How to price AI-made creative without losing margin

I write this for the owner or head of a small agency, two to thirty people, who sells ads, social content, short video or brand visuals. You have started using AI image and video tools. Your cost to make things has changed, but your price list has not. This page gives you a way to reset the price using your own numbers, not a vendor's.

One warning first. Most of the pricing advice you will find on this topic comes from software vendors who sell agency tools. Their percentages ("60 to 80 percent faster", "93 percent margin") come with no method and often contradict themselves inside the same post. I treat every such number as a vendor claim and so should you. Everything with a dollar sign below is either from a vendor's own price page (and dated) or clearly labelled as made up.

The efficiency penalty: why hourly billing punishes you

If you charge by the hour, every improvement in your tools lowers your revenue. A task that took ten hours now takes four. At $100 an hour your invoice fell from $1,000 to $400, and the client got the same result. You did the clever work and were paid less for it.

Clients see this too. In a 2025 Digiday report, seven independent agency leaders described how they handle it. One agency keeps billing by the hour but delivers more assets for the same hours. Another mixes a fee for output with a fee tied to performance. A third, Luquire, had used a productized, fixed-output approach since 2019. Their shared point is that buyers increasingly expect AI savings to be passed on, so you need a plan before the client raises it.

The fix is not to hide the savings or to hand them all over. It is to stop selling hours. Sell a defined result (ten finished ads, a monthly pack of social posts, a launch video) at a price based on what the result is worth to the client and what it truly costs you.

What your AI-made asset really costs

Three costs sit under every finished asset. Write them on one line each.

1. Tool cost, including retries. AI tools charge per attempt, and not every attempt is usable. If it takes four tries to get one keeper, you pay for four. The number of tries per keeper is your retry rate.
2. Human time. The brief, writing the prompts (the typed instructions you give the tool), reviewing each result, fixing defects, and client revision rounds. Review time is the line people forget.
3. Overhead. Rent, software, management time, the share of a producer's day spent on this client. Use your loaded hourly rate (salary plus overhead, divided by billable hours) and this is already inside it.

The formula:

Cost per keeper = (tool cost per try x tries per keeper) + (human minutes / 60 x loaded hourly rate)
Margin = (price minus cost) / price

Where the tool cost comes from

Most tools sell credits, a balance you spend per generation. To turn credits into dollars, divide the plan price by the credits it includes, then multiply by the credits one try costs. Three examples read from vendor pages on 4 November 2026. Prices change often, so confirm each on the vendor page before you use it:

• Runway, a video generation service: the Standard plan was $15 a month for 625 credits. The page shows its Gen-4.5 video model at about 12 credits per second, so about 52 seconds of video a month on that plan, or roughly $0.29 per second of output.
• Adobe Firefly, Adobe's image and video generator: individual Pro was $23.99 a month for 4,000 credits, listed as up to 40 five-second videos. Credits reset monthly and do not roll over.
• Higgsfield, a platform that bundles many image and video models: credit cost depends on the model and action, and some are shown only on the Generate button. For agency seats, as of November 2026, the pricing page lists Team at $79 per seat monthly or $65 per seat on annual billing (2 to 9 seats) and Scale at $215 monthly or $150 per seat annual (5 to 15 seats). See how to spend Higgsfield credits wisely and measure your own cost per approved result. If you need one plain per-second price to put in a quote, Runway's published rate is easier to use than a platform where cost varies by model.

Do not judge a tool by its sticker price. A $60 plan with a low retry rate can cost less per keeper than a $15 plan where nothing comes out right.

A cost-per-asset sheet you can copy

Here is the sheet. Every retry rate, review time and price below is made up for the example. The video tool cost uses the Runway figure above ($0.024 per credit, 12 credits per second) because it is a published rate. The loaded rate is $60 an hour. Replace each input with your own.

Asset	Tool cost per try	Tries per keeper	Tool cost per keeper	Human minutes	Human cost	Total cost	Price to client	Gross margin
Social image	$0.15	3	$0.45	25	$25.00	$25.45	$90	72%
Product image with on-screen text	$0.15	6	$0.90	45	$45.00	$45.90	$140	67%
10-second video ad	$2.88	4	$11.52	90	$90.00	$101.52	$250	59%
Pack of 20 video variants	$2.88	3 each	$172.80	180	$180.00	$352.80	$900	61%

Three things stand out, and they are true whatever your real numbers are.

• The tool is the small part. In the video row the tool cost is about 11 percent of the total. People who say "AI makes creative nearly free" are looking at that one column.
• Retries multiply the tool cost, not the human cost. Retries matter most for video and for anything with on-screen text or a real product, where garbled letters and wrong shapes force extra tries. Track tries per keeper by asset type for two weeks. A starting guess to test, not a fact: images 2 to 4, short video 3 to 6, text or product-accurate work 5 or more.
• Review time can match generation time. Twenty variants at six minutes of checking each is two hours. If review time per asset is more than about 25 percent of what the same asset took to produce the old way, you are not saving much, and you should not discount.

Add a fix budget as well: set aside roughly 10 to 20 percent of the asset's human time for defects a person must repair (broken text, off-brand colors, a wrong product detail). That range is an assumption to calibrate, not a benchmark.

Four ways to price it

1. Per asset. A fixed price for each finished image, video or variant, with a stated number of revision rounds. Easy for clients to compare and to approve. Risk: clients order many small things and your admin cost grows.
2. Productized tier. A named package with a fixed monthly output, for example "Growth: 12 images, 4 videos, 2 revision rounds, one month". Predictable for both sides. The agency software vendor Apaya says two rounds of revisions per content batch are included and extra rounds are billed hourly, with a three-month minimum term as a separate suggestion. Treat that as one vendor's suggestion, not a standard, but the shape is sound.
3. Retainer with overage. A monthly fee covers a stated capacity; work past it is billed at a set rate. For the mechanics of retainers in general, see agency retainer pricing; I do not repeat them here.
4. Outcome-linked. Part of the fee depends on results such as leads or sales. Be careful. If you control only the creative and not the targeting, budget or landing page, you are carrying risk you cannot manage. A small agency should keep the outcome-linked part to a minority of the fee, or skip it.

Pick this model if

Your situation	Use
Same deliverables every month (social packs, ad variants)	Productized tier
Scope is uncertain (rebrand, campaign concept)	Project fee with a cap, or hourly with a ceiling
Client wants a single price per item to compare	Per asset, with a minimum order
Ongoing relationship, volume swings month to month	Retainer with overage
Client owns the whole funnel and you also run the media	Small outcome-linked bonus on top of a base fee
Strategy, direction and creative judgment	Day rate or fixed fee for the thinking, kept apart from production

A simple rate card has three anchors: a pilot (a small fixed deliverable, such as five assets, so the client can try you), a monthly tier, and a day rate for direction. That covers most small agencies.

Keep the fee, raise the output

When tools get faster, the easiest move to defend is to keep the price and increase what the client receives: more assets, more variants, a faster turnaround, or all three. One agency in the Digiday piece described this pattern. The saving is visible to the client as extra value, and your revenue does not fall.

Write it into the scope of work so it does not drift. A sample line you can adapt:

"Each monthly package includes 16 finished assets (previously 10) and a three-working-day turnaround for first drafts. Two revision rounds are included per batch. Further rounds, or changes to direction after round two, are quoted separately."

If a client's buyer insists on a lower fee, offer a trade: a lower price for a smaller scope, a longer minimum term, or a faster approval process on their side. Do not cut the price and keep the scope. For the change-request side of this, see agency scope creep.

Do not sell what you cannot promise

Agencies often charge a premium for exclusive creative. AI output makes that risky. The Higgsfield terms, for example, say outputs may not be unique across users and give no guarantee of originality or exclusivity. In the US, the Copyright Office has said that prompts alone do not make someone an author. So I would not price "exclusivity" as an extra unless a contract term truly supports it. Sell speed, volume and the quality of your direction and review instead. The contract wording is covered in what to put in your contract when you use AI tools. This is information, not legal advice; check local law and each tool's current terms.

What to tell clients about AI use

Do not hide it, and do not apologize for it. A short, plain statement works: "We use AI tools to speed up drafts and variants. A named person on our team reviews every deliverable before it reaches you, and we do not enter your confidential material into tools without your agreement." The full question of when you must tell a client, and when a platform requires a label on the ad itself, is in should you tell clients you used AI.

For price, the honest framing is: "The tools make the first draft quicker. You are paying for the direction, the checking and the responsibility for what goes live." If a client says "AI makes this free", walk them through the cost sheet above in two minutes. The tool line is small; the people line is the product.

Absorb the tool fee or pass it through?

A rule of thumb to test (not a benchmark): absorb software subscriptions when they are below about 2 to 3 percent of the client's monthly fee. Pass through, at cost or with a small handling margin, any credit-heavy video bursts, with a monthly cap written in the contract. Log credits per client so the bill matches reality; the sheet for that is in track AI spend by client. That page is about logging and billing credits. This one is about setting the price and the margin target.

Moving existing clients without a fight

• New clients and new scopes first. Put the new model on every new proposal, so you learn what works before touching current accounts.
• Existing clients at renewal. Do not change prices mid-term. At renewal, show what they received last term, what they will receive next term (more output, faster), and the new price.
• Use a pilot for the nervous ones. Offer five assets at a fixed price for 30 days. If they like the result and the speed, move them to a tier.
• Spend the saved hours on strategy. Use time you get back for measurement, testing and planning. That is what clients cannot easily get from a tool and it supports a price rise.

A 30-day pricing test with pass and fail rules

Do not set your new price from a vendor's claim. Set it from your own month.

1. Pick one repeating service for one client. Record the old time per asset from past timesheets.
2. For 30 days, log for every asset: tries per keeper, tool cost, minutes to brief, review and fix, and client revision rounds.
3. Compute cost per keeper with the formula above and compare with the old cost.
4. Set the price so gross margin hits your target (many small agencies aim for gross margins of roughly 40 to 70 percent; treat that range as a rough guide, not a rule).

Pass: measured hours saved are 30 percent or more, revision rounds per asset did not go up, and the client accepted the work at the same quality. Fail: hours saved under 30 percent, or review and fix time ate the gain, or revision rounds rose. If you fail, do not discount. Narrow the use of AI to the asset types where it passed. To cut revision rounds specifically, see reduce revision rounds with AI.

Mistakes that quietly kill margin

• Unlimited revisions. Cheap generation tempts teams to say yes to endless tweaks. Limit rounds in writing.
• Ignoring retries. Quote on the tool fee and one try, then pay for six.
• Selling text-in-image work at image prices. Garbled lettering needs a person. Price it as the 5-or-more-tries row.
• Passing all savings to the client in month one. You cannot take it back later.
• Not counting seat costs. Per-seat plans grow with headcount. Compare a team plan with a shared workspace before you add people one by one; see best AI video generators for agencies and scaling video output without hiring.

Who can skip this

• A solo freelancer billing a single client by the project: a simple fixed quote and a note of your retries is enough.
• An agency whose clients all sit on a fixed retainer with no AI in the scope: the sheet matters once AI output enters the scope.
• Anyone hoping for a market-rate table. The per-video and per-word rates on vendor pages come with no method, so I give none.

How I'd approach this

I would not start with a new price list. I would start with the 30-day log on one client, because it replaces opinions with numbers. Then I would move to two or three productized tiers with a fixed revision count, keep the price, raise the output, and write the new scope into the contract. The part that breaks first in most small agencies is tracking: nobody records retries or review minutes, so the margin story is a guess. Time spent training the team on the tools also counts, and there is a one-week training plan.

If you want that tracking built into the way your team already works, Blur Studio builds repeatable workflows, client approval portals and shared asset libraries around these tools. You can read about it on the business page, with no pressure to buy anything.