The Short Answer
Intercom prices its Fin AI agent per resolution — a fee charged each time the AI closes out a conversation — on top of a seat plan for the underlying helpdesk. A custom AI chatbot from us is a fixed build with a flat monthly running cost that does not move when your conversation volume does.
That single structural difference decides the answer. Below a modest, steady volume of AI resolutions a month, per-resolution pricing is simpler and less hassle. Above it, you are paying a toll on your own customer service that grows every time the AI works well, and a custom build starts to pay for itself.
This guide works through the two pricing models so you can see which one suits your volume, using the vendors' own published pricing structures. Build costs are typical ranges we quote, not fixed prices, and any comparison should be made against a live quote from both sides.
What You Are Comparing
These are not the same product, and the honest comparison starts there.
Intercom Fin is an AI agent bolted onto a full customer service platform. You get the inbox, the ticketing, the reporting and the AI together. It answers from your help centre content and can run workflows.
Tidio Lyro is the same idea at the small business end, bundled with live chat.
A custom chatbot is software we build for you. It answers from your own documents, prices, stock or booking system rather than only a help centre, it lives wherever you want it, and nobody charges you per answer.
If you have no helpdesk at all, Intercom sells you a helpdesk with AI attached, which is a genuine reason to choose it. If you already have a helpdesk, or your questions need live data rather than articles, you are paying for a platform you do not need.
How the Two Pricing Models Actually Work
Intercom charges Fin per outcome, where an outcome is counted when the customer confirms the issue is resolved, does not ask again, or a workflow completes, on top of a seat plan for the underlying helpdesk that scales with the tier of features you need. Tidio's Lyro AI conversations work the same way in structure: a number of free conversations, then paid usage that rises in bands as your volume grows, on top of its own seat plans.
A custom build works differently. The engineering — connecting to your systems, designing the conversation flows, testing it against real questions — is a one-off cost. The ongoing running cost is model usage and hosting, which scales gently with volume rather than charging per successful answer, so it stays broadly flat whether the bot handles a trickle of conversations or a high volume of them.
That difference in structure is the whole comparison. Platform AI pricing scales with success: the better it works, the more you pay, conversation by conversation. A custom build's running cost does not carry that multiplier. Ask any platform vendor for their current published rate card before you commit, since these figures move, and ask any custom build provider for their fixed price and their flat monthly running cost in writing.
Working Out Your Own Crossover Point
You do not need the vendors' exact figures to know which side of this you sit on — you need your own conversation volume and a rough sense of how much of it an AI could resolve.
| Monthly AI resolutions | What happens to platform cost | What happens to a custom build's cost | Generally cheaper |
|---|---|---|---|
| Low and steady | Stays low; usage barely moves the bill | The flat build cost has little volume to spread across | Platform |
| Moderate and growing | Climbs in a straight line with every resolution | Stays flat; only hosting and model usage move | Roughly level — depends on your rates |
| High and consistent | Becomes the largest line in the support budget | Stays flat, so the build cost is fully absorbed | Custom, usually by a wide margin |
The pattern matters more than any single figure. Platform AI costs scale with success: the better it works, the more you pay, conversation by conversation. A custom build costs the same whether it answers ten questions or ten thousand, because the running cost is model usage and hosting rather than a per-answer fee.
Count the support conversations you handle in a typical month and estimate what share are the same handful of questions an AI could plausibly close out — most support teams find this is around half to two thirds once they look. Ask a platform vendor to model your actual monthly volume against their current rate card, and ask us to do the same for a custom build's running cost. Comparing two real numbers, not two headline structures, is the only way to know which is cheaper for your business specifically.
Where Fin Genuinely Wins
We build custom chatbots for a living and we still tell people to use a platform when it is the right answer.
You need a helpdesk anyway. If you are running support out of a shared mailbox, Intercom solves two problems at once and the AI cost is modest at low volume.
Your volume is low and steady. A small, steady stream of AI-resolved conversations a month will rarely justify a build.
You want it live this week. Point it at your help centre and it works. A custom build takes weeks, not days.
Nobody internally owns software. A platform is maintained for you. A custom bot needs someone, us or you, to look after it.
Where a Custom Build Wins
Your answers are not in articles. If the question is "where is my order", "do you have this in a 12" or "when is my next appointment", the answer lives in a database, not a help centre. A custom bot reads the live record. This is why the pattern recurs in retail, covered at /industries/ecommerce.
Volume is real. Past a meaningful number of resolutions a month, per-answer pricing becomes the largest line in the budget and grows as you grow.
The bot must do things. Book, reschedule, refund within a policy, update a record. Actions across your systems are where platform AI hits its limits fastest.
Data residency matters. With a custom build you decide where conversations are processed and stored. UK or EU hosting under a data processing agreement is a requirement in several sectors, and it is a design decision rather than a plan tier.
You want the asset. You own a custom bot. Its logic, prompts and integrations are yours, and they do not stop working if a vendor changes its pricing structure.
The Cost Nobody Puts in the Comparison
Both routes carry a cost that never appears in a pricing table, and it is the same one: somebody has to keep the answers right.
On a platform, the AI answers from your help centre. If an article is out of date, the bot is confidently wrong at scale, and you will not find out from the dashboard. Somebody has to read conversation transcripts every month, spot the questions it fumbled, and go and fix the underlying article. In our experience this is where platform deployments quietly decay: the bot is set up well, nobody owns the content, and six months later the support team is overriding it constantly.
With a custom build the same duty exists, but the failure is usually more visible because the bot is reading live data rather than prose, so a wrong answer tends to mean a broken connection rather than a stale sentence.
Budget an hour or two a month for whoever owns this, on either route. It is the cheapest hour you will spend, and skipping it is why so many businesses conclude that AI chatbots do not work when what actually happened is that nobody maintained one.
Questions to Settle Before You Choose
- 1. How many support conversations do you handle a month today?
- 2. What share of them are the same handful of questions?
- 3. Do the answers live in articles, or in a system?
- 4. Do you already pay for a helpdesk you would keep?
- 5. Would you need the bot to take an action, or only to answer?
- 6. Where must conversation data be stored?
- 7. What happens to your costs if volume doubles?
If your answers point at articles, low volume and no helpdesk, buy the platform. If they point at live data, real volume and actions, build.
The Hybrid Nobody Mentions
Starting on a platform and moving later is a perfectly good plan, and it is what we would suggest to most businesses handling a modest volume of conversations a month. Use Intercom or Tidio to learn what customers actually ask, export those transcripts after a few months, and use them as the specification for a build. You will get a far better custom bot for having waited, because the questions will be real rather than guessed.
The mistake is drifting into a large recurring platform bill without ever doing that arithmetic against what a fixed-cost build would run.
How We Approach It
We quote a fixed price after a free discovery call, and on that call we will tell you plainly if a platform subscription is the better buy for your volume. When a build does make sense, we connect it to your own data, design the handover to a human first, and set it up so you own the result.
For a wider view of the build options, see /blog/how-to-build-an-ai-chatbot-for-your-uk-business, and for the custom versus off-the-shelf decision in general see /blog/custom-ai-chatbot-vs-off-shelf-uk-2025.
Want the crossover worked out on your real numbers? Book a free discovery call at /contact.
Frequently Asked Questions
At what volume does a custom chatbot become cheaper than Intercom Fin?
It depends on your own conversation volume and current rates from both sides, but the pattern is consistent: below a modest, steady volume of AI resolutions a month, per-resolution pricing usually wins on cost and effort; above it, a custom build with a flat running cost starts to win because platform fees keep climbing with success while the build cost does not. Count your monthly support conversations and we can work out your specific crossover point on a discovery call.
Does a custom chatbot have running costs too?
Yes, but they are flat rather than per-answer — covering model usage, hosting and monitoring — and do not rise because the bot succeeded more often. We give a fixed monthly figure once we know your setup.
Can a custom bot connect to the helpdesk we already use?
Yes. We integrate with Zendesk, Freshdesk, HubSpot, Salesforce and most systems with an API, so the bot answers what it can and hands the rest to your existing queue rather than replacing it.
What if we start on a platform and want to move later?
That is often the sensible order. Run a platform for a few months, export the transcripts, and use the real questions as the specification for a build. You get a better bot because the requirements are evidence rather than guesswork.
Who holds the conversation data with a custom build?
You do. We design where conversations are processed and stored, typically the UK or EU under a data processing agreement, with a retention period you set. Confirm your own obligations with whoever handles data protection for your business.