To price AI services, stop billing by the hour and price the output, the outcome or a subscription. Set the price from your full delivery cost (people plus tools) and a target margin, check it against the value to the client, and price custom AI builds separately from ordinary work your team does faster with AI. Review the price every six months as your costs and results change.
This guide is for agency owners and operators deciding what to charge now that AI is part of production. It walks through the decision in seven steps: what you are actually selling, which pricing model fits, how to calculate cost once, a worked price and margin example, how to price custom AI builds, how to handle "AI discount" requests, and how to review prices.
Key takeaways:
- Separate two different things: AI-assisted delivery (your normal services, produced faster with AI) and AI implementation (building AI systems for the client). They are priced differently.
- Hourly billing is the wrong default for AI-assisted work because hours and output have come apart.
- Choose output, outcome or subscription pricing based on how clearly results can be attributed to your work.
- Count tool and model costs once, inside the unit cost. Price = cost / (1 - target margin).
- Keep efficiency gains as margin when the client gets the same deliverable; charge more only when AI creates something new for the client.
Step 1: Decide what you are selling
Most pricing confusion comes from mixing two kinds of AI work.
| AI-assisted delivery | AI implementation | |
|---|---|---|
| What the client buys | Your usual service (content, ads, design, reporting), produced with AI in the workflow | A system: a chatbot, an internal tool, an automated workflow, a retrieval setup on their data |
| What they judge | The deliverable and its results | Whether the system works reliably in their business |
| Main cost | People's time for strategy, review and editing, plus tool usage | Discovery, build, integration, testing, security review, handoff, maintenance |
| Main risk | Clients asking for a discount because AI "did the work" | Underestimating integration and maintenance |
| Pricing approach | Output, outcome or subscription (Steps 2 to 4) | Paid discovery, then fixed fee by milestone, plus a support retainer (Step 5) |
Write down which one each of your offers is. A content retainer that uses AI for drafts is AI-assisted delivery. A support chatbot you build and maintain for the client is AI implementation.
Step 2: Choose a pricing model
Why not hourly
Hourly billing assumes hours and value move together. AI-assisted work breaks that. For example, if a blog post took 8 hours at $150 an hour ($1,200) and now takes 3 hours, hourly billing turns the same deliverable into a $450 invoice. The client gets the same or better post; all of the efficiency goes to them, and you are trained to cut prices every quarter as the team gets faster.
When the hours no longer track the value, the price should not track the hours either.
The three models that work
| Model | How you charge | Works when | Watch out for | Example (illustrative) |
|---|---|---|---|---|
| Output-based | Per asset: per post, landing page, email, video, report | The deliverable is well defined and reviewable | Define what one unit includes, and limit revisions | $1,500 per long-form post including research, drafting, editorial review, two revision rounds and on-page SEO |
| Outcome-based | Against a measurable result: leads, sign-ups, ranked pages | Attribution is clean and you control the main inputs | Attribution disputes; results you do not control | $4,000 base plus $40 per qualified inbound lead |
| Subscription (productized retainer) | A flat monthly fee for a defined set of outputs | The work is recurring and repeatable | Scope creep without clear limits | $6,500 a month for 8 long-form posts, 30 social posts and a monthly report |
Pick by attribution clarity
| Service line | Attribution clarity | Suggested model |
|---|---|---|
| SEO and content | Medium | Output or subscription |
| Paid media | High | Outcome or percent of spend |
| Email and lifecycle | Medium-high | Outcome or subscription |
| Brand and creative | Low | Output or fixed fee |
| Web and product builds | High | Fixed fee |
| Recurring deliverables | High | Subscription |
Avoid pure outcome pricing where attribution is fuzzy or where the client controls key inputs (their sales team, their site, their budget).
Value-based pricing: the ceiling, not the formula
Value-based pricing sets the price from what the client gains, not what the work costs you. Use it to find the ceiling of your price range:
- Quantify the outcome with the client. What is a qualified lead worth to them? What would a one-point rise in conversion rate mean in revenue? What would hiring in-house cost?
- Price as a share of that value, low enough that the client keeps a clear return. The right share depends on how certain the outcome is and what alternatives the client has.
- Guarantee the floor, share the upside. A base fee that covers your cost and margin, plus a bonus when results beat an agreed target.
Example (hypothetical): instead of 40 hours a month of content at $150 an hour ($6,000), a content program priced at $5,000 a month for defined deliverables, plus $1,000 in any month after organic traffic has grown more than 10% quarter over quarter. With AI in the workflow, delivery takes fewer hours, so both the base and the bonus earn a better margin, and the client pays for results rather than time.
Productize repeatable AI-assisted services
AI makes productization easier because it makes production more consistent. A productized service has a fixed scope, clear deliverables and one price, so:
- Your cost per deliverable becomes predictable, which makes a fixed price safe.
- Delivery scales without adding people at the same rate.
- Clients decide faster on a defined package than on an open estimate.
Set a productized price between a cost-plus floor (Step 3) and a value-based ceiling. New services usually start nearer the floor and move up as you collect results. The service packages guide has a worksheet for building tiers, and productized services in AgencyPro lets clients buy them from a catalog with recurring billing.
Step 3: Calculate the delivery cost once
Build the cost of one unit (or one month of a subscription) from these lines, and add each cost only once:
| Cost line | What to include |
|---|---|
| Production time | Prompting, drafting, editing, design or build time, at your loaded hourly cost |
| Review and QA | Fact-checking, brand review, accessibility and legal checks |
| Account management | Briefs, calls, approvals, reporting, share of the account lead's time |
| AI and tool usage | Model or API usage for this unit, plus the share of fixed tool subscriptions |
| Overhead share | Software, rent and admin, if not already in your loaded hourly cost |
Two rules:
- Count tooling once. If your loaded hourly cost already includes software subscriptions, only add the per-unit usage (for example, API calls). Adding it twice inflates the price and makes you less competitive for no reason.
- Time the real process. Record hours for the first five to ten units. AI often speeds up drafting and slows down review, and the estimate you make before you start is usually wrong.
How to treat tool costs on the invoice
| Approach | How it works | When to use it |
|---|---|---|
| Bundle into the price (default) | Tool costs sit inside the unit or retainer price | Most AI-assisted delivery; keeps the conversation on outcomes |
| Pass-through with a handling fee | Tools billed at cost plus an agreed percentage | Clients who require cost transparency, or heavy usage that varies month to month |
| Separate line at cost | Tools listed at cost on the invoice | Rarely; it invites clients to negotiate each tool |
Usage-heavy work, such as large-scale generation or a client-facing chatbot with unpredictable traffic, is the exception. There, pass usage through or cap it in the contract, because one busy month can wipe out the margin on a fixed price.
Step 4: Work the price and margin example
Price = delivery cost / (1 - target gross margin)
A 40% markup is not a 40% margin: $300 cost x 1.4 = $420, which is only a 29% margin.
Worked example (illustrative, not a market benchmark): pricing one long-form article in an AI-assisted content program.
| Line | Calculation | Cost |
|---|---|---|
| Research and outline | 1 hour x $50 loaded cost | $50 |
| AI-assisted draft and editing | 2 hours x $50 | $100 |
| Fact-check and brand review | 1 hour x $50 | $50 |
| SEO and publishing | 0.5 hours x $50 | $25 |
| Account management share | 1 hour x $50 | $50 |
| Model and tool usage | Per-article API usage and tool share | $25 |
| Delivery cost | $300 | |
| Price at 40% target margin | $300 / 0.60 | $500 |
| Price at 55% target margin | $300 / 0.45 | $667 |
Then validate it:
- Against value: if one article typically brings the client more than its price in pipeline, the price is defensible even near the top of the range.
- Against alternatives: an in-house writer's loaded cost per article, freelance rates, and what similar agencies publish.
- Against capacity: at 5.5 hours per article, one person with about 120 delivery hours a month can produce roughly 20 articles. Price the subscription so the team that delivers it is not oversold.
If the cost-based price is far below the value-based ceiling, you have room to price higher. If it is above what the market pays, cut cost (fewer revision rounds, tighter scope) before you cut price.
Example gross margin targets
Planning targets to adapt, not survey data. Set yours from your own service-line results.
| Service line | Example target gross margin |
|---|---|
| Content and SEO | 55 to 70 percent |
| Email and lifecycle | 55 to 70 percent |
| Paid media (excluding spend) | 50 to 65 percent |
| Brand and creative | 50 to 65 percent |
| Web and product builds | 40 to 55 percent |
| Custom AI builds | 45 to 60 percent |
| Productized retainers | 55 to 75 percent |
If a service line keeps landing below your floor, reprice it, reduce scope or stop selling it. The agency financial reporting guide covers how to track margin per service line.
Step 5: Price custom AI builds separately
Custom AI work (a support chatbot, an internal assistant on the client's documents, an automated workflow across their tools) is software delivery. Price it in phases, with fixed fees against defined outputs rather than hourly:
| Phase | What it produces | How to price it |
|---|---|---|
| Paid discovery or readiness assessment | Use case, data sources, success measures, architecture, risks, a fixed quote for the build | Fixed fee; always paid, because it is a real deliverable |
| Build | The working system against the discovery spec, with test results | Fixed fee by milestone, from your estimated hours x loaded cost / (1 - target margin) |
| Stabilization and handoff | Fixes after launch, documentation, training | Fixed fee or included warranty period with a defined end date |
| Ongoing support | Monitoring, model and prompt updates, integration fixes, usage reporting | Monthly retainer, with model and API usage passed through or capped |
What makes build estimates run over: messy client data, integrations with tools that lack good APIs, security and privacy review, evaluation (testing the system against real examples until it is reliable enough), and changes to the underlying models or APIs after launch. List each as an assumption in the proposal, and price change orders for anything outside it.
For genuinely new AI services where neither you nor the client has a reference price, the discovery phase does double duty: it shows what the system is worth to the client, which tells you where to set the build price. Compare it with the client's alternatives (hiring a specialist, buying off-the-shelf software, other agencies), then adjust after the first few projects as you learn your real delivery cost.
Step 6: Communicate the price and handle objections
Decide when to charge more and when to keep the gain
| Situation | What to do with the price |
|---|---|
| Same deliverable, produced faster | Keep the price; keep the efficiency as margin |
| Internal improvements the client never sees | Keep the gain |
| Small enhancements that are becoming standard everywhere | Include them; do not price them separately |
| Deeper analysis, more coverage or more variations than before | Charge more, or offer it as a higher tier |
| Faster turnaround that matters to the client | Price speed as a premium option |
| A capability you could not offer before | Price it as a new service |
The grey area (the same article, but with more thorough research) depends on your market. If competitors already include it at current prices, include it too. If you are ahead, it supports a higher tier.
Lead with the outcome, and be honest about the method
Present what the client gets, not how you make it:
"Our content program delivers 12 research-backed articles a month for your target topics, with performance tracking and a monthly strategy review. The investment is $5,000 a month."
That is not the same as hiding AI. If a client asks how you work, or your contract or their policies require disclosure, explain plainly where AI is used and where people review and decide. Clients who find out later that AI was involved tend to feel misled, while clients who know upfront and see the results rarely object. See should agencies disclose AI use to clients for the disclosure side.
When the client asks for an "AI discount"
Three responses, in this order:
- Acknowledge the tools are cheap. "You are right that the AI tools themselves cost little. What you pay for is the strategy that decides what to create, the review that keeps it accurate and on-brand, and the optimization that drives results. A design tool subscription costs less than an hour of a designer's time; the value is in the person using it."
- Move the comparison to value. "We price on the deliverable and the results, not on hours. The fair comparison is what the program produces for you."
- Offer more, not less. "Instead of lowering the price, we can add two more articles a month at the current fee." If the budget really is fixed, offer a smaller scope at a lower price rather than the same scope for less.
Avoid cutting prices to "share AI savings." It sets a precedent you will face every quarter.
Price for the buyer you want
- Productized buyers want clear deliverables and a predictable monthly cost: output or subscription pricing.
- Outcome buyers want measurable results and will share upside: outcome pricing, if attribution is clean.
- Embedded-team buyers want a team they can direct: a monthly retainer with a defined team allocation.
Most agencies do better choosing one or two of these and declining work outside them.
Step 7: Review prices every six months
AI costs and speeds change faster than most agency costs. Twice a year, for each AI-assisted service:
- Re-time delivery. Compare actual hours per unit (from time tracking) with the hours in your price.
- Check tool and model spend per unit and per client, including usage that grew without anyone noticing.
- Look at margin by service line against your floor.
- Look at results and win rates. Rising results support a higher price; losing deals on price at a healthy margin suggests the package needs a smaller tier, not a discount.
- Decide one change per service: raise, hold, rescope or retire. Apply price changes to new clients first and to existing clients at renewal.
Common pricing mistakes
- Hourly billing on AI-assisted work. Every efficiency gain becomes a price cut.
- Discounts to "share AI savings." Clients learn to expect one every quarter.
- Counting tooling twice, once in the loaded hourly cost and again as a line item.
- Outcome pricing without clean attribution.
- No revision limits in subscription contracts.
- Pricing a custom AI build like a content job, with no discovery, no evaluation time and no maintenance budget.
- Never re-timing the work after the first price was set.
Frequently asked questions
Should we lower prices because AI made us faster?
Rarely. If the client gets the same deliverable, keep the price and the margin. Use the time you gain to deliver more, deliver better or add services. Lower the price only when the market price has clearly moved and you would otherwise lose good clients.
How do we handle clients who ask for an "AI discount"?
Acknowledge that the tools are cheap, explain that the price covers strategy, review and results, and offer more scope at the same price rather than the same scope for less. If the client's budget is genuinely fixed, offer a smaller package.
Should AI tooling cost be a separate line item or bundled?
Bundle it into the deliverable price in most cases. Pass usage through, or cap it in the contract, when usage is heavy and unpredictable, such as a client-facing chatbot.
What pricing model works best for AI service lines?
Output-based pricing for well-defined deliverables, outcome-based pricing when attribution is clean, and subscription pricing for recurring work. Choose by how clearly results can be tied to your work and by what your buyers prefer.
How do we price custom AI builds for clients?
In phases: a paid discovery or assessment, a fixed-fee build by milestone against the discovery output, a defined stabilization period, and a monthly support retainer with model usage passed through or capped. Price each phase from estimated hours at loaded cost and your target margin.
How often should we review AI service prices?
At least every six months. Re-time the work, check tool spend and margin per service, and apply changes to new clients first and existing clients at renewal.
Want to sell AI services as subscriptions, credits or fixed outputs? AgencyPro supports recurring billing and productized services on every plan, with credit-based and hourly services, time tracking and capacity management on the Pro plan and per-client profitability reports on Plus. Compare plans or book a demo.