How Much Does It Cost to Build an AI Chatbot? Pricing Guide

How much does it cost to build an AI chatbot? A hosted chatbot can range from free to hundreds of dollars a month, while published custom-build estimates range from about $12,000 to more than $260,000 upfront. You can also start with a free hosted plan if its limits fit your needs. Those figures describe different scopes, not a reliable industry average. A useful budget also includes recurring model, hosting, maintenance, setup, seat and usage costs.

AI chatbot costs: custom build versus hosted platform

A hosted platform usually charges a recurring subscription, while a custom chatbot can require a substantial upfront development budget plus continuing operating costs. Published estimates place self-serve plans at roughly $20 to $500 per month, custom LLM chatbot builds at $12,000 to $70,000, and complex enterprise builds at $90,000 to $260,000 or more.

These estimates come from different providers and methods, so treat them as planning ranges rather than quotes. One estimate of custom development uses a specific team composition and regional rate card; another gives a broader range based on project scope. A separate published estimate places custom chatbot development between $15,000 and $220,000 or more. The variation is a reminder to compare what each estimate includes, not just the headline total.

Approach	Published cost indication	What the price generally represents
Hosted self-serve platform	About $20–$500 per month	Subscription access, usually with defined usage and feature limits
Custom LLM chatbot	About $12,000–$70,000 to build	A project that may include retrieval over a business knowledge base
Enterprise multi-agent system	About $90,000–$260,000 or more to build	A larger, more complex project with deeper integrations
Custom contract	Annual price set by vendor	A negotiated scope based on factors such as volume, channels and integrations

A subscription is not automatically “all in.” The included allowance may be limited, and seats, overages, add-ons or staff time can add to the bill. A custom build shifts more responsibility to your team: you need to arrange and pay for development, hosting, monitoring, testing and knowledge upkeep.

The right comparison depends on what you need the bot to do. If it answers common questions from approved content and hands unresolved conversations to your team, a hosted support platform may cover the workflow without a bespoke build. If it must operate inside internal systems or follow requirements that a platform cannot meet, a custom project may be worth evaluating. That does not make one approach universally cheaper: scope and operating responsibilities differ.

When reviewing a custom estimate, ask what “build” means. Does the scope cover discovery, retrieval from your documents, testing, a support handoff, deployment and training for the people maintaining it? If not, those tasks may sit outside the quoted figure. Ask the provider to separate development from ongoing service costs so the upfront number does not hide the cost of running the chatbot.

Compare pricing models by what triggers the bill

The pricing model determines what makes the bill rise: more agents, more conversations, more AI replies, more resolved outcomes or a larger negotiated scope. For the same stated workload of 2,000 conversations a month, one published comparison gives a flat-plan example at $129 monthly and an outcome-based example at $1,077 monthly with three seats.

Pricing model	What triggers the bill	What to inspect
Per seat	Number of human agents with access	Which AI features are included and whether AI usage is billed separately
Flat plan	Subscription tier and included allowance	Conversation, reply, agent and feature limits
Per resolution or outcome	Number of issues the AI resolves	Definition of a billable outcome and any included allowance
Per conversation	Number of conversations, resolved or not	What counts as a conversation and the cost of additional blocks
Credits or usage	Replies, tokens or another metered unit	How units are counted and how quickly they are used
Custom contract	Negotiated volume, channels and integrations	Commitment, scope, renewal terms and usage boundaries

A per-seat model is easiest to forecast when your team size is stable, but hiring can raise the total even if chatbot volume stays level. A flat plan makes the subscription more predictable within its allowance. However, you still need to know what happens when usage exceeds that allowance, or when you need additional channels, seats or features.

Outcome-based pricing ties the bill to successful AI resolutions. That may sound attractive because payment follows a result, but the bill can increase as more cases are resolved. Check how the provider defines a resolution, how it treats conversations that need a human, and whether a usage cap is available.

Conversation pricing charges for conversations rather than only successful resolutions. That distinction matters if the bot often asks a question and then hands the case to a person: those conversations may still count. Credit-based pricing needs a similar check. If a credit is used for each AI reply, a long exchange can consume more credits than a short one.

The $129 and $1,077 examples are not a like-for-like guarantee for every service. They are examples from one published comparison of different pricing models. The second includes three seats and outcome pricing; the first is a flat-plan example. Use the examples to see why the bill driver matters, then calculate your own workload using each provider’s actual limits and terms.

What pushes a custom chatbot quote higher

A custom quote tends to rise when the chatbot needs more than a narrow question-and-answer flow: complex behavior, several integrations or channels, sensitive-data controls, extensive knowledge preparation, or a carefully designed route to human support. Each addition brings engineering, testing and documentation work, though the cost depends on the project and the provider’s assumptions.

Ask for a written scope that names the work. For example, “connect to our CRM” is not specific enough to budget reliably. Identify which data the chatbot will read or change, how access is controlled, what happens when the connection fails, and who tests the integration. Custom connections to business systems may require development and ongoing troubleshooting.

Channels also affect scope. A chatbot for a website is a different project from one that needs to work across multiple channels with consistent context and handoff. Ask which channels the quote covers and whether channel-specific behavior, testing and maintenance are included.

Knowledge preparation is another practical cost, even when there is no separate line item. Someone needs to decide which documents are current, resolve conflicts between policies, and make sure the chatbot can retrieve useful passages. If your return rules are scattered across old PDFs, help articles and internal notes, resolving those contradictions is work the project must account for. No generic quote can tell you how much effort your particular knowledge base will need.

Human handoff should be specified as part of the workflow rather than treated as an afterthought. Decide what details the visitor should provide, what the team sees when a case arrives, and how the chatbot behaves when it cannot answer. Test common edge cases: a refund request that needs review, a missing order detail, or a customer asking a question outside the approved policy.

Security and sensitive-data requirements can add development, testing and documentation effort. Describe the requirements early, and ask the provider to state which ones are included in the estimate. Avoid treating a general “secure” statement as a scope definition. The quote should explain the controls, responsibilities and evidence the project will deliver.

Published cost estimates are not universal rates. One custom-build estimate is based on a specific team and regional rate card; other estimates use different project scopes. Use them to frame a conversation, not to assume that a particular integration, channel or security requirement has a fixed price.

Budget for model usage, hosting, and upkeep after launch

A custom chatbot has recurring work after launch: model inference, hosting, monitoring, keeping the knowledge current, reviewing unanswered questions and adjusting the system when requirements change. A custom-build estimate puts typical inference at a few cents per conversation depending on the model and conversation length, and gives a broad monthly operating estimate of $300 to $2,000 for many mid-size deployments, plus planned maintenance of 15% to 20% of build cost per year.

Those are one provider’s estimates, not a universal operating budget. Your usage and architecture may differ. Ask for the assumptions behind any operating-cost figure: model, conversation length, traffic, hosting, monitoring and the time allocated to maintenance. If the quote includes model charges, ask whether they are fixed, metered or subject to a limit.

Model choice can change inference consumption. A system that uses a less costly model for routine questions and reserves a larger model for harder cases may have a different bill from one that routes every question to the same model. Conversation design matters too: a short exchange can use less inference than a long one. Do not optimize for brevity alone, though. A chatbot that fails to clarify a customer’s issue may create more work for the support team.

Plan time for these recurring checks:

• Refresh the knowledge. Remove outdated shipping, return and product information; make sure the chatbot is using the current policy.
• Review unanswered questions. Look for recurring gaps, unclear content and requests that should go to a person.
• Check handoff quality. Confirm that the team receives enough detail to continue the conversation without making the customer repeat everything.
• Test important changes. When policies, systems or prompts change, check that the chatbot still answers routine questions correctly and avoids unsupported promises.
• Monitor operating costs. Compare actual usage with the budget and investigate unexpected increases in conversation length or model consumption.

For a hosted platform, some operating work may be handled by the provider, but your team still needs to maintain its content and decide how to handle unanswered questions. For a self-built chatbot, your team or contractor also owns the hosting and monitoring arrangement. Make those responsibilities explicit before comparing a subscription with a custom build.

Hidden costs that make the advertised price incomplete

The subscription or build fee is only one part of the total cost. Check for overages, add-ons, extra seats, implementation work, integrations and the staff time needed to set up and maintain the chatbot. These costs can change the economics even when the headline plan appears affordable.

Before signing up, ask the provider to explain:

• Usage limits: What counts as a conversation, reply, credit or resolution? What happens after the included amount is used?
• Overages and add-ons: Are extra usage or features charged separately, and how are those charges calculated?
• Seats: How many people can manage conversations? Does each additional teammate change the monthly price?
• Setup: Is implementation included, or is there a separate project fee?
• Integrations: Which connections are included in the quoted scope? Who pays for custom development or ongoing troubleshooting?
• Contract terms: What is the billing period, commitment and renewal arrangement? How do annual terms compare with monthly terms?
• Maintenance: Who refreshes content, reviews failures, monitors the system and makes changes after launch?

Also consider the cost of switching later. If a platform no longer fits, you may need to rebuild workflows, reconnect channels and train the team on a new process. That future work is hard to price before you know what will change, but you can still ask whether your likely requirements are supported today and what data or content you can take with you.

Be careful with any estimate that combines implementation, software and usage into a single total without showing the assumptions. Request an itemized quote and keep unpriced work visible in your budget. “Not included” is more useful than a low estimate that leaves important responsibilities unclear.

Worked first-year budget for a small support team

A published example for 2,000 monthly conversations uses an outcome-based price of $0.99 per resolution and three seats at $29 each: 1,000 outcomes cost $990, the seats cost $87, and the monthly total is $1,077, or $12,924 for a year. That figure excludes unpriced implementation, maintenance and any other fees, so it is not a complete first-year project budget.

Here is the arithmetic:

Cost line	Monthly	First year
1,000 AI outcomes at $0.99 each	$990	$11,880
Three seats at $29 each	$87	$1,044
Software and seats in this example	$1,077	$12,924
Setup, maintenance and other fees	Not priced	Not priced

The example assumes 1,000 paid outcomes, not that all 2,000 conversations become paid resolutions. Its purpose is to make the pricing mechanism visible. If more cases are resolved, an outcome-based bill could rise; if fewer are resolved, it could fall. The provider’s definition of a billable result and any allowance or cap still matter.

For comparison, a flat-plan example in the same published comparison costs $129 a month, or $1,548 over a year on monthly billing. The comparison’s separate yearly-billing example is $948 a year. These are not the same plan terms as the outcome-based example, so do not read the difference as a direct quote for an identical package. The lesson is to compare the limits and included seats alongside the price.

For a practical first-year estimate, add the costs you can price and list the rest as explicit unknowns:

Budget line	What to enter
Subscription or platform fee	The chosen billing term and plan
Usage	Expected conversations, outcomes, replies or credits
Seats	The number of teammates who need access
Setup	Quoted implementation and content-preparation work
Integrations	Included connections and any custom work
Maintenance	Staff time or contracted support for ongoing upkeep
Hosting and monitoring	Relevant for a custom deployment
Contingency	Any usage or scope buffer your team decides to budget

Do not invent a setup or maintenance figure just to complete the table. Ask for a quote, or mark the line as unpriced until your team can estimate the work. That makes the comparison less polished but more useful: you can see which option has a known recurring fee and which still has significant unknowns.

For teams evaluating a flat support-desk fee, momo charges a fixed plan price in euros with an included number of AI conversations and no per-resolution fee. To compare other support tools and their pricing, see our AI chatbot business buying guide and small-business chatbot comparison. Its plans range from Free at EUR 0 to Scale at EUR 399 per month; yearly billing is 11 times the monthly price. The helpdesk inbox is included on every plan, and its AI answers from business content, citing sources when confident and opening a ticket when it is not sure. The right plan depends on the included limits and features your workflow needs.

A practical checklist before choosing a plan or custom build

Before choosing a platform or funding a custom build, count the work the chatbot will handle and identify the people and systems involved. Then price the first year and a longer-term period using actual limits, fees and responsibilities rather than the subscription headline alone.

Start with your support workload:

• Count the conversations that are appropriate for an AI answer, not just all incoming contacts.
• Estimate how many replies a typical conversation needs and how often the team must take over.
• Count the teammates who need to view or answer conversations.
• List the channels and integrations the workflow requires.
• Identify the business content the chatbot should use and who will keep it current.

Then check the commercial details:

• Confirm what the provider counts as a conversation, resolution, reply or credit.
• Record the included usage, seat count, channels and plan features.
• Ask what happens when usage or seat needs exceed the allowance.
• Separate recurring fees from implementation and integration charges.
• Check the billing term and any commitment or renewal conditions.
• For a custom build, clarify hosting, monitoring, model usage, knowledge upkeep and future changes.

Finally, compare the cost with the support work the chatbot is likely to take off your team. Count the repetitive questions, estimate the staff time they consume and consider how much human review remains. A bot that produces low-value answers or sends poorly prepared tickets to the team may not justify its cost, even if its subscription is inexpensive. A pilot or staged rollout can help you inspect real questions, escalation quality and usage before expanding the scope.

For a longer-term comparison, add plan fees, overages, add-ons, seats and staff time for setup and maintenance across the period you are considering. For custom development, include the upfront build and recurring operating costs. Keep assumptions visible, especially where an estimate depends on a particular vendor’s methodology or project scope.

Frequently asked questions

Can I build my own AI chatbot for free?

You can start with a free hosted plan if its limits fit your needs. For example, momo’s Free plan costs EUR 0 and includes 30 AI conversations as a one-time allowance, one agent, one team member, a website widget and a helpdesk inbox. A free plan does not mean a custom chatbot has no cost: custom development still requires someone’s time, and a self-built system may involve model usage, hosting and maintenance.

How much does a custom AI chatbot cost to maintain?

There is no single maintenance price that applies to every chatbot. A published estimate for many mid-size custom deployments gives operating costs of $300 to $2,000 per month and maintenance planned at 15% to 20% of build cost per year. Treat those as one provider’s estimate, not a universal rate. Ask what the figure includes, especially model inference, hosting, monitoring and staff time for updating knowledge.

Is a flat monthly chatbot plan cheaper than paying per resolution?

It can be, but the answer depends on the workload, the plan’s limits and what each vendor includes. One published comparison gives a $129 flat-plan example and a $1,077 outcome-based example with three seats for a stated 2,000-conversation workload. The examples use different pricing structures, so compare the outcome definition, allowances, seats and overages before deciding.

What should a chatbot implementation quote include?

Ask for a scope that identifies the knowledge sources, channels, integrations, handoff workflow, testing, deployment and ongoing responsibilities. It should separate development and setup from model usage, hosting, monitoring and maintenance. Ask who prepares and updates the content, how unanswered questions are reviewed, and what happens when an integration or handoff fails. Any requirement that is excluded should be named rather than left implicit.

How do I estimate the cost of chatbot model usage?

Start with the expected conversation volume, likely conversation length and model choice, then ask how the provider measures and bills usage. A custom-build estimate describes inference as depending on model choice and conversation length. For a hosted product, check whether usage is included in the plan, counted as conversations or replies, or charged separately. Compare the estimate with actual usage after launch and review it when conversation patterns change.

Compare the workflow before committing

A custom build makes sense only when its added control and capabilities justify the build and ongoing work. A hosted plan may suit a team that needs a defined support workflow without taking on a custom system. If you are weighing self-hosting, review live chat deployment costs and trade-offs. To see whether a flat-fee support desk fits your needs, try momo Free.