Chatbot Cost to Build: $15,000 to $50,000 in 2026
What a chatbot costs to build in 2026 by route: platform per-resolution fees, one engineer, or a custom build at $15K to $50K. Dated table and what moves the number.
Jun 30, 2026
Chatbot Cost to Build: $15,000 to $50,000 in 2026
Short answer: a production chatbot that answers from your own knowledge, hands off to a person, and connects to one or two systems costs $15,000 to $50,000 to build with a senior team, and $5,000 to $10,000 per month to keep improving afterwards. A chatbot that also takes actions inside several systems (updates the CRM, issues refunds, books appointments) or handles regulated data runs $50,000 to $100,000. If you only need answers from a help centre, a platform product does it for a per-seat or per-resolution fee and no build at all. Those are KUMO's published bands; the platform prices below were read from the vendors' own pages on 29 September 2026.
For a range on your own scope, the AI development cost calculator asks four questions and gives you one in about a minute.
Chatbot cost in 2026, by route
| Route | What you get | Cost | Source, checked 29 Sep 2026 |
|---|---|---|---|
| Platform chatbot on your helpdesk (Intercom Fin) | AI answers from your help content inside an existing helpdesk, hands off to agents | $0.99 per resolved conversation ("Fin outcome"), plus $29, $85 or $132 per agent seat per month on Intercom's plans, billed annually | intercom.com/pricing |
| Build it yourself with an AI coding tool | A working demo in days; production readiness depends on what you check before customers use it | Tool subscription plus a paid review | See the production readiness review linked below |
| Hire one engineer | One person to build and operate it, no design or QA cover | US median software developer wage $135,980 per year before benefits and recruiting | US Bureau of Labor Statistics, Occupational Outlook Handbook, May 2025 wage |
| Custom chatbot, first production version (KUMO Starter Build) | Your knowledge, your systems, human handoff, evaluation, deployed in your cloud, code yours from day one | $15,000 to $50,000, typically 2 to 3 months | KUMO published band |
| Custom chatbot that takes actions across several systems (KUMO) | Write access to CRM, helpdesk, payments or bookings, permissions, audit trail, regulated data | $50,000 to $100,000 | KUMO published band |
| Ongoing improvement after launch (KUMO) | Unanswered-question review, retrieval tuning, integration upkeep, monitoring | $5,000 to $10,000 per month | KUMO published band |
The platform row is cheap when your questions are already answered in a help centre and your helpdesk is where the conversations live. It stops being cheap when the chatbot has to look up an order, check an account, or do something, because that is integration work whichever logo is on the widget.
What moves the number
Six things decide where inside a band a chatbot lands. Most quotes hide which of them are doing the moving.
1. Where the answers come from. Public help articles are the easy case. Private documents, account data and policies that change weekly need retrieval with permissions, citations and a way to say "I do not know". This is the difference between a small build and a Starter Build.
2. Whether it acts or only answers. Reading an order status is one integration. Creating a ticket, changing a booking or issuing a refund means validation, duplicate prevention, approval rules, audit records and rollback. Each write-enabled system adds weeks.
3. Who is talking to it. Anonymous website visitors are simple. Logged-in customers with different plans and permissions need identity, and the chatbot must never show one customer another customer's data.
4. Human handoff. Every production chatbot needs a route to a person with the conversation summary attached, a queue that someone owns, and a way back. Designing this properly is a fixed cost that cheap quotes leave out.
5. Testing before launch. A demo is not evidence. The build needs a set of real, awkward and unsafe questions with agreed pass criteria, and someone to run it before release and after every change.
6. Channels. Website chat is the baseline. WhatsApp, voice, email and in-app each add integration and testing. WhatsApp AI on Claude shows what the WhatsApp channel involves.
Three quotes for the same chatbot brief
A buyer typically gets three quotes that look like $8,000, $35,000 and $90,000 for "a support chatbot". They are honest quotes for three different products.
- $8,000: a platform set-up or a thin wrapper over a model, answering from a fixed set of pages, no account lookup, handoff by email. Fine if that is the job.
- $35,000: a Starter Build as written: your knowledge with citations, one or two systems read, proper handoff to your helpdesk, an evaluation set, deployment in your cloud, handover documentation.
- $90,000: the roadmap priced as one project: actions in several systems, customer identity, multiple channels, audit and compliance. Sometimes right; usually better as two phases with real usage between them.
Compare quotes on what is in writing: which systems are read and which are written to, who owns the code and the accounts, how handoff works, what testing is included, and what happens the month after launch. How to compare software proposals has the checklist.
Running costs after launch
The build is not the total. Budget from month one for model usage (usually tens to a few hundred dollars a month at Starter scale, rising with conversations), hosting and vector storage, the platform or helpdesk seats you already pay for, and engineering time to review unanswered questions, refresh knowledge and keep integrations working. Without engineers of your own, that last item is the $5,000 to $10,000 per month band, or a smaller retainer once the chatbot is stable. Production AI support cost breaks down what that time is spent on.
What the Starter band buys at KUMO
A KUMO Starter Build ($15,000 to $50,000) for a chatbot is a launched first production version, not a prototype: retrieval over your knowledge with citations, one or two integrations read or written under stated rules, human handoff into your helpdesk, an evaluation set with pass criteria, deployment into your own cloud account, and a handover another engineer can continue from. Two to three months is typical. The code is yours from day one.
KUMO's founders built Volopay's first production version as its founding engineers, and CampaignHQ is KUMO's own product, in production since 2022, which is where the habits around handoff, monitoring and evaluation come from. How KUMO builds AI products is on AI product and platform engineering; if the chatbot is one part of a wider automation, see AI workflow automation.
If you already built a chatbot with an AI tool and want to know whether it is safe to put in front of customers, the vibe-coded app production readiness review lists the checks.
How to get a real number this week
- Write one page: who talks to the chatbot, the top ten questions it must answer, the systems it must read or update, and what is out of scope for version one.
- Run it through the cost calculator to see which band it lands in.
- Send the same page to two or three vendors and ask for scope in and out, ownership, handoff design, testing, and the month-after-launch plan, in writing.
- Prefer the quote that answers all of that, even if it is not the lowest.
If you would rather have that page scoped with you, book a free 30-minute call with an engineer and bring whatever you have. You will leave with a band and the reasons for it.
FAQ
How much does it cost to build a chatbot? A production chatbot that answers from your own knowledge, hands off to people and connects to one or two systems costs $15,000 to $50,000 with a senior team. One that takes actions across several systems or handles regulated data costs $50,000 to $100,000. A platform chatbot on an existing helpdesk costs a per-resolution fee (Intercom Fin: $0.99 per outcome, read 29 September 2026) plus seats, with no build.
How much does a chatbot cost per month? After launch, expect model usage and hosting in the tens to low hundreds of dollars a month at Starter scale, plus $5,000 to $10,000 per month for ongoing engineering if you have no engineers of your own, or a smaller retainer once it is stable.
What is the cost of implementing a chatbot on an existing helpdesk? On Intercom, Fin AI Agent is priced at $0.99 per resolved conversation and plans run $29 to $132 per agent seat per month billed annually (read 29 September 2026). Implementation is configuration, not a build, as long as you do not need account lookups or actions.
What makes a chatbot more expensive? Private or changing knowledge, write access to systems, customer identity and permissions, human handoff design, evaluation depth, extra channels such as WhatsApp or voice, and regulated data.
Is a custom chatbot worth it over a platform one? Only when the chatbot has to look things up in your systems or do things for the customer. If your answers are already in a help centre and your helpdesk is where the conversations live, use the platform.
How long does a custom chatbot take to build? A Starter Build typically takes two to three months from scoped brief to launched first version. Multi-system builds take longer, and the timeline should be in the quote with milestones.
Last verified 29 September 2026. Intercom prices from intercom.com/pricing and the software developer median wage from the US Bureau of Labor Statistics, both read on that date. KUMO bands are KUMO's published prices.
