full refund. That's not bravado. That's what pricing looks like when you trust your own machine. Across accounts our team managed, the best month produced $419,000 in contract value signed in January 2026 off ~$87K ad spend — when your system can produce that, charging for hours isn't humble, it's malpractice against your own P&L.
Productized: fine at the door, not the engine. Fixed-scope, fixed-price offers are great entry points — cheap for the buyer to say yes to, cheap for you to fulfill with a system. But as a core model they get shopped on price by definition, which drags you right back into cost-anchored thinking. Sell products at the door. Sell the machine inside the house.
Retainer vs value-based pricing: how do you choose in the AI era?
You choose based on what your machine can prove — not on what a pricing article says is modern. Performance pricing without a system underneath it isn't bold, it's gambling with payroll.
Here's the honest stage map:
No real system yet — deliverables are hand-made, results vary month to month, you couldn't state your cost per result without opening four dashboards. Stay on flat retainers anchored to a named outcome, and put your energy into building the machine, not redesigning the rate card. Pricing outcomes you can't reliably produce is how agencies write refund checks. The rebuild has an order of operations — I've laid out where to start — and it's shorter than most owners think.
System running, proof still thin — you have data flowing daily, kill rules you actually enforce, a few months of game tape. Move to hybrid: a base fee that covers operating the system, plus upside tied to the outcome. The base keeps the lights on while the upside trains both of you to talk in results instead of activity.
System proven — you know your numbers cold before the month starts. Our team has managed millions in ad spend and generated 2,700+ booked calls, and the number I'd defend in any room is $27.22 — our best cost per booked call. When you can quote your machine's output like that, weight the deal toward performance, and put a guarantee on it if you've got the nerve. The spread between your price and your collapsed delivery cost is the prize — and it widens every quarter the system runs, because the machine gets cheaper while the outcome holds.
The through-line: when delivery cost collapses, the margin goes to whoever owns the machine. Own nothing — rent tools, sell labor — and that margin leaks to your clients as discounts and to software vendors as subscriptions. Own the system — the data, the test history, the decision rules, the daily numbers — and you become the vendor that can't be cut, charging for something no cancellation gives back.
And here's how I know the market is already repricing this exact question. I wrote yesterday about who's actually buying our $27 playbook on building the system we run — agency owners, not the founder-led small businesses I wrote it for. They're not buying a PDF. They're running the make-vs-buy math on the machine itself — and deciding owning the engine beats renting labor that imitates one. If you want to run the same math, the playbook is $27 — the system, the kill rules, the morning numbers. Worth reading before you set next year's rates, not after.
FAQ
Should you give clients an AI discount?
No. Discounting because your delivery got cheaper concedes that you were selling effort all along — and it resets the negotiation from a lower base every time the models improve. The answer to "shouldn't this cost less now?" is to show the system and the judgment the subscription doesn't come with. I broke down the full margin math — and the ChatGPT objection script — in Will AI Kill Agency Retainers?
Is usage-based or hybrid pricing right for agencies?
Hybrid, yes — usage-based, mostly no. Usage pricing (per asset, per campaign, per credit) re-anchors the conversation to inputs, which is the exact thing AI is deflating. A base fee for operating the system plus upside on the outcome is the strongest practical structure: downside protection for you, outcome alignment for the client.
How much should an agency charge for AI services?
Don't sell "AI services" as a line item at all — any line item with AI in the name gets priced against a $20/month subscription, and you lose that comparison every time. Price the business outcome the system produces and keep the tooling invisible. The client is buying booked calls, revenue, and a machine they can watch run — not your stack.
When should an agency switch from retainers to performance pricing?
When your numbers get boring — when you can state your cost per result before the month starts and the variance is small enough that a guarantee doesn't scare you. Until then, run flat retainers anchored to outcomes and build proof. The switch itself works best in stages: add an upside component at each renewal instead of repapering every deal at once.
Cost-plus pricing was a truce, and AI ended it. The margin it freed up is sitting on the table right now, in every engagement you have — and it's going somewhere. To your clients as discounts. To the software vendors as subscriptions. Or to you.
The pricing model isn't a billing preference. It's a bet on what you own. Own the machine underneath the business — and stop betting against yourself.
