AI agents for small business: what they cost and what they actually do

What an AI agent is (and isn't)

An AI agent is software that answers questions or takes actions using your own content and systems — usually combining a large language model with retrieval (RAG) and API integrations. It is not a magic employee: it is a narrow tool that is reliable only within the data and rules you give it.

Use cases that work for small teams

• Answering repetitive support questions from your docs
• Qualifying and routing inbound leads
• Internal Q&A over policies, contracts or product data
• Automating a defined back-office workflow (reports, tagging, data entry)

What it costs in 2026

• Off-the-shelf SaaS chatbot: $29–$99/month
• Custom FAQ / support bot: $1,500–$5,000 one-time
• Retrieval (RAG) Q&A over your documents: $10,000–$30,000
• Agent wired into live systems (CRM, ERP): $20,000–$80,000+
• Ongoing model / API usage: usually $50–$2,000+/month, separate from the build

Where AI agents fail

• Bad or missing source data — the model can only be as good as your docs
• No owner: nobody refreshes content or reviews answers after launch
• No evaluation: nobody measures whether answers are actually correct
• Over-promising: treating a pilot as a finished product

Build or buy?

If a $29/month tool already does what you need, buy it. Build when the agent must use your private data, follow your rules, or write back into your systems — that is where off-the-shelf tools stop and a small custom project starts.

A sane first step

Start with one narrow, measurable workflow over a defined document set. Ship it, watch the answers, then expand only what proves useful. Most failures come from starting too broad.