How to sell AI employee setups to small businesses: the integrator model

You delivered the pipeline. Stages are set, fields are required, the sales team is trained, the invoice is paid. A month later the client calls, and the question is not about the pipeline. It is about who answers WhatsApp at eleven at night, when the lead from the ad arrives and the salesperson has gone home.

The client asks the integrator because there is nobody else to ask. It is also the easiest sale you have: they already pay you, already trust you, and the thing they are asking for already sits in their head next to the CRM setup you did. The only thing missing is a line on your price list.

Below: what a setup of an AI employee consists of, what people charge for, where the recurring income comes from, and how to hand the client a finished project so that they own it and you are not left holding their passwords. Disclosure: we build CRAIM, and the transfer mechanics described here are its. The setup itself is the same work for any AI employee, wherever it lives.

Why this work lands on the integrator

The platform vendor has a product and documentation. The client has a folder of materials and an idea of what they want. Between the two sits work neither of them will do: turning price lists, an old website and the owner's notes into knowledge an AI can answer from without inventing anything, and writing the instruction that keeps it inside the lines.

That is exactly the work you already do when you set up a CRM, on a different subject. You already know how to get pipeline stages out of a client who did not have them in their head. Getting the rules for answering customers out of them is the same job.

Three things make it easier than a CRM setup. The client sees the result in minutes, not a quarter: the employee answers its first message while they watch. Nothing is programmed: the role and the instruction are written in plain language. And a mistake is visible immediately, because in copilot mode every reply waits for a person.

What a setup consists of: the checklist

The list follows what gets created in the project, in the order it makes sense to do it.

1. Reading the materials. The website, price lists, files, a link to a Google Sheet. From these, twelve knowledge documents are built: company profile, buyer profiles, playbooks, offers, objections, tone of voice, pipeline rules, response times, catalogue, guides, legal notes, case studies. The client hands over half of it on day one; the other half you pull out with questions.
2. The catalogue. If the client sells products or services with prices, they go into a table with columns, search and filters. That is what the employee reads when a customer asks whether there is anything cheaper. Update the table and the answers update, with no change to the instruction.
3. The instruction. The role, what it handles, what it never promises, when it calls a person. This is the most expensive hour of the whole job and the most underrated: the client thinks it is configuring a bot, and it is writing the sales department's rulebook, which they did not have before you.
4. Channels. WhatsApp, Telegram, Avito, Max, Meta, a phone line. One channel binds to exactly one employee, so two employees cannot share one number: that takes separate numbers or bots.
5. The process. Who acts and when: a message arrives, the employee answers, the customer goes quiet for two days, a call goes out. Conditions and waits are drawn in the builder, not written into the instruction. The instruction governs how the employee acts within its turn; the process governs when the turn comes.
6. Autonomy and approvals. Copilot by default: every reply waits for a person. Autopilot goes on when the drafts stop surprising anyone. Specific actions, a discount for instance, can wait for approval permanently.
7. Two weeks of reading drafts. This is part of the setup, not support: two weeks show what the knowledge is missing, and the instruction gets corrected against real conversations.
8. Migration from the previous CRM, if there is one. From amoCRM the import pulls contacts, deals, tasks, companies, the pipelines with stage mapping, and the SMS and call history logged on the records; messenger chats stay behind, because amoCRM keeps them in a separate Chats API. From Bitrix24 and HubSpot: contacts, deals, companies and tasks. From a CSV, only the records. The migration is done into the project before the handover, not after.

What to charge, and how the money adds up

You set the setup price. The platform is not part of it and takes nothing from it, so the benchmark is your own price list for CRM setups. In volume of work, an AI employee setup is comparable to a mid-complexity pipeline; in how much the client notices the result, it is well above, and that belongs in the price.

Three lines of income make up one model. The table has it; the middle row is the platform's real numbers, the other two are yours.

The referral line is easy to compute exactly, because the plans are public. Growth is $29 a month, so one client on Growth brings $5.80 a month and $69.60 a year. A client on Team, $49, brings $9.80 a month and $117.60 a year. A client on Business, $99, brings $19.80 a month and $237.60 a year. None of that replaces the setup fee, but ten clients on Team over a year is nearly $1,200 that nobody invoiced.

One condition gets forgotten: the commission is credited only for a client who signed up through your link. A client who registered on their own and then accepted a project from you is not attributed to you afterwards. So the order is strict: link first, transfer second.

How to hand over the project without keeping the client's passwords

An integrator usually has two bad options. The first is to create the account under the client's email and work in it. Then the password is shared, the action history is mixed, and the client cannot tell who pressed what. The second is to build everything in your own account and then redo the setup in the client's. Then the work is done twice.

In CRAIM a project moves from one workspace to another the way a domain does. You build it in your workspace, next to your other clients' projects. When it is ready, on your workspace home you press Transfer to a client and enter their email. The offer lives seven days. Until the client accepts, you are the owner and the payer, and the project keeps working.

The client sees the offer in their own account, on the same home page. They pick which of their workspaces and which plan the project lands in, and tick a box for whether you stay as technical administrator. The box grants access to that one project, with no access to their billing or their other projects. Then they press Accept project.

From that second the project is theirs. What happens at that moment is worth knowing in advance, because the client will ask. The full sequence, with every check, is on the project transfer page in the knowledge base.

• The project's owners become the owners of the client's workspace. Every previous member of the project is suspended, and pending invitations are revoked.
• Every active session in the project is signed out. The password you used during the setup no longer works there.
• If the box was ticked, you are re-added as an administrator of the project.
• Your billing period for the project closes and the client's opens, stamped with the same second. Model usage is charged to their balance from then on.
• The project is not copied. Employees, knowledge, catalogue, channels, deals, conversations, processes and history stay where they are and change owner.

Two checks can stop a transfer, and it is better to pass them beforehand. First, the client's plan has to fit the project's active AI employees: one on Start, five on Growth, fifty on Team. A project with two employees accepted onto Start is refused, with the reason stated. Second, while a call, an AI task or a message delivery is running in the project, the transfer waits for it to finish.

And one thing that does not exist: an email to the client. When you send the offer, the platform notifies nobody. You tell the client yourself, as you would about any other delivery. That is deliberate: a project holding a company's conversations and phone lines should not move on the strength of an email that can be missed or forged.

What it costs you

While the project is yours it counts against your plan, specifically against its limit on active AI employees. Start, at $9 a month, holds one active employee. Growth, at $29, holds up to five across every project in the workspace. Team, at $49, up to fifty. Hand a project over and its employees free up a place for the next client.

Your workspace is free for the first 14 days at the Business level, with a $5 credit for model usage. The trial is granted once per email address, ever: there is no second one for you or for the client, and opening workspaces under different addresses to get around that is pointless, because the project leaves for the client's plan anyway.

Who this is not for

A client with no materials. If the company has no price list, no description of its services, not even a message history that shows how it answers, the employee has nothing to answer from. Materials first, then the setup, and that is a separate piece of work worth selling too.

A business with five enquiries a week in one channel. The employee will pay for itself, but the client will not feel the difference, and you will be left supporting a cheap project expensively.

An integrator who wants white-label. The client sees CRAIM and knows where their project lives. There is no reselling of the subscription through the integrator: the client pays for their own plan, you get your setup price and the referral share.

The order for the first project

1. Register your workspace. The first two weeks are free, which is enough for a whole first project.
2. Send the client your referral link from the account menu and ask them to sign up through it, under the email they will actually use. This can happen on day one; they only see the project when it is transferred.
3. Create the client's project in your workspace. Load the materials, build the knowledge and the catalogue, write the instruction.
4. If the client is leaving amoCRM, Bitrix24 or HubSpot, migrate the history now, while the project is yours.
5. Connect the client's channels and start the employee in copilot mode. Read the drafts with the client for two weeks.
6. Press Transfer to a client, enter their email, and tell them. Agree beforehand whether they tick the box for your technical access and what the support covers.
7. After acceptance, check that you see the project as an administrator and that billing has moved to the client.

The short version

Setting up an AI employee is the same work as setting up a CRM: getting the rules out of the client and turning them into a system. It sells to the same people, the same way, at your own price. On top of it sit 20% of the client's subscription payments for a year and a monthly support fee. The project is built in your workspace and handed to the client with one button, which they press themselves. The mechanics, the prices and the usual questions are on the integrators page.