Chatbot Development Cost: Build vs Buy
Building a chatbot can cost a few hundred a month or six figures a year. The variable is not the model, it is who builds and maintains it.
Prices verified 18 July 2026. Vendors change pricing often, so every figure links to its source.
There are four ways to end up with a chatbot, and they differ by roughly two orders of magnitude in cost. This page lays out what each really involves, including the part most estimates leave out: the work after launch.
One caveat before the numbers. Unlike our vendor pricing pages, where every figure is read off the vendor's own page, development costs are genuinely ranges. Contractor rates vary by market and scope varies by project. Treat everything below as an estimate to sanity-check a quote, not a price list. matram.ai's own numbers are exact.
The four routes, side by side
Cost, time to launch, and who owns it afterwards are the three things that actually separate these.
| Route | Upfront (estimate) | Ongoing (estimate) | Time to launch |
|---|---|---|---|
| DIY on an LLM API | Engineering time only | API + hosting + maintenance | Weeks to months |
| Freelance developer | Low thousands to tens of thousands | API + hosting + fixes | Weeks |
| Agency build | Tens of thousands upward | Retainer, often monthly | Months |
| SaaS chatbot | None | $29 to $199 / mo (matram.ai) | Minutes to hours |
The row that surprises people is the last one. A modern SaaS chatbot trains on your existing content, so the thing you would have spent weeks building (retrieval over your docs) is the default behaviour rather than the project.
Building it yourself on an LLM API
The API bill is rarely the expensive part. Engineering time is.
A competent in-house build means retrieval over your content (chunking, embeddings, a vector store), prompt design, a chat UI, escalation to a human, logging, and an admin surface so non-engineers can fix wrong answers. That is a real project, not an afternoon.
- Model API usage: metered per token, so it scales with conversation volume. Check the current rate card for the model you pick.
- Hosting and a vector database: modest, but non-zero and permanent.
- Engineering time: the dominant cost. Loaded cost of the hours matters more than the infrastructure bill.
- Ongoing: model updates, prompt regressions, new content, edge cases, security patches.
This route makes sense when the chatbot is genuinely core to your product, when you need behaviour no vendor offers, or when data cannot leave your infrastructure.
Freelance and agency builds
You are buying someone else's hours, plus a handover risk.
Freelance
Cheaper than an agency and often faster. The risks are concentrated at the end: code quality varies, documentation is often thin, and support after handover depends on one person's availability. Insist on a written scope, a maintenance arrangement, and a repository you own from day one.
Agency
You pay materially more and get process: design, integration work with proprietary systems, compliance review, documentation, and a support agreement. That is worth it for regulated industries, complex internal integrations, or when there is no in-house engineering to own the result. Expect an ongoing retainer, because a chatbot is not a deliverable that sits still.
How to decide
The question is not build versus buy in the abstract. It is whether your requirements are actually non-standard.
- Standard support, FAQ or lead-capture use case, no in-house engineering: buy. Almost always.
- Chatbot is a differentiating part of your product: build, and staff it properly.
- Strict data residency or compliance that no vendor meets: build, or agency with a compliance review.
- Unsure: buy first. A month on a SaaS plan tells you exactly which requirements are real, and that is cheap discovery for a build you might still do.
For reference, buying looks like this: matram.ai is $29, $69 or $199 a month flat with unlimited seats, crawls your site to train, cites its sources, and installs with one line of JavaScript. There is a 7-day free trial and no credit card, and an Enterprise tier quoted on request, invoiced, for volume above the published plans. Run your own numbers with the chatbot ROI calculator before committing to a build.