Course 18 walked through usage-based, credit, bring-your-own-key, and managed pricing for a single product you own, where the core question was who bears the AI provider’s variable cost. Pricing client services carries a related but distinct question: not just who bears the AI provider’s cost, but how you get paid for your own time and expertise delivering, maintaining, and supporting that work on someone else’s site. The same underlying tension, cost predictability versus flexibility, shows up again here, just applied to your own labor and the client relationship instead of a product license.
Lesson 1’s productized-service framing. This lesson assumes you have a defined scope to price against, pricing an undefined scope is a separate problem this lesson doesn’t solve.
Step 1: The four models, compared for client services specifically
| Model | How it works | Who bears your time risk | Who bears AI provider cost |
|---|---|---|---|
| Project-based | A fixed price for a fixed scope (Lesson 1’s productized package), paid on delivery or in milestones | You, if the work takes longer than estimated | Usually the client, through their own provider account or a pass-through line item |
| Retainer | A recurring monthly fee for ongoing work: maintenance, monitoring, small feature additions, support | Shared, a well-scoped retainer caps your hours but a vague one exposes you to scope creep | Usually the client, ongoing usage is easiest to bill through their own account |
| Usage/credits passed through | You bill the client based on their actual AI usage volume, sometimes with your own margin added on top of the provider’s cost | You, for the setup and monitoring work, low ongoing risk otherwise | The client, directly proportional to their own usage |
| Fully managed | One flat recurring fee covers everything, delivery, hosting relationship, AI provider costs, and support, client never sees a provider bill | You, fully, for both your time and the AI provider’s bill | You, fully, unless usage limits are built into the agreement |
This is the same structural question Course 18 raised for product pricing (who bears the variable cost, and how much friction does shifting it create), applied here across two separate variables at once: your own delivery time and the client’s ongoing AI provider bill. A model can pass one of those risks to the client while you keep the other.
Step 2: Project-based pricing fits one-off setup work
A fixed project price fits Lesson 1’s productized packages well, because the scope is already defined and the delivery timeline is already known from having built the same package before. The real risk is entirely on you: if a client’s specific content model turns out to be messier than a typical engagement, you absorb that extra time unless the SOW’s exclusions (Lesson 1) are tight enough to justify a change order. Most agencies handle AI provider costs in a project-priced engagement by having the client supply their own API key or provider account from day one (the same bring-your-own-key pattern Course 18 covered for products), which keeps ongoing usage cost entirely off your books once delivery is complete.
Step 3: Retainers fit ongoing maintenance and support
A retainer is the standard recurring-fee arrangement for work that doesn’t end at delivery: monitoring an MCP server’s error logs, updating abilities as a client’s content model evolves, responding to a client’s support requests about the AI feature’s behavior. A retainer needs a clearly stated scope the same way a project does, a common structure is a fixed number of hours or a fixed set of included tasks per month, with anything beyond that billed separately or bumped to the next cycle. An imprecisely scoped retainer (“ongoing AI support” with no further detail) is the single most common way agencies end up doing unbounded work for a bounded fee, so the same discipline Lesson 1 applied to project scope belongs in a retainer’s scope too.
Step 4: Usage/credits and fully managed sit at the two ends
Passing usage or credits through to the client keeps their AI provider cost visible and proportional to their own usage, which is easy to justify since it maps directly to something the client can see (how much their assistant gets used). A fully managed model removes that visibility entirely: one flat fee, no provider bill the client ever sees, which is easier to sell to a non-technical client but puts you in exactly the position Course 18 flagged for managed product pricing, fully exposed to variable AI cost unless the agreement includes explicit usage limits. The same rate-limiting patterns Course 4 covered for security reasons are the practical safeguard that makes a fully managed client offering financially survivable rather than a standing risk of an unbounded provider bill landing on you.
If you offer a flat-fee, fully managed AI service to a client, state a usage ceiling in the agreement itself, not just in your own rate-limiting code. A client whose usage grows well beyond what the fee assumed should trigger a real conversation about upgrading their plan, not a silent margin loss on your side that you only notice months later when reviewing provider invoices.
Step 5: Why most agencies mix models rather than picking one
In practice, the common pattern is a project fee for initial setup (Lesson 1’s package, priced and delivered once) followed by a retainer for ongoing maintenance and support, with AI provider costs handled through the client’s own account in both phases. This mirrors Course 18’s observation that many real products offer more than one pricing tier rather than forcing every customer through the same model, just applied across the lifecycle of a single client relationship instead of across different customer segments.
Recap
Client AI/MCP pricing has to answer two separate questions Course 18’s product pricing only had one of: who bears the AI provider’s variable cost, and who bears the risk that your own delivery time runs over. Project-based pricing fits one-off, already-productized setup work. Retainers fit ongoing maintenance and support, and need the same explicit scope discipline as a project to avoid unbounded hours for a bounded fee. Usage/credits and fully managed pricing sit at the two ends of who sees the AI provider’s bill, with a fully managed offer needing real usage limits in the agreement, not just in code, to stay financially safe. Most agencies end up combining a project fee with an ongoing retainer rather than forcing a client into one model for the whole relationship.
Resources & further reading
- Pricing Models for AI Features, Course 18, the product-pricing version of this same tradeoff
- WordPress MCP Security & Authentication, Course 4, the rate-limiting patterns a managed offer depends on