How Much Does an AI Chatbot Cost? What Actually Sets the Price

The short answer: two bills and six cost drivers

Every AI chatbot comes with two bills. The first is the one-time cost of getting it designed, connected and launched. The second is the ongoing cost of running it: platform fees, AI model usage, and the hours someone spends keeping its answers accurate. A vendor who quotes only the first bill is telling you half the story, and so is one who advertises only a low monthly subscription.

Both bills are shaped by the same handful of decisions:

1. Build path. Configuring a subscription product, extending a platform with custom integrations, or assembling an assistant from components you own.
2. Content readiness. Whether your policies, service descriptions and answers already exist in clean, current form, or have to be written and reconciled first.
3. Integrations. Each system the bot reads from or writes to, such as a CRM, a scheduler, an order database or a help desk.
4. Channels. A website widget alone, or also text messaging, social messaging and voice, each with its own rules and testing.
5. Volume and model choice. How many conversations it handles and which AI model processes them, since model providers bill by usage.
6. Maintenance ownership. Who reviews transcripts, updates knowledge and repairs broken connections once it is live.

That is why a figure you find in an article or an ad rarely matches the quote for your business. The rest of this page explains how each driver behaves, so you can decide deliberately which ones are worth spending on.

Three ways to get a chatbot, and where each one’s money goes

Most chatbot projects fall into one of three build paths. They differ less in how clever the AI is and more in who controls the data, the integrations and the recurring bill.

None of these is automatically the cheaper option. A subscription tool that fits out of the box is usually the lowest-cost route to launch. The same tool can become the expensive route if its per-conversation pricing climbs with your traffic, or if you keep paying for workarounds because it cannot reach the system your customers actually ask about. A custom build costs more up front and can cost less over several years, but only when someone is committed to maintaining it.

Whether the spend pays back is a separate calculation from what it costs. Our guide to AI chatbot ROI for small business covers how to measure the return once you know the investment.

What the one-time build actually pays for

When a quote lists a build phase, the hours are spread across work most people never picture when they imagine a chatbot. Writing the system prompt is a small slice of it.

Discovery and scoping

Deciding which questions the bot will answer, which it will decline, when it hands off to a person, and what success means in measurable terms. A narrow, well-defined scope is the single biggest lever on total project cost.

Content preparation

This line item surprises people most. An AI assistant answers from the material you give it, and many businesses discover their website, PDFs and staff knowledge disagree with one another. Old pricing pages, retired services and conflicting return policies all have to be found and reconciled, or the bot will confidently repeat the contradictions. A company with a clean, current knowledge base spends far less here than one whose answers live in three employees’ heads.

Conversation design

Writing the flows for the handful of conversations that matter most: qualifying a lead, answering a policy question, collecting details for a callback. It includes the fallback behavior when the bot does not know something, which is where visitor trust is won or lost.

Integration work

Connecting the bot to other systems through their APIs. Read-only connections, such as checking whether an order shipped, are simpler than write connections, such as creating a lead record, because writes need validation, duplicate protection and error handling.

Testing and launch

Running the bot against realistic and hostile questions, checking that it stays inside its scope, confirming handoffs reach a real person, and deploying it to the chosen channels.

What drives the recurring cost once it is live

The ongoing bill is where estimates drift most, because part of it moves with your traffic.

• Model usage. AI model providers typically bill per token, a small unit of text, counting both what is sent to the model and what it writes back. A bot that attaches a long document to every question costs more per conversation than one that retrieves only the relevant paragraphs. Design choices made during the build set this number for years.
• Model tier. Larger models cost more per token than smaller ones. Plenty of turns, like greeting a visitor or confirming business hours, do not need the most capable model, and routing simple turns to a smaller one is a legitimate cost control.
• Platform and hosting fees. Subscription tiers, messaging channel fees, and the server or cloud resources a custom assistant runs on.
• Conversation volume. More conversations mean more usage. Seasonal businesses should expect the bill to follow their busy months.
• Human upkeep. Someone has to read a sample of transcripts, correct wrong answers, and update the knowledge base when a price, policy or service changes. This rarely appears on a vendor invoice, yet it is a real cost.

Eight scope decisions that move a chatbot quote up or down

Use this table while you plan. Every row is a choice you can make on purpose rather than a surprise on an invoice.

Compliance deserves a special note. A bot that could touch protected health information, payment data or other regulated records needs vendors willing to sign the appropriate agreements, logging that never stores what it should not, and a review of what the bot is allowed to ask. That work is necessary, and it belongs in the quote from the first conversation rather than being discovered after launch.

Costs that rarely appear on the first quote you receive

• Texting registration. Business text messaging in the United States involves carrier registration and consent handling. It takes paperwork and lead time that a website-only quote never mentions.
• Model changes. Providers retire and update models. When the underlying model changes, the bot’s behavior can shift, and it needs re-testing against your most important conversations.
• Integration breakage. When your CRM or help desk changes its API, or account permissions change, a connection can fail quietly. Monitoring that alerts someone is far cheaper than discovering a week of lost leads.
• Knowledge drift. Every change to a price, policy or service needs the same change in the bot’s sources. Businesses that skip this pay in wrong answers instead of hours.
• Exit costs. If your flows, transcripts and knowledge base live only inside one vendor’s product, leaving means rebuilding. Ask before signing what you can export.
• Your team’s time. Discovery calls, content review and testing all take staff hours. Nobody invoices you for them, but they are part of the real cost.

Worked example: scoping a chatbot for an online specialty retailer

The business below is invented to show how scope decisions change a quote. No prices appear because the reasoning is the point, not a number.

An online retailer of specialty kitchen equipment receives a steady stream of repetitive emails: where is my order, does this pan work on an induction cooktop, how do I start a return. The owner’s first request is a chatbot that does everything, including processing returns, recommending products and answering in both English and Spanish.

What the full wish list would require. Write access to the store platform to create return authorizations, product data cleaned and structured for recommendations, two languages of testing, and a handoff into the support inbox. Each item is reasonable. Together they produce the largest possible quote.

How the scope was split into phases.

1. Phase one: answer compatibility and shipping-policy questions from cleaned product pages, and look up order status read-only using the order number and the email on file. Anything about returns collects the details and opens a support ticket for a person.
2. Phase two, only if phase one earns it: guided return initiation with write access, designed after transcripts show how customers actually phrase return requests.
3. Phase three, optional: Spanish support, once the English answers are stable enough to translate and test.

What changed in the quote. Phase one dropped the write integration, the multilingual testing and the recommendation data work. The owner also agreed to rewrite the product care pages herself, which removed most of the content preparation hours. The recurring cost stayed predictable because the bot retrieves a few relevant product paragraphs per question instead of stuffing the whole catalog into every request.

The lesson generalizes: the most affordable reliable chatbot does a few jobs completely rather than many jobs halfway.

Ways to lower the cost without building a bot nobody trusts

• Launch on one channel. Start with website chat and add texting or social messaging after the answers are proven.
• Prepare your own content. Owners who write or approve the source answers in advance cut the most expensive build hours.
• Start read-only. Let the bot look things up before it is allowed to change anything.
• Match the model to the task. Reserve the larger, costlier model for questions that genuinely need reasoning.
• Keep the first handoff simple. An email or ticket with the conversation attached is often enough at launch.

What not to cut: testing, a clear fallback for questions the bot cannot answer, and a named person who reviews transcripts. Those three separate a bot that saves staff time from one that creates cleanup work.

How EVOTECH IT LLC turns your requirements into a fixed-scope quote

EVOTECH IT LLC is a US-based team in the Houston area delivering web, software and AI projects remotely to businesses nationwide. We have been in business for more than 20 years and are rated 5.0 on Google. We do not publish chatbot prices, because a number without a scope is a guess. Here is what happens instead:

1. Free consultation by phone or video. We ask which questions reach your team most often, where your answers currently live, which systems hold the data customers ask about, and what should happen when the bot cannot help.
2. A build-path recommendation. We tell you plainly whether a configured subscription tool would serve you or whether custom integration or a custom build is justified. Sometimes the honest recommendation is the smaller project.
3. A written fixed-scope quote. It names the questions in scope, the integrations, the channels, the handoff method and the testing, and it lists the recurring third-party costs you should expect apart from our build work.
4. Phased delivery where it helps. We can propose a first phase that proves its value before you commit to the rest.
5. A clear upkeep plan. We explain what maintenance the bot will need and who can do it, whether that is your team or ours.

To start, call (832) 359-2425 or book a free consultation. Bring a list of the questions your team answers most, and we will show you what a sensible first phase looks like. You can also read more about our AI chatbot development work.

Related services

Frequently asked questions