← All articles
Content MarketingBobos AI

The Math Behind a $3,000/Month AI Retainer

You built an AI assistant. They love it. Now they ask what it costs monthly. You guess $200 and end up working for free. Here's the actual math behind a retainer that works.

You built an AI assistant for a client. They love it. Now they ask: what does it cost to keep this running?

You say something like $200 a month because you're scared to sound expensive. Two months later you're answering Slack messages at 9 PM for free, updating prompts when their sales process changes, and onboarding their new hire on how to talk to the assistant.

The problem is not that you undercharged—it's that you guessed. Here's how to build a retainer with actual math behind it.

Start with What the Client Uses Every Day

A retainer is not a discounted project. It's access to AI tools the client's team uses daily.

Before you think about price, list what you're giving them: how many assistants, which roles can use them, what data sources they connect to. Be specific.

Example: Three assistants (sales, operations, finance), five users, connected to their CRM and accounting system. That's the offer.

The base of your retainer is consistent access and uptime, not custom builds. You're not selling a project that ends—you're running infrastructure. Clients pay monthly because the AI is in their daily workflow, not because you're on call waiting for tickets.

What this means for you: If you can't describe what the client gets every month in one sentence, you don't have a retainer yet. You have a support contract.

The assistant that checks every inbound lead against their CRM and flags duplicates? That runs every day. The one that pulls last month's financials and writes the board update? That runs on the first of the month. Daily use is what keeps them paying.

Add the Work You're Already Doing for Free

You update prompts when a process changes. You add a new data source when they acquire a company. You train a new hire on how to use the assistant.

Right now that's goodwill. In a retainer, it's a line item.

Here's what to include:

  • Prompt tuning and assistant updates: Included up to 4 hours per month
  • New role or data source requests: One per quarter included, $500 each beyond that
  • Training and onboarding: Two sessions per year included

These are not upsells—they're the reason the retainer exists. Without you, the AI stops matching how the business actually works. The assistant that was perfect in March is wrong by June because they changed how they qualify leads.

Your clients don't know how to fix that. You do. That's what they're paying for.

What this means for you: The invisible labor you're doing for free today is worth $100-150/month per client. Put a name on it and bill it.

Account for Platform and Model Costs

You pay Dispatcher per active client. The client pays for their own AI usage through their keys. Your cost is predictable; theirs scales with how much they use the AI.

Build both into the pricing conversation.

Here's the math:

  • Dispatcher: $29 per active client per month (after the first two, which are free)
  • Model spend: Client brings their own Anthropic or OpenAI keys; you set a spending cap and show them the bill
  • Example: $29 platform fee + $40 estimated model spend = $69 in costs you need to cover

Model spend depends on usage. A client with three people using two assistants daily might spend $30-50/month. A client running an assistant that processes every inbound email might spend $100. You won't know until month two.

That's fine. Set the cap at $100 to start. If they hit it, you get an alert and you talk about whether to raise it or tune the prompts to use fewer tokens. Either way, you're not eating the overage.

What this means for you: Your platform cost is fixed at $29 per client. Model spend is variable but visible. You can price with confidence because you know your floor.

The $3,000 Retainer Breakdown

Here's what a $300/month retainer looks like for one client, and what happens when you have ten of them.

One Client

  • Revenue: $300/month
  • Costs: $29 Dispatcher + ~$40 model spend = $69
  • Margin: $231 per client (77%)

Ten Clients

  • Revenue: $3,000/month
  • Costs: $290 platform + ~$400 model spend = $690
  • Margin: ~$2,310 (77%)

At this margin, you can afford to spend 8-10 hours per month per client on updates, training and support and still be profitable. That's two hours a week across your book of business to keep ten clients happy.

Compare to project work: ten $5,000 builds = $50,000 once, then nothing. Ten retainers = $36,000/year, every year. By year two, the retainers have made you $72,000 from the same ten clients.

What this means for you: A $300 retainer is not expensive. It's the minimum price that lets you do the work well and still have margin left over.

If you're spending more than 10 hours per client per month, either your price is too low or your delivery is inefficient. Fix one or the other.

What Keeps Them Paying

Clients don't stay because you handed them a workflow. They stay because their team uses the AI every day and you make sure it keeps working.

The retainer funds that relationship.

Usage drives retention. If three people use the assistant daily, they won't cancel. If one person used it twice in March and forgot about it, you'll lose them by May. Your job is to make sure the first thing happens.

Your monthly check-in is not a status call—it's you showing them what the AI did this month. "Your sales assistant answered 340 questions, flagged 12 duplicate leads, and saved your team from calling the same prospect twice. Here's the record."

Governed access means you see every tool call, so you know when adoption is slipping. If usage drops, you find out why and fix it before the client notices. That's what separates a retainer from a handoff.

What this means for you: Retention is not about being nice on Slack. It's about making sure the AI gets used. Track it, report it, and act when it drops.

Start with Two, Price for Ten

Your first retainer will feel expensive to propose and cheap six months later. That's fine. Price it so you can say yes to the second client without rethinking your model.

Here's how to start:

  1. Use the same price for your first three clients. Learn what the work actually costs before you start customizing. If you charge client one $250, client two $350, and client three $300, you have no idea what your model is.
  2. Dispatcher is free for two active clients, so your first two retainers are pure margin minus model spend. Use that room to learn.
  3. By client four or five, you'll know if $300 is right or if you should be at $400. Don't change pricing mid-contract, but adjust for the next one.

The Founding 50 get white-label free for six months, which means you can remove the "Powered by" badge and run everything under your brand from day one. If you're signing your first retainer clients now, that's $299/month in cost you don't pay while you figure out your delivery model.

What this means for you: Don't wait until you have ten clients to start charging like you have ten clients. Price for the business you're building, not the one you have today.

Run the Numbers, Then Run the Business

A $3,000/month book of business is ten clients at $300 each, $690 in platform and model costs, and $2,310 in margin you can actually run a company on.

The math works because you're not selling hours—you're running AI infrastructure your clients use every day. Build the retainer around what they use, charge for the work you're already doing, and start with two.

Dispatcher is free for two active clients, no card required. Build your first retainer, test the math, and see if it holds. Start free with 2 clients—you'll know by client three if you priced it right.

We’d like to use analytics and advertising cookies (Google Analytics, Google Ads, and LinkedIn Insight, loaded via Google Tag Manager) to understand how visitors use this site and to measure our advertising. Nothing loads unless you accept. You can change your choice any time via Your Privacy Choices. Details in our Privacy Policy.