# AI automation agency: what it is and how to start one in 2027

> An AI automation agency without the course hype: what it sells, how delivery works from discovery to go-live, startup costs, pricing, the niches that buy and the Meta and data risks.

- **HTML version:** https://tolky.to/en/blog/ai-automation-agency-what-it-is-how-to-start
- **Author:** [Marlos Carmo](https://tolky.to/en/author/marlos), Co-founder of Tolky
- **Published:** November 20, 2026
- **Reviewed:** November 20, 2026
- **Language:** en

## Summary

An **AI automation agency** is a services business that designs, deploys and runs AI agents for clients, most often in customer service and sales over WhatsApp. It sells discovery, implementation and monthly operations, and the decision that matters most at the start is where to build the agents: custom code, n8n or a ready-made platform. The risks are unofficial APIs, ignoring Meta's rules and handling client data without a data processing agreement.

The AI automation agency has become a business category. The term shows up in videos, courses and ads, usually with a revenue figure in the headline and little explanation of what actually gets delivered to the client. If you run a marketing or tech agency and research the topic, you find more promise than method.

This guide takes the opposite route. It explains what an AI agency sells, how delivery works from discovery to go-live, where to build the agents, what it costs to start, how to charge, which niches buy and which risks sink projects. No income claims and no shortcuts: only what you need to know before starting one, since you are the one who will answer to the client for the delivery.

*Updated in November 2026.*

**In short:**

- **An AI automation agency designs, deploys and runs AI agents for businesses.** The best-selling service is customer service and lead qualification on WhatsApp, followed by database reactivation and campaigns.
- **Revenue comes from three lines:** discovery, implementation and monthly operations. Operations is what turns a project into recurring revenue.
- **The technical choice decides how much of your team's time goes to maintenance.** Custom code and n8n give control and demand upkeep; a ready-made platform cuts the technical work and keeps the agency focused on the client.
- **The risks are well known:** unofficial WhatsApp APIs get numbers banned, Meta has banned general-purpose AI assistants since January 15, 2026, and data protection law requires a processing agreement with the client.
- **A marketing agency is already halfway there:** client base, relationships and funnel knowledge. What is missing is a delivery method.

## What is an AI automation agency?

**An AI automation agency is a services business that designs, deploys and runs artificial intelligence agents for other companies, charging for the project and, almost always, for monthly operations.** It does not build language models: it uses existing models and platforms and delivers the result configured for the client's business.

The core product is the [AI agent](https://tolky.to/en/glossary/ai-agent): software that reads the customer's message, understands the intent, answers based on what the business taught it and takes actions, such as checking an order, booking or logging an opportunity in the CRM. The guide on [what an AI agent is](https://tolky.to/en/blog/what-is-an-ai-agent-complete-guide) covers the types and how they work.

The difference from a software house is focus. A software house builds custom systems; an AI agency sets up, tunes and monitors agents that solve a business problem, such as unanswered leads or service that stops after hours. You will also see the term **AI agency**, which often covers workflows between systems on top of conversational agents.

## What does an AI automation agency sell?

**An AI automation agency sells four main services: a customer service agent, a lead qualification agent, database reactivation and campaigns answered by AI.** All of them happen where the end customer already talks, which in Brazil and much of Latin America is WhatsApp.

| Service | Problem it solves | What the agency delivers | How to measure |
| :--- | :--- | :--- | :--- |
| Customer service agent | Queues, delays and repeated questions | Agent with a knowledge base, handoff rules and business hours | AI resolution rate and first response time |
| Lead qualification | Paid-media leads that go cold without a reply | Agent that answers instantly, asks qualifying questions and logs to the CRM | Qualified leads per period and time to first contact |
| Database reactivation | Old contacts gone quiet | Segmentation, a campaign with an approved template and an agent for replies | Replies and opportunities per campaign |
| AI-answered campaigns | Broadcasts that generate replies nobody handles | Campaign plus an agent that talks to whoever replies | Conversations handled and opportunities per campaign |

### Customer service agent

This is the most common entry service. The agent answers frequent questions, looks up information in the client's systems and hands off to a person when the case calls for it, with the history attached. That handoff with context is called [handoff](https://tolky.to/en/glossary/handoff), and it is what separates a serious project from a bot that traps the customer. The [WhatsApp customer service](https://tolky.to/en/solutions/whatsapp-customer-service) solution shows how this works on a platform.

### Lead qualification

For agencies that already run media, this is the service closest to the client's results. The agent answers the lead in seconds, asks the questions that separate browsers from buyers and hands the salesperson an opportunity with context. The concept is in the [lead qualification](https://tolky.to/en/glossary/lead-qualification) glossary entry.

### Reactivation and follow-up

A dormant database is dormant revenue. The agency segments old contacts, builds a campaign with a Meta-approved template and lets the agent talk to whoever replies. The guide to [lead follow-up on WhatsApp](https://tolky.to/en/blog/how-to-follow-up-with-leads-on-whatsapp) shows cadences that work.

### AI-answered campaigns

[WhatsApp bulk messaging](https://tolky.to/en/features/whatsapp-bulk-messaging) creates reply spikes. Without an agent, replies wait for hours. With one, every reply becomes a conversation, and the campaign is measured in opportunities, not just sends.

## How does an AI agency deliver, from discovery to go-live?

**AI agency delivery follows six steps: discovery, agent design, knowledge base, integrations, testing and go-live with monitoring.** Skipping a step, especially testing, is what causes the most rework and unhappy clients.

1. **Discovery.** Conversation volume, most frequent questions, peak hours, who handles service today and where leads get lost. This produces the scope and the success metric.
2. **Agent design.** Mission (serve, qualify, reactivate), tone of voice, what it can and cannot do and when it hands off to a person.
3. **Knowledge base.** Documents, approved answers and business rules. The guide on [training the agent with a knowledge base](https://tolky.to/en/blog/train-ai-agent-knowledge-base) shows how to organize it.
4. **Integrations.** CRM, calendar, order system. Start with what the agent needs to look up to resolve the most common cases, as the post on [AI without integration becoming a FAQ](https://tolky.to/en/blog/ai-without-integration-becomes-faq) explains.
5. **Testing.** Simulate real conversations, including the hard ones, before publishing. On platforms with simulations and versions, as the [trustworthy AI](https://tolky.to/en/trustworthy-ai) page shows, each change is tested before it reaches the end customer.
6. **Go-live and monitoring.** Publish with clear hours and rules, read the first week of conversations and adjust the knowledge base. This is where monthly operations begin.

What defines good delivery is not the first version of the agent but the review routine after go-live. The [AI resolution rate](https://tolky.to/en/blog/ai-resolution-rate-how-to-measure) is the metric that shows the client whether the agent improves month after month.

## Should you build on custom code, n8n or a platform?

**An AI automation agency can build agents in custom code, in an automation tool such as n8n or on a ready-made platform, and the choice decides how much of the team's time goes to technical maintenance.** There is no single answer: it depends on who is on your team and how many clients you want to serve.

| Criterion | Custom code | n8n or automation tool | Ready-made platform |
| :--- | :--- | :--- | :--- |
| Time to first client | Long | Medium | Short |
| Who maintains it | The agency's developers | Whoever built the workflow | The vendor maintains the base; the agency configures |
| Inbox for the client's team | Must be built | Needs another tool | Included |
| CRM and contact history | Must be built or integrated | Must be integrated | Included in conversational CRM platforms |
| Testing before publishing | Must be built | Manual | Depends on the platform; check for simulations and versions |
| Scaling to many clients | Depends on the architecture | Each client becomes a set of workflows | One account per client, with central access |

n8n is great for proving the concept and connecting systems, and the guide to a [WhatsApp AI agent with n8n](https://tolky.to/en/blog/whatsapp-ai-agent-n8n) shows how to build one. The cost shows up as the client base grows: each client becomes a set of workflows only its builder understands, and the client's team still needs an inbox to take over conversations. The full comparison is in [n8n or a platform for agencies](https://tolky.to/en/blog/n8n-or-platform-for-agencies).

In practice, many agencies combine both: a platform for the agent, inbox and CRM, and automations to connect systems the platform does not cover. Before choosing, check the available [integrations](https://tolky.to/en/integrations).

## How much does it cost to start an AI automation agency?

**Starting an AI automation agency costs mostly team time to learn and build the first projects; the tool can be cheap or even free at first, and Meta's per-message fee is usually paid by the client.** What weighs most is the period when the team works before the first contract.

The components of startup cost:

- **Team training.** At least one person who understands agent design, knowledge bases and WhatsApp rules. Platform partner programs often include training; in Tolky's program it is free for member companies.
- **A practice account.** Many platforms have a free plan. At Tolky, the Individual plan is free and includes webchat and AI; WhatsApp starts on the Starter plan, listed on [/en/precos](https://tolky.to/en/pricing) at US$ 90 per month on the date of this review.
- **Pre-sales time.** Discovery and demos for the first clients, before any revenue.
- **Client costs, not agency costs.** Platform, Meta's per-message fee and the number connected to the official API usually sit on the client's account. The guide to [WhatsApp AI agent pricing](https://tolky.to/en/blog/whatsapp-ai-agent-pricing) breaks down each component.

Anyone who builds everything in custom code also adds hosting, the AI model and development hours before the first contract.

## How should you price AI projects?

**The most common approach is to separate implementation, charged once, from operations, charged monthly, and to leave the platform and Meta's fees as client costs, priced separately.** Bundling everything into one fixed price hides from the client what they pay for and squeezes the agency's margin as volume grows.

Three billing lines:

- **Discovery:** paid or credited against implementation, with a clear deliverable (scope, metrics, plan).
- **Implementation:** a per-project fee, proportional to the number of agents, integrations and channels.
- **Monthly operations:** conversation review, knowledge base updates, reports and new automations. This is the retainer that keeps the agency inside the client's operation.

Amounts depend on the market and the client's size. The guide on [how much to charge for a WhatsApp AI agent](https://tolky.to/en/blog/how-much-to-charge-for-whatsapp-ai-agent) shows how to price each line, and the post on the [agency retainer with AI](https://tolky.to/en/blog/agency-retainer-with-ai) shows how to build the recurring package.

## Which niches buy AI agents the most?

**The businesses that buy most have high WhatsApp conversation volume, expensive leads and service needs outside business hours.** Those three signals, more than the industry itself, show where an agent pays off fastest.

Industries where these signals often appear together:

- **Healthcare:** scheduling, confirmations and repeated questions, with extra care for sensitive data. See the [healthcare](https://tolky.to/en/industries/healthcare) segment.
- **Education:** student recruitment with lead spikes during enrollment. See the [education](https://tolky.to/en/industries/education) segment.
- **Real estate:** portal and ad leads that need a fast reply. See the [real estate](https://tolky.to/en/industries/real-estate) segment.
- **Vehicle protection and insurance:** quotes, sign-ups and assistance over WhatsApp. See [vehicle protection](https://tolky.to/en/industries/vehicle-protection).
- **Law firms and professional services:** case triage and scheduling. See [law firms](https://tolky.to/en/industries/legal).

For a marketing agency, the best niche is usually the one it already serves with media. It knows the funnel, knows the cost per lead and can show, with its own campaign data, how much the client loses without a fast reply.

## What does this change for your agency?

**For an agency, becoming an AI agency means no longer delivering just the lead, but also the conversation that turns the lead into an opportunity, with recurring revenue from operations.** The change is in scope, not identity: the agency keeps selling results, now with more control over them.

What changes in practice:

- **A new service for your current clients.** Start with clients who already complain about leads that never answer or never buy. They already feel the problem.
- **An operations retainer, not just a media fee.** Running the agent every month creates a revenue line that does not depend on the ad budget.
- **Reports that reach the sale.** With the lead logged in the CRM from the first message, the agency shows what happened after the click, not just cost per lead.
- **Retention.** An agent running every day inside the client's operation builds a longer relationship than a campaign.
- **New roles for the team.** People who handle client service or account management today can take over agent operations, with training.

To sell the first project, the guide on [how to sell AI agents to businesses](https://tolky.to/en/blog/how-to-sell-ai-agents-to-businesses) covers discovery, demo and proposal.

## What are the risks of running an AI agency?

**The three risks that sink AI agency projects most often are using an unofficial WhatsApp API, ignoring Meta's AI rules and handling client data without a data processing agreement.** All three have known fixes, and all three land on the agency when they go wrong.

**Unofficial WhatsApp API.** Tools that connect through WhatsApp Web, without the official API, are cheap and quick to set up, but the number can be banned, and the one who loses the number is your agency's client. The post on the [unofficial WhatsApp API](https://tolky.to/en/blog/unofficial-whatsapp-api-risks) explains the risks and the path to the official API.

**Meta's AI rules.** Since January 15, 2026, the [WhatsApp Business Platform terms](https://www.facebook.com/legal/Meta-Terms-for-WhatsApp-Business-Platform) ban AI providers whose main function is the AI itself, such as general-purpose assistants. Agents that serve a company's customers remain allowed. In Brazil, the antitrust authority CADE suspended the rule in January 2026. What applies to your project is in [does WhatsApp ban general-purpose AI chatbots?](https://tolky.to/en/blog/whatsapp-bans-general-purpose-ai-chatbots). In practice: sell agents with a defined mission for the client's business, not "ChatGPT on WhatsApp".

**Data protection and each party's role.** Under Brazil's data protection law, [Law No. 13,709/2018 (LGPD)](https://www.planalto.gov.br/ccivil_03/_ato2015-2018/2018/lei/l13709.htm), checked on November 20, 2026, the controller makes the decisions about data processing and the processor handles data on the controller's behalf, following its instructions (articles 5 and 39). In an agent project, the client is usually the controller and the agency acts as processor. Other data protection laws use similar roles, so put this in a data processing agreement and see the [LGPD](https://tolky.to/en/glossary/lgpd-in-customer-service) glossary entry and the [security](https://tolky.to/en/security) page of the platform you use.

A fourth risk is commercial: promising full automation. An AI agent resolves part of the conversations and hands the rest to people. A proposal that promises 100% becomes an unhappy client in the first month.

## How does a marketing agency become an AI agency?

**A marketing agency becomes an AI agency when it adds to what it already does (media, content, relationships) the delivery and operation of agents that answer the leads its media generates.** It already has the clients, their trust and an understanding of the funnel; what is missing is the technical method and a first case.

The safest path is gradual: train one or two people, pick a client already struggling with unanswered leads, deliver a lead qualification agent and measure. Then turn the case into an offer for the rest of your clients. How to make that transition without alarming the clients who pay your retainer today is covered in [marketing agency with AI](https://tolky.to/en/blog/marketing-agency-with-ai). The [AI for marketing](https://tolky.to/en/departments/marketing) page shows the client's side.

## Where does Tolky fit in an AI agency?

**Tolky is a conversational AI CRM: AI agents, a built-in CRM and WhatsApp campaigns through the official API, in one platform, which an agency can use to deliver the four services in this guide.** The Partner Program gives the agency training, certification and support on initial projects.

![Real screenshot of Tolky's Conversations screen, with period filters on the left and the list of WhatsApp conversations in the center, each showing the contact name, the time and a summary of the request](https://tolky.to/screenshots/conversas-whatsapp-en.png)

What the platform gives the agency to deliver:

- **Agents configured by mission** (support, sales, engagement), with handoff to people with context.
- **A built-in CRM** with contacts, opportunities, pipeline, history and automatic Follow Up.
- **Simulations and versions** before publishing, to test every change.
- **Channels:** WhatsApp through the official API, webchat, email, Microsoft Teams and voice, which is under activation. Instagram and SMS are coming soon.
- **Integrations:** native HubSpot, 1,000+ apps in the Integration node of Flows, REST API, webhooks and MCP tools.

**What Tolky does not do:** it does not automatically link the ad of origin to the sale of a WhatsApp lead (that takes a client process or a custom integration, to be confirmed in a demo), and it has no native integration with Salesforce or Pipefy.

## Frequently asked questions

### What does an AI automation agency do?

An AI automation agency designs, deploys and runs artificial intelligence agents for businesses, mainly in customer service and sales over WhatsApp. It charges for the project and, in most cases, for monthly operations that keep the agent up to date.

### How does an AI agency work?

It follows a cycle of discovery, agent design, knowledge base, integrations, testing and go-live. After go-live, the agency reviews conversations, adjusts the knowledge base and reports metrics every month.

### Is an AI automation agency profitable?

It depends on the pricing model and how much of the team's time goes to technical maintenance. Agencies that charge for monthly operations and use tools that reduce repetitive work tend to have more predictable revenue. Be wary of any revenue figure promised without context.

### Do I need to know how to code to start an AI agency?

Not necessarily. On ready-made platforms, the agent is configured through its mission, knowledge base and rules, without code. Knowing automation logic and APIs helps with more complex integrations.

### What is the difference between an AI agency and an automation agency?

An AI agency usually focuses on agents that talk to customers. An automation agency also covers workflows between systems, such as moving form data into the CRM. In practice, the terms overlap and many agencies offer both.

### Is it allowed to use AI on WhatsApp to serve customers?

Yes. Since January 15, 2026, Meta's terms ban general-purpose AI assistants, but agents that serve a company's customers remain allowed. Always use the official WhatsApp API.

### Who pays the WhatsApp fees in an AI agency project?

Usually the client, because the WhatsApp Business account and the number belong to them. The agency should make this explicit in the proposal, separate from implementation and operations.

### Can an AI agency use n8n?

Yes, and many start with it. The point to watch is maintenance: each client becomes a set of workflows, and the client's team still needs an inbox and a CRM to take over conversations.

### Can an AI agency work under its own brand?

Yes, if the platform offers white label. At Tolky, white label builds on the "Your brand and domain" feature of the Enterprise plan, and the topic is covered in [white label AI agent](https://tolky.to/en/blog/white-label-ai-agent-for-agencies).

---

## Conclusion

An AI automation agency sells discovery, implementation and operation of agents that serve, qualify and reactivate customers, mostly on WhatsApp. What separates a solid agency from a course promise is a delivery method, recurring billing for operations, the official API, respect for Meta's rules and a data agreement with the client. Start with a client who already feels the problem and turn that first case into an offer.

## Your agency can deliver this with Tolky

**The Tolky Partner Program is for agencies that want to bring AI agents, CRM and WhatsApp campaigns to their own clients, with recurring earnings.** On the Solution track, your agency sells, implements and supports clients under the Tolky brand and charges for its own implementation and operations services; the other tracks are referring Tolky or running the platform under your own brand. Training and certification are free for program members, and the Tolky team supports every initial project in 1:1 sessions.

[Explore the Partner Program →](https://tolky.to/en/partners) Let's talk?

---

This is the Markdown version of the page. The full version, with images and navigation, is at https://tolky.to/en/blog/ai-automation-agency-what-it-is-how-to-start

Tolky: https://tolky.to
