AI chatbot cost: what you pay for SaaS, per-resolution and custom builds

What are the three ways to pay for an AI chatbot?

Every AI chatbot price you will see falls into one of three models, and the cheapest one changes with your volume. Comparing a $49 plan with a $40,000 build makes no sense until you know which model each quote uses and how it scales.

Seat-based SaaS charges per support agent per month and usually bundles some AI usage. Per-resolution pricing charges a fee each time the bot closes a conversation without a person. A custom build charges once for the engineering and then bills you for usage directly from the model provider, the vector database and the host. Many vendors mix the first two, so a single helpdesk invoice can carry both seat fees and resolution fees.

• Seat-based SaaS: predictable, tied to headcount, AI usage capped or sold as an add-on.
• Per-resolution: you pay only when the bot succeeds, but the bill grows with every resolved conversation.
• Custom build plus usage: a large first payment, then a monthly bill that grows slowly with volume, on accounts you own.

What does AI chatbot pricing look like on SaaS platforms?

Published SaaS AI chatbot pricing starts below $50 a month and rises with seats, conversations and AI resolutions. The figures below come from each vendor's own pricing page as we read them on October 16, 2026. Vendors change these often, so check the page before you sign.

Two details matter more than the headline price. First, what counts as billable: Intercom bills one Fin outcome per conversation, counted when the customer confirms the answer, does not ask for more help, or Fin completes a workflow. Second, what is included: Zendesk's Suite Team plan includes 5 automated resolutions per agent per month and Suite Professional includes 10, after which each one is charged.

How much does it cost to build a custom AI chatbot?

Custom AI chatbot development cost runs from about $5,000 for a scripted FAQ bot to $300,000 or more for a generative AI system wired into several business systems. Monocubed's 2026 cost guide gives the bands in the table below, with timelines. Softude's 2026 guide puts specialist AI agency projects in general at $25,000 to $150,000, which overlaps the middle bands.

The bands are wide because the label "chatbot" covers very different builds. A bot that answers from 200 help articles on a website is a small project. A bot that answers from SharePoint with per-user permissions, looks up orders in Shopify, opens tickets in Zendesk and logs everything for audit is a much bigger one, even though both look like the same chat window to the user.

What drives the price of a custom chatbot build?

Four things move a custom quote more than anything else: how many knowledge sources you connect, whether answers must respect user permissions, how many channels the bot runs in, and whether it only answers or also takes actions. Model choice matters far less, because the model call is a small part of the code.

Ask each vendor to price these items separately. It makes quotes comparable, and it shows you where to cut scope if the number is too high. Dropping the second channel or deferring write actions to a later phase is a real saving; asking for a general discount is not.

• Sources: each of SharePoint, Google Drive, a Zendesk help center, Confluence or a PDF library needs its own connector, sync schedule and handling of deleted files.
• Permissions: an internal bot that must only answer from files the user can open needs access rules carried into the search index, which is real engineering.
• Channels: a website widget, a Slack app, a Teams app and a Zendesk sidebar each have their own login, formatting and handoff rules.
• Actions: looking up an order in Shopify or creating a HubSpot ticket turns a chatbot into an agent, adding API work, confirmation steps and error handling. Monocubed's guide prices CRM integrations at $5,000 to $25,000 each.
• Accuracy bar: a bot that drafts replies for an agent to edit needs less testing than one that answers customers directly about refunds or eligibility.
• Compliance: protected health information requires vendors that sign a Business Associate Agreement, which narrows model and hosting choices.

What does a custom AI chatbot cost to run each month?

After launch, a custom chatbot costs a usage bill plus upkeep, and most of it is billed straight to accounts you own. For a bot that answers from company documents, the bill has four lines: model tokens, a vector database for retrieval, hosting and logging, and maintenance time.

Monocubed's 2026 guide estimates model API tokens at $0.005 to $0.075 per conversation, a vector database at $50 to $500 or more a month, cloud hosting at $200 to $1,000 a month, and annual maintenance at 15 to 20 percent of the build cost. Token cost per conversation is the line you can shrink most: a smaller model for routing and simple questions, retrieval that sends only the few passages needed rather than whole documents, and caching for repeated questions all cut it.

Maintenance is the line buyers forget. Documents change, model vendors retire versions, new question types appear and the eval set needs updating. Budget for it from the start, or the bot's answers drift out of date within months.

When does a custom chatbot become cheaper than per-resolution pricing?

A custom chatbot beats per-resolution pricing once your monthly resolution fees are larger than the custom bot's running bill by enough to pay back the build in a period you accept. Below is a worked example using the published figures above. The $40,000 build price is an assumption chosen from inside Softude's published agency range so the arithmetic is visible; it is not a quote.

Assume 5,000 conversations a month, of which 3,000 are resolved by the bot. Intercom Fin at $0.99 per outcome costs about $2,970 a month; Zendesk pay-as-you-go automated resolutions at $2.00 cost about $6,000. The custom bot pays tokens on all 5,000 conversations ($25 to $375 using Monocubed's per-conversation range), plus $50 to $500 for the vector database, $200 to $1,000 for hosting and $500 to $667 a month for maintenance (15 to 20 percent of $40,000 a year). That gives a custom running bill of roughly $775 to $2,542 a month.

Against Intercom, the custom bot saves about $428 to $2,195 a month, so the $40,000 build pays back in roughly 18 months at best and nearly eight years at worst. Against Zendesk pay-as-you-go, it saves about $3,458 to $5,225 a month and pays back in roughly 8 to 12 months. At 10,000 resolutions a month from about 16,700 conversations, Intercom fees reach about $9,900 a month while the custom bill rises only to about $833 to $3,420, and payback against Intercom falls to roughly 4 to 6 months.

The lesson is that volume decides. At a few hundred resolutions a month, a SaaS plan is almost always the cheaper choice. At several thousand a month and growing, the per-resolution line becomes the biggest number on the support budget, and a build you own starts to make financial sense.

Which option fits your situation?

Pick the pricing model from your volume, your data and what the bot must do, not from the lowest headline price. The table below is the short version of the decision we walk through in a free review.

When should you not pay for a custom AI chatbot?

Skip a custom build when your volume is low, your content is public and your helpdesk vendor already offers an AI agent that answers well on a trial. Paying $40,000 to save a few hundred dollars a month is a poor trade, and a packaged tool can be switched off with little loss.

Also hold off when the knowledge itself is the problem. If help articles are out of date or policies contradict each other, any chatbot, packaged or custom, will repeat those errors with confidence. Fixing the top 50 documents people ask about is cheaper than any bot and makes whichever one you choose better. Our guide to why AI projects fail covers this and the other common reasons builds stall.

How do we price an AI chatbot build?

We quote a fixed price after a free review, not an hourly estimate. The review looks at your sources, the questions people actually ask, the channels you need, the permission rules and any actions the bot should take, and the quote lists what is in scope and what is not.

You own the model, hosting and vector database accounts, the code, the prompts and the index, so the running bill comes to you directly from each vendor and you can move the system if you ever leave us. A person stays in the loop on anything touching money, customers or regulated data, and low-confidence answers go to a handoff rather than a guess. Our AI chatbot for business page explains how we build document-grounded bots with citations, and the AI implementation cost guide covers build and running costs across other kinds of AI projects.