AI Chatbot Development Company, Freelancer or Platform? A Small-Firm Choice

Most small businesses should start with a chatbot platform, because it covers questions-from-your-documents bots with human hand-over for a modest monthly fee. Hire a freelancer for a defined custom job, and a development company only for complex integrations or strict hosting needs. Whoever builds it, you should own the accounts, the documents and the prompts.

The choice is rarely "which is best". It is "which is the least you need". Every step from platform to freelancer to company adds a person between you and the bot, which adds cost, delay and a dependency. Add that person only when the platform has demonstrably failed the job you need done.

Follow me on Instagram@sagnikteaches

Platform, freelancer and company compared

Start by writing the scope in plain words

Whichever route you take, scope decides the cost. A one-page scope also exposes whether you need development at all. Here is a filled-in illustration for a tutoring centre; every line is a choice, not a given.

Connect on LinkedInSagnik Bhattacharya

CHATBOT SCOPE (tutoring centre, illustrative)
PURPOSE: answer parents' common questions out of hours and pass
         booking requests to the office
CHANNEL: website chat widget only (no messaging apps in phase 1)
ANSWERS FROM: fees page, timetable, trial-lesson policy, FAQ (14 pages)
MUST DO: give fees from the published list; offer trial-lesson times
         from a list; collect name, child's year and email
MUST NOT: promise places, discuss a child's progress, give refunds,
         give advice outside the documents
HAND-OVER: to the office inbox with the chat transcript, within hours;
         message at night says when the office will reply
INTEGRATIONS: none in phase 1 (email only)
VOLUME: about 300 chats a month
SUCCESS: 60% of chats resolved without staff; no wrong fee quoted
LANGUAGES: English only

This scope fits a platform without any development. Notice how few lines need custom code. "Integrations: none" is the line that keeps cost down. The moment a bot must read a live booking system or write to an accounting package, a freelancer or company enters the picture.

Subscribe on YouTube@sagnikteaches

Route one: a chatbot platform

Platforms let you upload documents, write instructions, set hand-over rules and paste a widget into your site. Prices from vendor pages, checked in November 2026:

• Chatbase. No free plan, but a 7-day trial. Hobby is $40 a month ($32 billed yearly) for 700 message credits. Basic is $150 a month ($120 yearly) for 4,000 credits, and its plan card lists the Chatbase Helpdesk inbox; Business is $500 ($400 yearly) for 15,000 credits.
• Tidio. The first 50 Lyro AI conversations are free as a one-time, lifetime allowance. The Lyro AI agent, which lists human hand-off among its features, starts at $32.50 a month for 50 conversations; the page has a monthly or annual toggle, so check which billing that figure assumes.
• Intercom Fin. From $0.99 per outcome, where an outcome is a confirmed or assumed resolution or a completed workflow, including hand-offs. It runs on Intercom seat plans (listed at $29, $85 and $132 a seat a month, with a new-customer discount on the cheapest when checked) or alongside your current helpdesk with a minimum monthly commitment, such as 50 outcomes.

Prices and plan names move often, so check the live pages. The pattern matters more than the numbers: you pay by subscription, by message credits or by outcome, so estimate your monthly chat volume first. At 300 chats a month, an outcome-priced tool at $0.99 costs roughly $300 at most if every chat were billed, and usually less. A credit-based plan may be cheaper or may run out mid-month; read the overage terms.

Where a platform wins

• You can test it this week for a few tens of dollars.
• You own the account, so there is nothing to hand over.
• The vendor fixes model changes and security patches.
• You can switch vendor by exporting your documents and instructions.

Where a platform fails

• It cannot reach a system the platform has no connector for.
• It cannot apply business logic, such as "quote differs by order size and delivery type".
• Data must stay in a particular place or environment that the vendor does not offer.
• Your volume makes per-outcome pricing worse than a fixed-price build.

If a platform only fails on one connection, you may be able to bridge it with an automation tool rather than custom development. A two-week pilot plan for a small-business chatbot shows how to find those limits cheaply.

Route two: a freelancer

A freelancer suits a defined job: configure a platform properly, write the instructions, connect a booking or order system, test it, hand it over. Published market ranges give a feel for the cost. One freelancer-hiring page for Zapier work (Contra) lists $30 to $150 an hour, $500 to $1,500 for simple automations and $2,000 to $10,000 or more for complex ones, with retainers from $1,000 to $5,000 a month. Chatbot work is not identical, but the range shows the width of the market. Treat it as a snapshot to verify with real quotes.

What to ask a freelancer

1. Which platform will you build on, and will the account be mine from day one?
2. Can I see a live chatbot you built and its transcripts?
3. What will you do when it answers wrongly, and how quickly?
4. What is the fixed price, and what is excluded?
5. Are you free to support it for three months after launch, and at what rate?
6. Do you earn a commission from the platform you recommend?

The final question matters. Platforms run partner and affiliate schemes, and some publish rates: Make advertises 35 per cent for 12 months to affiliates, n8n 30 per cent for 12 months, and HubSpot's affiliate programme 30 per cent monthly recurring commission for up to a year, per their affiliate pages in November 2026. A commission is not wrong, but it is a reason to ask what else they considered.

Key-person risk

One person means holidays, illness and career changes. Insist on a handover document, a recorded walkthrough and credentials stored in your password manager. Then a freelancer leaving is an inconvenience, not a crisis.

Route three: a development company

Companies bring teams, project managers and capacity. They suit work a freelancer cannot carry: integration with several internal systems, custom interfaces, strict hosting or compliance needs, voice channels, or a bot that is part of a larger product. They also bring process: scoping documents, change requests, and invoices that grow.

The risks specific to companies

• Overbuilding. A custom stack where a platform subscription would do. Ask them directly: "Why not a platform for phase 1?" A good company will give a reason, or admit one.
• Lock-in. Code in their repository, hosted on their account, built on their framework. Specify ownership and handover in the contract.
• Retainer creep. A monthly fee for "maintenance" without a defined list of what it includes.
• Selling the next phase. Phase 1 is cheap, phase 2 is where the money is. Ask for a total cost of ownership across three years.

For quotes on larger builds, what to ask an AI agent development company and what a quote covers goes deeper, and custom build or off-the-shelf covers the wider build-or-buy question.

Contract questions that protect you

This is practical guidance, not legal advice. Have a qualified adviser read the final contract. These are the clauses I would look for, in plain words.

1. Ownership. All accounts, documents, prompts, configuration and any code belong to your business. The developer holds a licence to use them only to perform the work.
2. Accounts in your name. The platform subscription is billed to your card, with the developer added as a user. Never the other way round.
3. Handover. On request, within a stated number of days, at no extra charge: exports, documentation and a walkthrough.
4. Ongoing fees. Every recurring charge named, with the amount, what it covers and how to cancel.
5. Data. What customer data the bot sees, where it is stored, who can access it, how long it is kept and whether any of it is used to train models.
6. Acceptance test. A written test, such as 40 sample questions with expected answers, that the bot must pass before final payment.
7. Warranty period. A fixed period during which defects are fixed free.
8. Change process. How extra requests are priced and approved.
9. Termination. What happens if either side ends the work early.

The data and handover clauses are the ones people skip and regret. If the developer goes quiet, a clause about handover and your own accounts is the difference between a stalled project and a lost one. See what if your AI vendor shuts down for the checks to make on the platform side.

A print shop's bot, costed three ways

Say a print shop wants a website chatbot to answer questions on turnaround, file formats and prices, collect quote requests and hand over to staff. Volume is about 400 chats a month. Illustrative costs follow and must be replaced with real quotes.

• Platform, self-configured. A mid-priced subscription, say $120 a month, plus 12 hours of the owner's time to set up and test, valued at $25 an hour, which is $300. First-year cost about $1,740.
• Platform, configured by a freelancer. The same subscription ($1,440 a year) plus a fixed fee of about $700 for set-up and a two-hour handover. First-year cost about $2,140.
• Custom build by a company. A project fee of several thousand dollars, a hosting cost and a retainer. Say a fee of $8,000 plus $200 a month for hosting and support: first year about $10,400.

The company route costs about five times the platform route, for a bot that answers the same questions. It earns its place only if the shop needs something the platform cannot do, such as pricing from a custom estimating system. If it does not, the extra $8,000 is paying for ownership of code nobody will maintain.

Five more cases, with a verdict each

A yoga studio: platform, self-set-up

Questions about class times, passes and first visits. Documents already exist. Verdict: a platform with a human hand-over, configured by the manager in an afternoon.

A wedding planner: platform plus a freelancer for one connection

The bot answers package questions and should book consultations into a calendar. The calendar connection is the only non-standard piece. Verdict: platform, with a freelancer for one afternoon to connect and test the booking.

A tutoring centre: platform, but do the data homework

Parents give children's details. Verdict: platform, but check the vendor's data and training terms, keep the bot away from anything about a child's progress, and limit what it collects to a name, a year group and an email.

A parts distributor with a live stock system: freelancer or company

The bot must check stock and delivery dates in an existing system. That is a real integration. Verdict: a freelancer if the system has a documented interface; a company if it does not or if several systems are involved.

A firm with strict hosting requirements: company, with eyes open

If contracts require data to remain in a specified environment, a standard platform may not qualify. Verdict: a company or a self-hosted route, with the ownership and handover clauses above, and a clear year-three cost.

Sample questions and the answers you should expect from a bidder

Asking is one thing; knowing what a good answer sounds like is another. Here are three of the questions above, with an illustrative good reply and a warning sign for each.

"Why not a platform for phase 1?"

A good reply: "Your orders sit in a system with no standard connector, so the platform could answer questions but not check a delivery date. We could launch the question-answering part on a platform in two weeks and build the delivery lookup as phase 2." A warning sign: "Platforms are limited and will not scale", with no mention of which limit applies to you.

"Who owns the code and the accounts?"

A good reply: "Everything is in your accounts; we are added as users. You get the repository on handover, free." A warning sign: "We host it for you, and that is included in the retainer", which usually means the bot cannot leave without a rebuild.

"What happens when it gives a wrong answer?"

A good reply names a process: a transcript review in week one, a log of wrong answers, a fix within an agreed time, and who decides what the bot must refuse. A warning sign: "The AI is very accurate", which is a claim, not a process.

The hosting clause nobody wrote

Illustrative: a small distributor paid a company for a chatbot and the contract said nothing about hosting. The bot ran on the company's server. Eighteen months later the company changed its pricing, doubling the monthly fee, and the distributor discovered that moving the bot meant paying for a rebuild, because the knowledge documents had been loaded into the company's own system with no export. The cost to leave was higher than the cost to stay. Had the contract required accounts in the distributor's name and an export on request, switching would have taken a week. Contract clauses feel like paperwork on the day you sign; this is why they matter.

When a custom build pays back: the sum

Use this to test whether a custom build is even plausible. Say a bot resolves 250 chats a month that staff would otherwise handle, each taking 6 minutes. That is 25 hours a month. At $25 an hour, the saving is $625 a month, or $7,500 a year. Against that, a custom build of $8,000 plus $2,400 a year in hosting and support has a first-year cost of $10,400, so it does not repay in year one. Net of the $200 monthly hosting, the bot saves $425 a month, so the $8,000 fee takes about 19 months to recover if nothing goes wrong. The same bot on a platform at $120 a month saves $505 a month net, so a $700 set-up fee is recovered inside two months. Custom becomes the better bet only when volume is several times higher, or the platform simply cannot do the job. These figures are illustrative; run the sum with your own.

Checking the platform against your scope before you talk to anyone

An afternoon of testing can save you a development fee. Take your scope document and run it through a free tier or trial. Upload your real documents. Ask the 40 test questions. Try to make it quote a wrong fee, promise something it should not or answer outside its documents. Test the hand-over: does a message reach a person, with the transcript, and what does the customer see at night?

Score the platform on each line of your scope: done, done with a workaround, or not possible. Anything in the third group is what a freelancer or company would have to build, and your scope for them is now exactly that list. Many firms discover the list is empty, or has one item. That is the best possible outcome of the exercise, and it costs you an afternoon.

How to test a quote before you sign

1. Ask each bidder to write the scope back in their own words. Differences show misunderstandings early.
2. Ask them to price a platform-only version and a custom version of the same scope.
3. Ask for the year-one and year-three totals, with every recurring line.
4. Write 40 test questions yourself, including ten awkward ones (a complaint, a refund, a question outside your documents), and make passing them the acceptance test.
5. Run a two-week pilot on a share of chats before full launch.

The "platform-only version" request is the most revealing. A bidder who refuses to price it is protecting their own work. One who prices it honestly and explains the gap is someone you can work with. What AI chatbot consulting and set-up cost gives a view of the other end of the price range.

How to tell you chose well

After the first month, check four things: the share of chats the bot resolved without staff, the number of wrong or invented answers found in a sample of 50 transcripts, how long it takes you to change an answer, and what you paid in total against the estimate. If you cannot change an answer yourself in ten minutes, your set-up depends too much on someone else. If the bot invents answers, tighten the instructions and sources before you add features.

If you would rather settle this with someone who has no commission on any platform, my AI implementation consultation covers choosing between these routes and handing over something your team can run. Preparing your questions first, as in how to build the FAQ your AI chatbot needs, makes any route cheaper.

Questions to settle before you hire anyone

When is a platform enough?

When the bot answers questions from your own documents, hands over to a person, and connects to tools the platform already supports. Most small firms fit this. Custom development earns its fee only for unusual integrations, strict data-hosting needs or logic a platform cannot express.

Who should own the chatbot?

You. The platform account, the knowledge documents, the prompts and any code should sit in accounts held by your business, with the developer added as a collaborator. Put this in the contract, along with handover on request and no fee to hand over.

What ongoing fees should I expect?

Platform subscription, usage or per-conversation charges, any AI model costs billed separately, and a support or maintenance retainer if someone looks after the bot. Ask for each as a line item, and for the year-two total, not just the build price.

How do I judge a developer's past chatbots?

Ask to see a live one, the transcripts of a test, and what went wrong in the first month. A developer who can describe a failure and its fix is more trustworthy than one who shows only polished demos. Ask permission to speak to a past client.

Further reads

• How Much Does a WhatsApp Chatbot Cost? Fees, Tools and Setup — Fees and tools if your customers use messaging apps.
• Do I Need an AI Consultant? 8 Signs It's Time to Get Help — Signs that outside help will pay for itself.
• AI Automation Setup for Small Businesses: DIY or Hire It Done? — The same build-or-hire question for automations.
• How to Evaluate an AI Implementation Proposal or Quote — Read any proposal line by line.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Chatbase, Tidio and Intercom (Fin) pricing pages; Make, n8n and HubSpot affiliate pages; Contra's Zapier-freelancer hiring page for market rates. All checked November 2026.