How to Package a Maintenance Plan for AI Builds
- A maintenance plan for AI builds sells the outcome of a system that keeps working - monitoring, small changes, and a guaranteed response time - as a flat monthly fee, so you are not quoting every fix and the client is never surprised by a bill.
- Price it at 10 to 20 percent of the build cost per month, tiered by response time and how many small changes are included. Frame it as insurance on a system the client now depends on, not as on-call labor.
- Offer it at handoff, when the build is fresh and the value is obvious, not months later when something has already broken. The plan turns a one-time project into predictable revenue that compounds across every client.
What a Maintenance Plan for an AI Build Actually Covers
A maintenance plan for an AI build is a flat monthly fee under which you keep a client's system running, make small changes as their business shifts, and respond fast when something breaks. It is not open-ended support and it is not a new project every month. It is a defined promise: the thing you built stays working, and you are the one who makes sure of it. The client pays for peace of mind on a system they now depend on, and you get revenue that shows up whether or not anything goes wrong.
The distinction that makes it sellable is outcome versus labor. If you pitch 'I'll be available for support,' the client hears a cost and tries to minimize it. If you pitch 'your automation keeps recovering leads every month and I make sure it never quietly stops,' the client hears insurance on something valuable. Same work, completely different conversation - and it is the same shift behind [value-based pricing](/blog/value-based-pricing-stop-charging-hourly).
AI builds specifically need maintenance in ways a static website does not. Models get updated, APIs change their responses, the client adds a service that the automation needs to know about, usage costs drift, and edge cases surface once real volume hits. A build that was perfect at launch will need small touches within weeks. That is not a flaw in your work - it is the nature of a living system, and it is exactly what justifies the plan.
The Three Things Every Maintenance Plan Must Promise
A vague plan is a plan the client will try to cancel. Make the promise concrete with three pillars, each of which the client can picture. We teach this as the Keep-It-Running Plan inside the Profit Room, and every strong maintenance offer has all three.
| Pillar | The promise to the client | What it actually costs you |
|---|---|---|
| Monitoring | We watch that the system is running and catch failures before you do | A logging dashboard and a periodic check, mostly automated once set up |
| Small changes | A set number of minor tweaks a month - a prompt, a script line, a rule | A few hours a month, batched, on your schedule not the client's |
| Response time | If something breaks, we respond within a guaranteed window | The real value the client pays for - your priority when it matters |
The Keep-It-Running Plan - the three pillars of a maintenance offer
The response-time promise is what the client is really buying. A business that depends on an AI system does not want to file a ticket and wait a week when it stops. A guaranteed window - next business day on a basic tier, same day on a premium one - is worth more to them than the actual hours it takes you to fix most issues. Sell the window, deliver the fix quietly.
How to Price a Maintenance Plan
The clean rule is 10 to 20 percent of the build cost, charged monthly. A $4,000 build carries a $400 to $800 a month plan; a $10,000 build carries $1,000 to $2,000. This is a rule of thumb from how established software and agency retainers are priced, not a guaranteed market rate - so treat it as a starting anchor and adjust to what the specific client's system is worth to them. The percentage scales naturally: bigger, more critical systems cost more to keep running and are worth more to protect, so the fee tracks both your effort and the client's dependence.
Tier it so the client chooses their level of protection rather than choosing whether to have any. A basic tier is monitoring plus next-business-day response and one small change a month. A premium tier is same-day response, several changes, and a monthly check-in call. Most clients pick the middle, which is exactly where you want them. The good-better-best structure is laid out in [productized service tiers](/blog/productized-service-tiers), and the recurring-revenue tradeoffs are in [retainer vs productized subscription](/blog/recurring-revenue-retainer-vs-subscription).
Do not underprice the plan to seem generous - an underpriced plan trains the client to treat you as unlimited on-call help for pocket change, which is worse than no plan at all. If the fee does not cover a monitoring habit plus a few batched hours plus the option value of your fast response, it is too low. The undercharging trap and how to escape it is in [why builders undercharge](/blog/why-builders-undercharge).
When and How to Sell the Plan
Sell the maintenance plan at handoff, not later. The moment the build goes live is when its value is most vivid to the client - they can see it working, they are already thinking about depending on it, and you have their trust from delivering. Wait three months and you are cold-selling an afterthought; offer it in the delivery conversation and it is the obvious next step. Build it into how you close out every project, the way [how to deliver a build so clients come back](/blog/deliver-a-build-so-clients-come-back) frames the handoff.
Frame it as the default, not the upsell. 'Here's the plan that keeps this running and gets you a fast response if anything ever breaks - most clients start on the standard tier' assumes the plan, which is very different from 'do you want ongoing support?' The first is a professional handing over a system responsibly. The second is a salesperson reaching for more money. Present the price without flinching, as covered in [how to present your price without flinching](/blog/present-your-price-without-flinching).
If a client declines, do not fight it - set a trigger instead. Note that their system will need attention within a few weeks, and follow up the first time something does. A client who watched their automation break and remembered you offered a plan is the easiest renewal you will ever get. Either way, the plan turns one-time projects into revenue you can forecast, which is the whole game of moving [from freelancer to agency](/blog/from-freelancer-to-agency).
Short, practical drops on offers, outreach, pricing, and closing clients with Claude Code. No spam, unsubscribe anytime.
Frequently asked
What should a maintenance plan for an AI build include?
Three things: monitoring so you catch failures before the client does, a capped number of small changes each month, and a guaranteed response time if something breaks. The response-time promise is what the client is really paying for - a business that depends on the system does not want to wait a week for a fix. Keep the line clear between maintenance, which is running and small tweaks, and new work, which is a change order or a new project.
How much should I charge for a maintenance plan?
A common starting anchor is 10 to 20 percent of the build cost per month, so a $4,000 build carries roughly $400 to $800 a month. That figure comes from how software and agency retainers are typically priced, so treat it as a starting point and adjust to what the client's system is worth to them and how critical it is. Tier it by response time and number of included changes, and never underprice it - an underpriced plan trains the client to treat you as unlimited on-call help.
When is the best time to sell a maintenance plan?
At handoff, when the build is live and its value is most vivid to the client. Waiting months means cold-selling an afterthought, while offering it in the delivery conversation makes it the obvious next step. Frame it as the default - 'here's the plan that keeps this running, most clients start on the standard tier' - rather than an optional upsell. If they decline, set a trigger and follow up the first time the system needs attention.
Why do AI builds need maintenance more than a regular website?
AI builds are living systems. Models get updated, APIs change their responses, usage costs drift, the client adds services the automation needs to know about, and edge cases surface once real call or message volume hits. A build that was perfect at launch will need small touches within weeks - that is the nature of the work, not a flaw, and it is exactly what a maintenance plan exists to cover.
How do I stop a maintenance plan from becoming unpaid full-time work?
Cap the small changes in writing - 'up to two minor changes per month, each under an hour' is a promise you can keep, while 'changes as needed' is an open door to scope creep. When a request exceeds the cap, convert it into a quick change order rather than absorbing it. Naming the line between maintenance and new work up front is what keeps the flat fee profitable instead of turning it into unlimited on-call labor.
Last reviewed September 15, 2026.
Co-founder of the Claude Code Profit Room. Built and sold AI services to real clients; writes about offers, pricing, outreach, and closing with receipts.
More from Duncan Rogoff →
