Website support · Decision guide & cost calculator
Build or Buy an AI Chatbot for Your Website?
For a standard website FAQ, start by checking whether a chatbot platform covers the job. Build a custom chatbot when a verified requirement, your existing technical resources or the full cost comparison justifies it. A freelancer can help with either route: configuring a platform and developing software are different services.
The decision in 30 seconds
Choose the software and the person doing the work separately. A platform can save development effort; it still needs someone to prepare the knowledge, check answers and maintain the setup. Building gives you more design control and more operating responsibility. Compare the same scope, include your time, and decide who owns the service after launch.
Four ways to get a website chatbot
“DIY” does not necessarily mean writing code. You can configure a hosted chatbot yourself, build your own application, pay someone to configure the hosted product, or commission custom software. The first choice concerns the system; the second concerns who implements it.
| Route | What you do or commission | What remains after launch |
|---|---|---|
| A · Configure a platform yourself | Choose the plan, prepare sources, configure the agent and add its widget. | Subscription, knowledge updates and checks. The provider operates its product. |
| B · Build it in-house | Assemble the interface, retrieval, model calls, hosting and operating controls. | Your team maintains the application and its dependencies. |
| C · Freelancer sets up a platform | Pay for configuration, content preparation, integration and agreed checks. | Platform subscription plus whatever maintenance you commission or retain. |
| D · Freelancer builds custom software | Commission a defined application, integrations, deployment and handover. | Hosting, model usage and an agreed owner for fixes, updates and incidents. |
A hybrid is also possible: a managed chatbot can call a small custom API. This can preserve the hosted interface while adding a specific business function. It is worth checking before commissioning an entire replacement.
Compare the same chatbot before comparing prices
A price comparison is useful only when the services are comparable. A public FAQ widget, a support inbox staffed by people and an agent that changes customer orders are different projects.
Our reference job: a small website FAQ
A text chatbot answers from a defined set of public pages, points visitors to supporting sources, handles an unanswered question and offers a contact link. The budget model assumes 200 conversations a month, with two model requests per conversation.
Private accounts, order changes, voice, extra messaging channels and a native helpdesk are outside this example. Add their real implementation and operating costs if your project needs them.
Write down your sources, languages, expected traffic, update frequency and fallback before requesting a quote. Then check the actual product plan or proposal against those requirements. An inexpensive subscription that lacks a required function is not a substitute for the service you need.
What does each route cost over 12 months?
Separate money paid out from the value of your own time. In the illustrative FAQ scenario below, configuring a platform costs $480 in payments and $540 in internal time over one year. Building a new application in-house has a similar technical bill, but the assumed extra work takes its economic cost to $2,612.16.
Illustrative USD budgets, not provider quotes or measured delivery times. Hours and rates are editable assumptions. These totals compare costs; they do not establish equal answer quality.
| Route | Payments | Internal time value | Economic cost |
|---|---|---|---|
| A · Platform, self-configured | $480.00 | $540.00 | $1,020.00 |
| B · Custom, built in-house | $482.16 | $2,130.00 | $2,612.16 |
| C · Freelancer + platform | $1,180.00 | $240.00 | $1,420.00 |
| D · Freelancer + custom build | $3,932.16 | $300.00 | $4,232.16 |
See all original hours, rates and cost assumptions
The internal hourly value is $30; the freelance rate is $50. Both are chosen inputs, not market averages. Setup is counted once; recurring costs and maintenance run for 12 months.
| Route | Internal hours: setup / monthly | Freelance hours: setup / monthly | Monthly service / other |
|---|---|---|---|
| A · Platform alone | 6 / 1 | 0 / 0 | $40 / $0 |
| B · Build in-house | 35 / 3 | 0 / 0 | $30 / $10, plus model usage |
| C · Freelancer + platform | 2 / 0.5 | 8 / 0.5 | $40 / $0 |
| D · Freelancer + custom | 4 / 0.5 | 45 / 2 | $30 / $10, plus model usage |
The $40 platform allowance uses Chatbase Hobby’s monthly-billing price as a reference. Chatbase pricing lists 700 message credits on Hobby. Annual billing is a different commitment. The plan must still meet your functional and usage requirements.
The custom example combines the $5 Workers paid subscription from Cloudflare’s pricing documentation with the first $25 Supabase Pro project from Supabase pricing. This is one possible component budget, not a required stack or a validated deployment. The extra $10 monthly is an assumed allowance for retrieval, ingestion and monitoring costs, not a quoted package.
Taxes, currency conversion, marketplace fees, unplanned incidents, migration and human resolution work are outside the base totals. Add AI coding subscriptions or extra services if you need them. The calculator keeps service budgets flat until you change them: it does not automatically model plan upgrades or every usage charge.
Calculate your own 12-month budget
Change the assumptions to match your project. Results update in the comparison table above. The original table and its assumptions remain readable without JavaScript.
Why a tiny API bill does not make a chatbot nearly free
The reference workload generates 400 model requests monthly. At 3,000 input and 300 output tokens each, Anthropic’s published Haiku 5.5 rates of $0.10 per million input tokens and $0.50 per million output tokens produce a $0.18 monthly token estimate. That uses standard global pricing, prompts under the 100,000-token threshold, and no caching or server-tool charges.
The application still needs a usable interface, retrieval, updates, monitoring and someone who fixes failures. The calculator therefore separates token usage from technical services and labour. A low-cost model is a component choice; it does not prove that the chatbot answers your questions reliably.
Platform billing also has its own units. Chatbase’s model comparison lists different credit tiers by model. Forecast model requests and check actual usage rather than assuming that 700 credits mean 700 complete visitor conversations.
Download the reference cost table (CSV) · Download the full original assumptions (JSON)
When building it yourself makes sense
Building becomes more attractive when you already have relevant skills, reusable infrastructure and a person responsible for operating it. It is a different calculation from learning the stack and creating a new application solely to avoid a subscription.
The existing-infrastructure case changes the result
Suppose route B reuses resources already funded for other work. Replace its setup with eight hours, upkeep with half an hour monthly, and its extra hosting/database budget with $5 monthly. Keep the $10 other allowance and $0.18 token estimate.
The 12-month result becomes $182.16 in extra payments and $420 in internal time: $602.16 combined. This is a marginal-cost scenario. It excludes the cost of acquiring those skills and funding the existing infrastructure. Select “Existing infrastructure and skills” in the calculator to load it.
A working prototype is the beginning of the operating job
An AI coding tool can help produce a prototype. Before treating it as production software, identify who reviews the code, protects credentials, controls spend, deploys changes and restores the service. Include that work in the estimate rather than treating generated code as a completed delivery.
Low-code does not remove this question. n8n’s self-hosting documentation covers configuration and keeping an instance running. Its backup guide explains that a complete recovery needs more than workflow exports alone. If you choose that architecture, maintenance and recovery belong in the project.
Custom development is worth considering when a written requirement is not met by the platforms you have checked, or when you need control over a component that materially affects the service. Specify the unmet requirement first; “more flexible” is too vague to price.
When a platform or a hybrid setup makes sense
A managed platform is a practical route to evaluate when the required interface, knowledge sources and fallback fit its capabilities, and you want to avoid owning the whole application. You still own the decisions about approved content, customer experience and escalation.
A business integration does not automatically require a fully custom chatbot. Chatbase Custom Actions can call your own API. That provides a possible hybrid: keep the managed conversation layer and commission only the missing integration. Access control, validation and error handling still need a defined implementation.
Check the feature gate
The right subscription depends on the workflow. For example, Chatbase’s current plan comparison lists API access on Basic rather than Hobby. The $40 FAQ reference budget should not silently become the budget for an API-based application.
Check the full journey
For a private lookup, decide how the customer is authenticated and which record may be returned. For an action, define permissions, approval and proof of completion. Price that journey whether it runs on SaaS, custom code or both.
For a closer product evaluation, read our Chatbase review covering pricing, credits and limits. If the project specifically concerns store orders and returns, the Chatbase Shopify integration guide covers that separate implementation.
Shortlisting a managed chatbot?
Check the plan against your required sources, functions and usage before paying for implementation.
Hiring a freelancer: platform setup or custom development?
Ask which system you will own or subscribe to after delivery. “Build a chatbot” can describe a configured SaaS account, an automation workflow or a separate application. A one-off fee does not establish what is included or replace recurring subscriptions.
For platform setup, ask for the account configuration, source preparation, installation, checking and handover. For custom development, also specify the repository, deployment instructions, dependency list and maintenance arrangement. Keep billing and administrative access under accounts you control.
A brief you can copy into a quote request
“We need a text chatbot on [website] that answers from [approved public sources], in [languages], for approximately [conversations/month]. It must cite the relevant source, handle an unanswered question and offer [contact route]. Please separate a platform-based proposal from custom development if both are suitable.
“Quote setup, recurring services, usage, maintenance and optional work separately. Name the accounts and plan levels required, the included source updates, the acceptance checks and the handover materials. Explain how another provider could take over. Private data and write actions are [excluded / specified separately].”
Ask the provider to state assumptions and exclusions alongside the price. If a quote promises handoff, specify whether it means a contact link, a ticket or a staffed live conversation. If it promises maintenance, name the tasks and response arrangements rather than accepting the word alone.
A small fixed-price package may be useful for a narrowly defined setup. It cannot be compared fairly with an application quote until you match integrations, recurring fees, ownership and post-launch work. Use the same brief for each provider.
Maintenance, ownership and leaving the system
The chatbot needs an owner after launch whichever route you choose. Allocate source updates, answer review, budget monitoring, connector changes and incident handling. Otherwise those jobs remain unassigned even if the initial deployment works.
| Responsibility | What to agree | Evidence to keep |
|---|---|---|
| Knowledge updates | Who changes approved content and checks that the agent uses it? | Source inventory, update steps and a repeatable question. |
| Budget and incidents | Who sees alerts, reviews usage and responds to a failure? | Access, limits, contact details and a recovery procedure. |
| Account and code access | Which accounts, configurations and code are handed over under the agreement? | Client-controlled administration and documented deployment. |
| Export and exit | What can be exported, in which format, and what must be rebuilt? | An actual sample export and a replacement plan. |
Chatbase Data Sources documents how knowledge is managed and kept current. Its Activity documentation describes conversation exports in JSON, PDF or CSV. Exporting records is useful, but it does not by itself recreate the prompts, integrations and behaviour of a working replacement.
For a custom application, code access alone is also incomplete. Ask whether a different authorised person can deploy it, locate the required services and restore the agreed data. Put any migration or recovery work that you expect to pay for into the cost model.
What to test before accepting delivery
Agree the questions and evidence before the build starts. The following is a proposed acceptance checklist to use with your provider, not a set of product test results.
| Check | Expected behaviour | Evidence |
|---|---|---|
| Known and unknown questions | Correct supported answer, or an honest fallback when the source does not answer. | Question, response and supporting source. |
| Changed or conflicting content | Agreed update and conflict rules work. | Source change, update steps and repeat question. |
| Contact, mobile and language | Visitors can use the widget, sources and fallback on the intended devices and languages. | Real-page checks and receipt of the intended message. |
| Quota or provider failure | The bot follows its fallback without claiming a task succeeded. | Controlled failure and the resulting visitor experience. |
| Private access or write action, if included | Only authorised data is shown; the intended action happens with the agreed approval. | Controlled accounts and a record in the destination system. |
| Handover and recovery | Another authorised person can follow the agreed operating instructions. | Accounts, documentation, export and a recovery demonstration. |
Pay attention to what a test actually executes. Chatbase’s Tests documentation distinguishes mocked Actions, which return a supplied response without running the operation, from live tests limited to read-only Actions. A successful simulated refund therefore does not prove that a refund occurred.
For write operations, arrange a controlled check in the destination system and inspect the resulting record. Re-run relevant questions when the sources, model or integration changes. The downloadable checklist provides fields for scope, expected behaviour, evidence and the person accepting the result.
Which route fits your website?
You want a public FAQ and have no technical owner
Evaluate platform capabilities first. Configure it yourself if you can prepare and review the content; request a platform setup quote if that work is the obstacle. Keep someone responsible for changes after launch.
You already operate the relevant stack
Calculate the incremental work and spending. Building can make sense when you reuse skills and infrastructure and can keep the service running. Use the marginal-cost scenario rather than pretending a new project is equally cheap.
You need a specific private integration
Scope the identity check, permissions and destination action. Compare a platform plus a custom endpoint with a complete custom build. Commission only the architecture that meets the verified requirement.
You cannot define a useful support journey yet
Improve the FAQ or contact flow first. A chatbot subscription and a custom build both create work; neither resolves unclear policies or missing answers on its own.
Our decision rule: first confirm the required service, then compare capable architectures, then choose who implements it. Use the cost calculator and the same brief for each quote. Select on fit, operating ownership and total cost together.
Questions before you decide
Can I create a website chatbot without coding?
Yes. You can configure a hosted platform instead of writing an application. You still need approved content, an unanswered-question fallback and someone to review the live service. If a workflow needs a custom integration, check whether the platform already supports it before commissioning extra code.
Is building a chatbot cheaper than buying one?
It depends on the scope and the resources you already have. In our illustrative 12-month scenario, self-configuring a platform costs $1,020 including internal time; a new in-house build costs $2,612.16. Reusing existing skills and infrastructure changes the custom route to a $602.16 marginal cost. These are editable scenarios, not market quotes.
Can a freelancer set up a chatbot platform for me?
Yes. Platform configuration is a different service from custom development. Ask which subscriptions remain, whose accounts are used, what content preparation is included and who maintains the setup. A setup fee does not normally pay for all future services unless the proposal explicitly includes them.
When does a custom chatbot pay for itself?
Calculate the extra upfront cost divided by positive net monthly savings, including internal time and maintenance. If the custom route has no monthly saving, that formula gives no payback. Feature fit, quality and migration costs still need a separate assessment.
Do private data or business Actions require a custom chatbot?
Not automatically. Some platforms can call a custom API, so a hybrid can satisfy the requirement. Verify the actual plan and integration, authentication, permissions and error handling. Choose a complete custom build when the capable platform options do not meet the requirement or the full economics justify it.
What should I receive from a chatbot freelancer?
Agree a working deployment, the appropriate client-controlled accounts, source inventory, configuration or code covered by the proposal, acceptance evidence and operating instructions. Separate subscriptions, maintenance, future changes and exit work in the quote.
Sources
Official pricing and documentation checked November 1, 2026. The cost model combines published component prices with the explicit hours, rates and budgets in our original assumptions. Use the linked provider pages to confirm the plan and billing conditions for your project.
- Chatbase pricing — billing options, plan features and credits
- Chatbase model comparison — message-credit tiers
- Chatbase Custom Actions — calling your own endpoint
- Chatbase Data Sources — knowledge and updates
- Chatbase Activity — conversation records and exports
- Chatbase Tests — mocked and read-only live Actions
- Anthropic API pricing — Haiku 5.5 token rates and prompt threshold
- Cloudflare Workers pricing — paid subscription and usage
- Supabase pricing — Pro starting price and project costs
- n8n self-hosting — configuration and operation
- n8n backup and restore — recovery scope