How agencies can productise AI creative into a fixed-price package

If you bill by the hour and your team gets faster, your revenue falls. That is the plain problem with AI tools for an agency that sells time. A productised service (a service sold as a fixed package rather than as hours) turns the same speed into margin. The client gets a known result at a known price. You keep the efficiency.

This page is about how to build one for AI-made creative: images, short video, social sets. It does not repeat the pages on retainer pricing or scope creep. It adds the parts that are new when AI is doing part of the work: retry costs, regeneration caps, consent and disclosure inside the package.

What productised means: one outcome, one price, one turnaround

A package has three fixed things:

• One outcome. "Ten social posts and three short loops", not "creative support".
• One price. A number the client can approve without a call.
• One turnaround. "Five working days from a complete brief."

Everything else is a wall around those three: what is included, what is not, how many rounds of changes, and what counts as a complete brief.

It is different from a retainer. A retainer reserves your time each month and is open about hours or deliverables. A package sells a defined output. Some agencies sell a package once; others sell it monthly. The shape is the same.

Why a fixed fee fits AI work

With hourly billing, every hour AI saves is passed to the client one for one. A job that took 10 hours and now takes 4 loses 60% of its revenue. With a fixed fee, the client pays for the result and the efficiency stays with you. Clients know this. A survey of more than 180 agencies by the software company Productive reported that about a third had been asked for an AI discount and nearly half expected to be, while most held their prices. It is a vendor-run survey, so treat the numbers as a signal and not a measurement.

The honest argument to a client is not "AI is free". It is "you are paying for a reviewed, on-brand result by Friday, and for the risk we carry if it is wrong." For the conversation itself, see what to say when a client asks "can't we just use AI?".

Pick the outcome clients already buy repeatedly

Do not invent a new product. Look at your last 12 months of invoices and find the work that repeats: the same type of asset, for the same type of client, with a brief that is mostly the same. Good candidates:

• Social content sets for retainer clients.
• Product image sets for small online stores (see the product photo package).
• Short-form video ads in a fixed number of variants.
• A pitch or concept preview (see mood video for a client pitch).
• Ad resizing and variants for existing campaigns.

Poor candidates: brand identity, a long film, anything where every project is genuinely different. Quote those per project.

Build the price: cost per package with retries and review time

Use this formula:

Price = (staff hours x loaded hourly cost + tool cost x retry multiplier + review time cost) divided by (1 minus target margin)

• Loaded hourly cost means salary plus overhead per working hour, not just pay.
• Retry multiplier is how many generations you pay for per keeper. If one in four passes, it is 4.
• Review time is the human check on every asset. It is the cost people forget.

Worked example, with made-up numbers: a "Launch Visual Kit" of one hero image set, 10 social variants and 3 short loops.

Line	Illustration
Brief and setup	1.5 hours
Creating and selecting assets	6 hours
Review and fixes	3 hours
Revisions (2 rounds)	2 hours
Staff total	12.5 hours at $50 loaded = $625
Tool cost with a retry multiplier of 4	$60
Cost	$685
Price at a 45% margin	$685 divided by 0.55 = about $1,245, rounded to $1,250

Two rules make this safer. First, price on the 90th-percentile retry rate, meaning the bad weeks, not the average. One stubborn asset with hands, text or product accuracy problems can eat a day. Second, track the real hours for your first three packages and adjust. For measuring tool spend per client, see tracking AI spend by client.

Five package examples for an agency that uses AI

All prices below are made up to show structure. Use your own numbers.

1. Launch Visual Kit. 1 hero image set, 10 social variants, 3 short loops. 5 working days. 2 revision rounds. $1,250.
2. Monthly Social Pack. 12 posts and 4 short videos, billed in 4-week cycles. 2 rounds per batch. Pause allowed.
3. Product Shot Pack. 20 images and 5 clips for one product line. Real packshots required from the client.
4. Ad Variant Sprint. One approved ad turned into 12 variants in 4 sizes, delivered in 3 days.
5. Pitch Preview. A 30-second concept video and 6 stills to show a prospect the idea. Fixed price, credited against the project if they sign.

Offer two or three, not twelve. A menu of five packages with three tiers each becomes a custom quote in disguise.

For comparison, two well-known design subscriptions publish their prices. As of November 2026, DesignJoy's site lists a monthly plan at $4,995 with one request in progress at a time, an average 48-hour delivery, and pause or cancel at any time. It says video, animated graphics and 3D are not supported. ManyPixels lists monthly plans from $699 to $2,599 with unlimited requests and revisions in a queue, and cancel with no fee or pause for $10 a month. These are vendor figures and change. Neither page says how its designers use AI, so I will not claim that. They show that clients already understand the "queue" form of subscription, which is worth borrowing.

Scope walls: inclusions, exclusions, revisions and the queue

Scope on the thing that costs you, not on "AI or human". Good scope levers are:

• Concept directions. "Up to 2 directions, then a change order."
• Final sizes and formats. Number of sizes per asset.
• Realistic people. Allowed or not. Realistic faces raise review time and rights questions.
• Rush. A premium for under three days.
• Revisions. A fixed count, with extra rounds priced separately. One software vendor suggests replacing unlimited revisions with a fixed number, for example 2, and gives an example overage of $50 per extra round. Use your own number.

Avoid "unlimited" anything. The honest form of unlimited is a queue: one request at a time, in order. If you do sell it, say so plainly.

Add a regeneration cap line: "We create as many drafts as we need. You review and approve final assets only, and change requests beyond the included rounds are quoted." That protects you from a client who asks for "just ten more versions" because they believe it costs nothing.

For the paperwork around changes, use a change order template and a scope of work template.

Rights, disclosure and consent inside the package

This is the part the standard productised-service guides skip. Put it in the package, not in a side email. This is information, not legal advice. Check local law and each platform's current rules.

• Exclusions. No real person's face or voice without written consent. No logos or brands you do not own or have a licence for. No images that show a product doing something it cannot do.
• A delivery note. Which tools were used, whether any client product photo or person's likeness was uploaded, and any platform label guidance. This also helps if a rights question appears later. See a plain policy on telling clients and the contract clause for AI use.
• Training data. Many AI platforms may use uploaded content to train their models unless you are on a special business agreement. Do not promise a client that nothing will train a model unless you have checked the tool's terms. See AI tools that do not train on your data.
• Platform labels. Platforms such as YouTube, Meta and TikTok have their own rules on labelling realistic AI content. Check the current rule for each channel the client uses.

State AI use in the scope. Selling a package as "no AI" to premium buyers, or hiding it from price-sensitive ones, makes every invoice look suspicious later. Sell the outcome and the turnaround, and be factual about how it is made.

Quality checks that run every time

A fixed price only works if quality is fixed too. Write one checklist and run it on every asset before the client sees it:

1. Facts, prices and claims match the client's approved sources.
2. Product, logo and brand colour are correct.
3. All text is legible and spelled correctly, placed from approved files.
4. People shown are consented, and none resemble a real person without written consent.
5. The platform's AI-label decision is made.
6. File is named and logged with tool, prompt, reviewer and date.

For video there is a longer method in the AI video quality check, and for sign-off mechanics see the creative approval workflow.

Selling it: discount requests and "why not just use AI"

Three questions will come. Have an answer for each:

• "Can we have a discount since it is AI?" Offer a smaller scope or a faster turnaround, not a lower price for the same scope. A fixed fee makes this easy: the package is already priced at the market, not at your cost.
• "Why not do it ourselves with AI?" They can. What you sell is the reviewed result, the brand fit, the legal care and the time. Show a before and after of a reject you caught.
• "Is it AI?" Yes, partly, and here is the scope note that says so.

Pause and cancel are features. Billing monthly packages in 4-week cycles with a pause option reduces the churn that comes from a client having nothing to ask for this month.

Pilot with three clients and a pass/fail

Do not launch on your website first. Sell the package to three existing clients at a pilot price (not free). Then check:

• Pass: every package delivered on the promised turnaround, margin above your target after counting review time, and no scope dispute.
• Fail: any one of those misses. If it fails, reduce the scope, not the price. Fewer concept directions, fewer sizes, fewer formats.

Run two more cycles after changes, then publish.

When a package is the wrong shape

• Every project is different, such as brand identity or a film. Quote per project.
• The client needs a human designer in touch every week. A design subscription from a specialist can fit better than a package you run yourself.
• You do not yet know your real hours. Track 4 to 6 weeks of time first.
• You cannot say no. A package with no walls is a retainer with a worse price.

How I'd approach this

I would pick one package, the one the agency already sells most often, and write it on one page: outcome, price, turnaround, what is excluded. I would price it with the formula above on my worst weeks, test it on three clients and only then publish it.

The repeatable delivery side is where I can help. Blur Studio builds the workflow around the tools: an intake form, a shared asset library, a client approval portal and a log of which tool and prompt made each file. It is described on the business page.