The honest answer for most UK businesses is: start with an off-the-shelf chatbot. It costs about £20 a month, works today, and solves more problems than owners expect. A custom AI system — software built around your own data and tools — is a real investment that earns its place only when a subscription genuinely can't do the job. This guide shows you where that line sits and how to tell which side of it you're on.
We build custom AI systems for a living, and we still tell most people to try a chatbot first. It's cheaper, it's faster, and the lessons you learn from it make any later build better. There's no shame in a subscription solving your problem. That's the whole point of the ladder.
What's the difference between a chatbot and a custom AI system?
Answer: A chatbot is a ready-made product you rent — ChatGPT, Microsoft Copilot, a website chat widget, a no-code bot builder. It answers questions from general knowledge or a small set of documents you upload. A custom AI system is software built for you: it reads your private data, connects to your existing tools, and runs a workflow. One is generic and rented; the other is specific and owned.
Think of it as the difference between hiring a temp who knows a bit about everything, and training a member of staff on exactly how your business runs. The temp is available immediately and cheap by the hour. The trained employee knows your systems, your clients, and your process — but you had to invest to get there.
A chatbot lives outside your business and talks to you through a text box. A custom system lives inside it, wired to the database, the CRM, the inbox. That wiring is where the cost — and the value — comes from. For the underlying vocabulary, our AI glossary covers the terms, and what an AI agent actually is explains the "system that takes actions" end of the spectrum.
When is an off-the-shelf chatbot enough?
Answer: A chatbot is enough when the task is answering questions, drafting text, or summarising documents — and the knowledge it needs is either general or small enough to paste in. If a £20-a-month subscription solves your problem, you don't need anything custom. Most first AI wins fall into exactly this category.
Cases where a chatbot is the right and only tool you need:
- Drafting and editing — emails, proposals, job adverts, first-draft copy.
- Summarising — turning a long document or meeting transcript into bullet points.
- Research and explaining — "explain this contract clause in plain English."
- Simple website support — a widget answering FAQs from a help-page you upload.
- One-person productivity — an assistant each staff member uses for their own tasks.
The test is simple. If the AI only needs to talk, and the information it needs is public or you can hand it the document each time, a chatbot covers it. Our guide on ChatGPT for business and AI tools for UK small business go deeper on getting value here without spending on a build.
When do you actually need a custom AI system?
Answer: You need a custom system when the AI must use your private live data, connect to your own tools and take actions in them, run reliably at scale, or automate a multi-step workflow end to end. If a chatbot keeps failing because it can't see your records or reach your systems, that failure is the signal to build.
Four honest triggers, any one of which justifies a build:
- Private data. The AI needs to answer from your live database, your document store, your customer records — not the general internet. A chatbot can't safely or reliably plug into that.
- Integration. It has to do things in your systems: update the CRM, raise an invoice, book an appointment, move data between tools that don't talk to each other.
- Reliability and scale. You need consistent, auditable answers across thousands of interactions, with guardrails and logging. Consumer chatbots don't give you that control.
- A real workflow. The job is several steps across several systems — triaging enquiries, chasing overdue invoices — not a single question and answer.
If none of these apply, you don't need a custom build, and we'll tell you so. Knowing your trigger also helps you choose your first AI project so you build the thing that pays back, not the flashiest demo.
How much does each cost?
Answer: A chatbot costs roughly £20 per user per month, with usable free tiers. A custom AI system is a software build — from £18,000, plus running costs per use. The gap is wide on purpose: one is a subscription, the other is bespoke software. That's why proving value with a chatbot first is the sensible order.
| Off-the-shelf chatbot | Custom AI system | |
|---|---|---|
| Typical cost | ~£20 / user / month (free tiers exist) | from £18,000 build + running costs |
| Ownership | Rented; vendor owns it | You own the system and the code |
| Data | General knowledge or small uploads | Your live private data, wired in |
| Integration | None, or shallow plugins | Deep — your CRM, database, tools |
| Reliability | Best-effort, no guarantees | Guardrails, audit logs, tuned for scale |
| Time to live | Minutes | Weeks to a few months |
Between "£20 chatbot" and "£18k build" there's a cheaper step worth knowing about: a Blueprint, from £4,500, a fixed-price plan with an honest build or don't-build verdict. It's how you avoid spending £18k to learn the chatbot would have done. For the full picture, see our pricing guide.
What about my data and security?
Answer: With a chatbot, your data leaves your business and sits on someone else's servers under their terms — fine for general use, riskier for personal or client data under UK GDPR. With a custom system, you control where data lives, who can see it, and how it's logged. If sensitive data is involved, that control is often the reason to build.
A few things worth being straight about:
- Consumer vs business chatbot plans differ. Free and personal ChatGPT tiers may use your inputs to improve models; business and enterprise plans contractually don't and add data controls. Check which plan you're on before pasting anything sensitive.
- GDPR still applies. Putting personal data — customer records, staff details — into a third-party tool is a processing decision the ICO expects you to have thought through. The ICO publishes guidance on AI and data protection.
- Custom means containment. A build can keep data inside your own Supabase or cloud account, scoped by row-level security, with an audit trail of every access. That's harder to get from a rented widget.
If your main worry is "can I put company data into ChatGPT," that question deserves its own answer before you commit either way — and it often nudges the decision toward a controlled custom build.
Can I start with a chatbot and upgrade later?
Answer: Yes — and it's usually the best route. Run a chatbot for a few months on the real problem. You'll learn where AI genuinely helps, which questions recur, and exactly where the subscription hits a wall. That evidence makes a later custom build cheaper, tighter, and far more likely to pay back.
The staged path looks like this:
- Try the chatbot on the actual task. Cost: near zero. Time: a day.
- Watch where it fails. Can't see your data? Can't take the action? Not reliable enough? Note the specific gap.
- Scope only the gap. You don't rebuild what the chatbot does well — you build the part it can't.
- De-risk the build. Start with a free AI Readiness Assessment, then a paid audit before committing to a full build.
Upgrading isn't throwing the chatbot away. Most businesses keep the subscription for drafting and research, and add a custom system for the one workflow that needed private data and real integration. If you're weighing who builds that, how to choose an AI development company covers what to look for.
What's the honest verdict?
Answer: Start with a chatbot. Prove AI helps, learn where it stops, and only then spend on custom software for the gaps a subscription can't fill. A custom system is worth it when private data, integration, reliability, or a real workflow are on the line — and not before.
Most people who ask us "should we build an AI system?" leave with "not yet — try this £20 tool first, come back if it can't do X." When X turns out to be private data plus a real workflow, that's a genuine build, and we'll scope it honestly. Spending £18k to replace a £20 subscription is the mistake worth avoiding, and picking the right first step is how you avoid it.