The cost of a custom AI agent depends mostly on scope, how many systems it must connect to, the state of your data, the channels it runs on, how much human handoff you need, and ongoing maintenance. Two quotes can differ widely because they assume different answers to those six questions.
Key takeaways: Price follows complexity, not the word "AI", so compare quotes by what each one assumes about the six drivers below. A clear brief is the cheapest way to get a realistic number.
Why do quotes for the "same" AI agent differ so much?
They differ because the agencies are rarely pricing the same thing. One quote may cover a chatbot that answers ten common questions on your website, while another covers an agent that reads your CRM, books appointments, escalates to staff and reports on its own results. Both get called "an AI agent", yet the work behind them is not comparable. Add different assumptions about who prepares your content, who maintains the system afterwards and how much testing is included, and a gap between quotes is normal. A very low number is not automatically a bargain, and a high one is not automatically padding. Your job is to find out what each number includes.
What are the main factors that drive the cost?
Six factors explain most of the variation, and each one changes how much design, building, testing and upkeep the project needs. The table below summarises them, and the sections after it go through each in more detail. None of them is exotic, which is good news: you can answer most of them yourself before the first call.
| Driver | Lower effort | Higher effort |
|---|---|---|
| Scope | One task, one audience | Several tasks, exceptions, multiple audiences |
| Integrations | Standalone or one tool | CRM, calendar, billing, store, internal tools |
| Data quality | Clean FAQs and policies | Scattered, outdated or contradictory documents |
| Channels | Website widget only | Website, WhatsApp, Instagram, email |
| Human handoff | Simple "contact us" fallback | Routing to staff with full context and rules |
| Maintenance | Occasional updates | Ongoing tuning, monitoring and reporting |
How does scope change the price?
Scope is the biggest driver because every extra task multiplies the cases the agent must handle correctly. "Answer opening-hours questions" is a small job. "Qualify leads, book a visit, reschedule it and follow up if they go quiet" is several jobs that interact, each with exceptions such as a customer who cancels twice or asks something off-topic. More behaviours mean more design, more test conversations and more ways to fail. Starting with one well-defined task is usually cheaper and safer, and our guide on what to automate first in a small business helps you pick it.
Why do integrations matter so much?
An agent that only talks is cheap compared with one that acts inside your systems. Every connection to a CRM, calendar, store or invoicing tool needs authentication, field mapping, error handling and testing against your real setup. Tools with good, documented APIs are quicker; older or heavily customised systems take longer and sometimes need workarounds. Ask the agency to list each system it will touch and what it will read or write in each. If you use HubSpot, for instance, the details in connecting a chatbot to your CRM show what that work involves.
How does the quality of my data affect the cost?
Messy source material turns into hidden work, because the agent can only be as accurate as the content it draws on. If your pricing is in three documents that disagree, or your policies live in someone's head, somebody has to sort that out before launch. That time is either billed by the agency or lands on your team, and a quote that ignores it is optimistic. Cleaning up a short, current set of FAQs and policies often saves more than any technical shortcut. The process is covered in how to train an AI chatbot on your own business data.
Do channels and human handoff change the price?
Yes, each channel has its own rules, message formats and setup, and handoff adds logic you have to design carefully. WhatsApp, for example, runs through Meta's platform with its own approval steps and conversation-based charges, which Meta documents in its WhatsApp Business Platform pricing documentation. Those platform fees are separate from the agency's fee, so check whether a quote includes them. Handoff is similar: passing a conversation to a person with the history attached, at the right moment, takes real design work. See what an agency sets up for the WhatsApp Business API and when to hand off to a human.
What about maintenance after launch?
An agent is not finished at launch, and a quote that treats it as finished is incomplete. Your products, prices and policies change, models and platforms get updated, and real customers will ask things nobody planned for. Someone needs to review conversations, fix gaps and keep integrations working. Usage costs also exist: model providers charge by usage, as OpenAI's API pricing page shows, so a busier agent costs more to run. Ask what is included after launch, for how long, and who pays for usage.
How do I compare two quotes fairly?
Put both quotes into the same checklist and mark every assumption that differs. This turns a vague "one is cheaper" into a concrete list of what you would be giving up or taking on. Use these questions on each agency:
- Which exact tasks does the agent handle, and which are explicitly out of scope?
- Which systems does it connect to, and is reading or writing data included?
- Who prepares and cleans the content the agent relies on?
- Which channels are covered, and are platform and usage fees included or billed separately?
- How does handoff to a person work, and what context travels with it?
- What testing happens before launch, and what support or tuning follows?
- What do you own at the end: accounts, prompts, workflows, documentation?
The answers often explain the price difference completely. If an agency cannot answer them, that is useful information too. For a broader view, how to choose an AI agent agency and freelancer or agency cover the choice of who to hire.
How can I get a realistic quote?
Arrive with a short brief, because the quality of a quote follows the quality of the information behind it. Describe the one problem you want solved, the volume of conversations or tasks, the tools involved, where your content lives and what a good outcome looks like. Agencies that receive this can estimate instead of padding for unknowns, and you spot misunderstandings early. The template in how to write an AI project brief gives you the structure. It also helps to understand the stages a project goes through, which what a custom AI agent project involves lays out.
Is the cost worth it for my business?
That depends on what the task costs you today, so estimate it before you ask for any quote. Count the hours your team spends on the task, the leads or sales lost to slow replies, and the errors that cost you money. Compare that with the build and running costs once you have a real quote, and be conservative with the savings. The worksheet in calculating process automation ROI before you build walks through it, and custom agent vs. no-code tools vs. SaaS may show that a cheaper option is enough for your case.
Frequently asked questions
Can I get a fixed price for a custom AI agent?
Often yes, once the scope, integrations and channels are clearly defined. Open-ended requests usually get estimates or phased pricing instead. A written brief makes a fixed quote much more likely.
Are there running costs after the agent is built?
Usually. Model usage is billed by providers such as OpenAI, messaging platforms may charge per conversation, and someone needs to maintain the agent. Ask which of these the quote includes.
Is a cheaper quote a red flag?
Not by itself. It may cover a smaller scope or exclude testing, maintenance or platform fees. Compare what each quote assumes before judging the number.
Should I start small or build everything at once?
Start with one well-defined task, measure the results, then expand. It limits your risk and gives you real data for the next quote.
What can I do to lower the cost of my project?
Narrow the first scope, tidy your FAQs and policies, and list the tools involved before the first call. Preparing the content yourself removes work the agency would otherwise price in.
At Kuan AI we work through diagnosis, design, implementation and then measuring and scaling what works, and pricing is agreed per project once the scope is clear. If you want to see how these drivers apply to your case, send us a short description through the contact form. We reply within 24 hours.
