Key takeaways
- An AI chatbot earns its place in a service business when it does something with the conversation: it qualifies the visitor, finds the right match in your own data, and puts a ready record into your CRM.
- A RAG chatbot answers from the business’s own documents and records, which is what lets it talk about your services and your criteria instead of the internet in general.
- At Cirrus Capital, a debt capital advisor, the website chatbot asks the qualifying questions, filters the investor database while the chat runs, and shows a ranked shortlist of matched investors before the conversation ends.
- An off-the-shelf chatbot tool is usually enough when the job is answering common questions. A built one starts to pay when the bot has to read your records and write back into them.
Search for an AI chatbot for small business and page one is chatbot vendors, plus lists of the best chatbot vendors written by those same vendors. Any of them will put a widget on your site by the afternoon. What none of them show is what happens to the person after the chat: whether they end up as a qualified record your team can act on, or as one more transcript nobody reads.
We build operations systems for service businesses, and chatbots are one of the AI features we run in production for clients. This guide uses one of them, at Cirrus Capital, to show what a chatbot looks like when it is wired into the rest of the business. We sell builds, so weigh that, and where a tool is the better answer we say so.
What does an AI chatbot do for a service business?
For a service business, an AI chatbot does three jobs: it answers questions about your services, it qualifies the person asking, and it hands a complete record to your team. The first job is what most chatbot tools sell. The other two are where the hours come back, because they replace the intake call and the data entry after it.
| Job | What the chatbot does | What it replaces |
|---|---|---|
| Answer | Responds from your own documents: services, process, policies | The same questions answered by hand, again and again |
| Qualify | Asks the questions your team would ask on a first call and checks the answers against your criteria | A first call with someone who was never a fit |
| Hand off | Creates the record in your CRM with everything it learned, and starts the next step | Copying notes from a transcript into the CRM |
Most of the value sits in the last two rows, and both depend on the chatbot being able to reach your data.
How does a RAG chatbot work?
A RAG chatbot looks things up before it answers. When a question comes in, it retrieves the most relevant pieces from documents or records you control, then hands those pieces to a language model, which writes the answer from them. RAG stands for retrieval-augmented generation, and the retrieval step is what keeps the answer tied to your business.
The method was described in a 2020 research paper from Facebook AI, and IBM Research’s explainer is a good plain-language version. For a business owner the practical point is simpler. The bot answers from your service descriptions, your criteria and your records, so it can say what you actually do, and you keep it current by updating those sources rather than retraining anything.
It can still be wrong. Its answers are only as good as the sources it retrieves from, which is why the Cirrus Capital team still validates the top matches with its own judgment.
What does a lead-qualifying chatbot look like in practice?
At Cirrus Capital, a Florida debt capital advisor, a company that arrives on the website talks to a chatbot instead of filling in a contact form. The bot asks what the business does, its stage, what kind of capital it needs, how much and when, and it filters the investor database while the conversation is still running.
- The conversation replaces the contact form, and the bot asks the questions an analyst would ask at intake.
- As the answers come in, it filters investors against their mandates: sector, ticket size, geography, structure and stage.
- Before the chat ends, a ranked shortlist of matched investors is already visible inside it.
- If the company wants to meet them, it is added to an introduction email sequence, and the opportunity enters the CRM already qualified and matched.
- The Cirrus team validates the top matches with its own judgment.
We built the chatbot on Botpress. Underneath it sits a custom CRM with separate databases for companies, investors and partners, and a matching algorithm that reruns whenever either side changes. Before the bot, the website produced traffic that left no trace: companies arrived, browsed and left, and Cirrus had no record of who they were.
“Our work with Catalytics will enable us to reach our 2024 and 2025 revenue and growth targets with 30–40% less headcount than anticipated.” Ryan Ridgway, CEO, Cirrus Capital
That figure covers the whole system, the CRM, the matching and the automated sequences, and not the chatbot alone. The full build is in the Cirrus Capital case study.
Should you buy a chatbot tool or have one built?
Buy a chatbot tool when the job is answering common questions about your business, because most tools do that well for a monthly fee. Have one built when the bot needs to read your own records, qualify against your criteria and write the result back into your CRM, because that work lives in your data rather than in the widget.
A tool is usually enough when:
- the questions are the same for everyone, like services, hours, process and policies
- a person qualifies every lead anyway, and the bot only needs to collect contact details
- nothing the bot says depends on records that change during the day
A build starts to make sense when:
- the bot has to check answers against criteria that live in your own data, the way Cirrus checks investor mandates
- the result has to land in your CRM as a complete, usable record and start the next step on its own
- the answer depends on live records, such as a match, an availability or the status of an account
For a lot of small businesses the first list describes them, and a tool is the right call.
What about FAQ chatbots?
An FAQ chatbot is the narrower version: it answers the questions your team gets every day from the business’s own documents, and passes anything else to a person. FAQ chatbots are one of the AI features we run in production for clients, and they are also the case where an off-the-shelf tool most often does the job on its own.
What does a chatbot need before it goes live?
It needs records worth answering from and a clean place to write to. A chatbot that reads stale documents gives stale answers, and one that qualifies leads into a messy CRM creates more cleanup than it saves. At Cirrus Capital, the companies, investors and partners databases were built before the chatbot was.
It is the order we follow for AI work in general: the data layer first, then the automations, then the AI on top. The AI automation page lists the other AI features we run for clients and how they fit together.
Frequently asked questions
How can I create an AI chatbot?
Chatbot platforms such as Botpress let you build one without code, connect it to your documents and publish it on your site. The other route is having one built into your own systems, which fits when it has to qualify against your criteria and write into your CRM. Either way, start by listing the questions it should answer.
Is ChatGPT a RAG?
Not by default. ChatGPT answers from what its model learned in training, and it works in a retrieval-augmented way when it is given documents or search results to draw from, for example when it searches the web or reads files you upload. A RAG chatbot built for a business always retrieves from that business’s own sources before it answers.
What is RAG vs AI agent?
RAG is a way of answering: retrieve the relevant material, then generate the reply from it. An AI agent is a way of acting: it takes steps toward a goal, such as updating a record or sending an email. The Cirrus Capital bot retrieves to answer, and it acts when it filters investors and starts the introduction sequence.
Can an AI chatbot update our CRM on its own?
Yes, if it is connected to it. The chatbot can create or update the CRM record through the CRM’s API, or through an automation platform such as Make or n8n, so a qualified lead arrives with everything the conversation collected. Keep a person on the decisions that matter, the way the Cirrus team still validates its top matches.