Three Ways to Build an AI Chatbot, and How to Pick One
There are three routes to an AI chatbot for a UK business: a no-code builder, an in-house build on an LLM API, or a UK chatbot agency. Each has a different ceiling and failure mode, and the right choice depends far less on budget than on what the bot actually has to do.
All three routes work. All three also fail regularly, usually because the business chose the route before deciding what the bot was for. This guide covers the decisions to make first, what each route involves, and a checklist that applies whichever you choose. For a fixed price on your own project, book a free discovery call at /contact.
What You Need to Decide Before You Build Anything
Every chatbot project we have seen go wrong skipped one of these questions.
Which jobs must the bot do? Be specific. "Answer customer questions" is not a job. "Answer the questions our support inbox receives most, quote for our standard packages, and book a call with sales" is. Rank the jobs by how much staff time each costs today. A bot that does a few jobs well beats one that does many badly.
Which channels? Website widget only? WhatsApp? Messenger? Inside your customer portal or app? Each channel changes the conversation style: a WhatsApp customer expects short replies and a fast handover; a website visitor will read a longer answer.
What data does it need? Static content (FAQs, product pages, policies) is easy. Live data (order status, booking availability, account details) means integrations, and integrations are where timelines move most.
What happens when the bot cannot help? This is the handoff, and most businesses design it last. Does the conversation go to a shared inbox, a live agent, a phone callback, or a helpdesk ticket? A bot with a bad handoff makes customers angrier than no bot at all.
Answer those and the right route usually becomes obvious.
Route 1: No-Code Chatbot Builders
These are the subscription tools that let you paste in your website URL, upload some PDFs, and have a working widget within an hour.
Widely used tools such as Tidio, Chatbase and Botpress bill on a subscription that scales with volume, seats and which AI model powers the bot. Some, including Intercom's Fin AI agent, charge per resolved conversation instead of, or on top of, a flat fee, so the bill moves with how many conversations the bot actually closes.
Pros: live in an afternoon with no developer needed, hosting and model access handled for you, good enough for straightforward FAQ deflection, and easy to switch off if it is not working.
Cons
- •Answers only from uploaded content; live lookups are limited or unavailable
- •Integrations restricted to the vendor's connector marketplace
- •Limited control over tone, guardrails and what the bot refuses to discuss
- •Transcripts sit on the vendor's infrastructure, often outside the UK — the ICO's guidance on international transfers treats this as a "restricted transfer" whenever a UK GDPR-covered business sends personal data to an organisation outside the UK, so check the vendor's terms and safeguards before you go live
- •Costs climb sharply as volume grows or you need the "enterprise" tier for basics like removing branding
When it is enough: if the bot's job is to answer common questions from your website content, capture a name and email, and hand over to a human, start here — you can move on later.
Route 2: Build It Yourself on an LLM API
For businesses with a developer who wants full control. You build on a model API from OpenAI, Anthropic or Google, or a self-hosted open-source model, and write the application around it.
What is actually involved: prompt design, writing and testing the instructions governing tone, scope, refusals and format, which takes longer than anyone expects; retrieval over your documents, a pipeline that chunks your content, stores it in a vector database, and retrieves the right passages for each question (this is the "RAG" you will see mentioned everywhere, and getting it accurate is most of the work); integrations connecting to your CRM, booking system, helpdesk or order database so the bot can act, not just talk; evaluation and monitoring, a test set run every time you change a prompt, plus logging, failure flags and cost tracking once live; and hosting and security, an API layer, rate limiting, authentication for live data, and a plan for provider outages.
The realistic effort, and the maintenance nobody budgets for
For a competent developer new to this, a first version with retrieval and one integration is genuinely time-consuming, measured in developer weeks rather than days, and it is rarely the version you keep. This is where DIY builds quietly die: models get deprecated, content goes stale, a new product launches and nobody updates the prompt, the developer moves on. Budget ongoing attention every month, and be honest about whether anyone will actually do it.
When it is right: you have a developer with capacity, the bot is core to your product rather than a support add-on, and you want to own the whole stack. It is wrong if the developer is doing this in the gaps between other work.
Route 3: Hire a UK Chatbot Developer or Agency
Pay a specialist to design, build, integrate and maintain the bot. This is what most UK SMEs end up doing once the bot needs to connect to real systems.
What a good build includes: a discovery phase producing the job list, channel plan and handoff design; retrieval over your own content with a process for keeping it current; integrations with your CRM, helpdesk, booking or e-commerce platform; guardrails covering what the bot refuses and when it escalates; a test set built from your real customer questions, run before launch; a UK GDPR review covering data location, retention and the model provider's terms; and monitoring with a monthly optimisation cycle after launch.
A chatbot that stores customer data or makes automated decisions can trigger obligations under UK GDPR, including Article 22's rules on solely automated decision-making and, in some cases, a Data Protection Impact Assessment. Confirm your specific obligations with your DPO or a data protection adviser before launch.
What sets the price is how many systems the bot must connect to and how many channels it runs on, not anything arbitrary. Ask any agency to walk you through exactly which of these drives their number.
| Type of bot | What you get | What sets the price |
|---|---|---|
| FAQ bot | Answers from your website and documents, lead capture, email handoff | Content volume, tone and channel |
| Integrated bot | Everything above plus one or two live integrations (CRM, booking, order lookup), WhatsApp or Messenger | Number and complexity of integrations |
| Custom AI agent | Multi-step tasks, several integrations, multiple channels, voice handoff, bespoke guardrails and evaluation | System count, guardrails, evaluation depth |
For a wider view of what AI projects involve in the UK, see /blog/ai-development-cost-uk-2025.
What to ask before signing: Is the price fixed, and what is in scope? Where is my data stored, which model provider is used, and is my data used for training? Who owns the prompts, the code and the content index at the end? What does the ongoing fee cover, and what happens if I stop paying it? Can I see a test set and results from a previous project, with client details removed? How is the handoff to my team designed, and what happens out of hours?
If a developer cannot answer these plainly, keep looking. Our fuller custom versus off-the-shelf comparison is at /blog/custom-ai-chatbot-vs-off-shelf-uk-2025.
The Three Routes Compared
| No-code builder | DIY on an LLM API | UK developer or agency | |
|---|---|---|---|
| What sets the cost | Volume, seats, model tier | Developer time, mostly | Integrations, channels, guardrails |
| Time to launch | Hours to days | Several developer weeks for a first version | A few weeks |
| Integrations | Vendor connectors only | Anything, if you build it | Anything in scope |
| Data control | Vendor's infrastructure | Full | Full, under your contract |
| UK GDPR | Check vendor terms and data location | Your responsibility entirely | Reviewed as part of the build |
| Maintenance | Vendor handles the platform, you handle content | You, indefinitely | Covered by the ongoing fee |
| Best for | Simple FAQ deflection | Product-critical bots with in-house developers | Bots that must connect to real systems |
A 7-Step Chatbot Build Checklist
This applies whichever route you take.
- 1. Define the jobs. Write the ranked job list and the channels. Anything not on the list is out of scope for version one.
- 2. Gather the source content. Collect the FAQs, policies and product information the bot will answer from, and fix what is out of date now.
- 3. Choose the model and channel. Pick a provider whose data terms you can live with, and launch on one channel before adding others.
- 4. Design the handoff. Decide where conversations go when the bot cannot help, who receives them, and how quickly.
- 5. Test with real questions. Pull genuine questions from your inbox and chat logs, run them through the bot and grade every answer before a customer sees it.
- 6. Measure containment and CSAT. Containment is the share of conversations resolved without a human. Track it alongside satisfaction, because a bot that contains most conversations by frustrating people into leaving is not a success.
- 7. Iterate monthly. Review transcripts, find the wrong answers, update content and prompts, rerun the test set.
Common Mistakes
- •Starting with the tool instead of the jobs. The subscription gets bought, the bot gets switched on, nobody can say what it is for.
- •Feeding it your whole website. Old blog posts and outdated content become answers. Curate what it draws on.
- •No handoff. The bot loops, the customer gives up, and the complaint arrives by email anyway.
- •Skipping the test set. If you have not graded its answers to real questions, you do not know how good it is.
- •Treating launch as the finish line. The bot is at its worst on day one and gets good through review.
How BigBerri Approaches Chatbot Builds
We are a Manchester AI development company and we build chatbots for UK businesses on a fixed price, agreed after a free discovery call, where we do the job list, channel and handoff work described above so the quote reflects what you need rather than a generic package.
Every bot we build retrieves answers from your own content, so it says what your business says and nothing else. Human handoff is designed in from the start, and after launch we run a monthly optimisation cycle: reviewing transcripts, updating content and rerunning the test set. We also build our own AI products, so the team building your chatbot has shipped and maintained production AI systems of its own, not just configured someone else's platform. Full service details are at /services/ai-chatbot-development.
If you are weighing these three routes, book a free discovery call at /contact.
Frequently Asked Questions
How much does it cost to build an AI chatbot in the UK?
It depends on the route: a no-code builder is a monthly subscription that scales with volume, a DIY build on an LLM API costs mainly developer time, and a UK agency build is priced on integrations, channels and guardrails rather than a flat rate. Book a free discovery call at /contact and we will give you a fixed price once we know what the bot needs to do.
Can I build an AI chatbot for my business without a developer?
Yes, if the bot only needs to answer questions from your existing content and capture leads. No-code builders can be live within an hour. Once the bot needs to look up orders, book appointments or write to your CRM, you will usually need a developer or an agency.
How long does it take to build an AI chatbot?
A no-code bot can be live in hours. A DIY build with retrieval and one integration typically takes several developer weeks for a first version. A UK agency build usually takes a few weeks from kickoff to launch depending on integrations and channels.
Is an AI chatbot compliant with UK GDPR?
It can be, if data location, retention and the model provider’s terms are checked and documented. With no-code tools, review the vendor’s data processing terms and hosting region, since sending data to a provider outside the UK is treated as a restricted transfer under the ICO’s guidance. With a DIY or agency build, decide where transcripts are stored, how long they are kept, and confirm the model provider does not train on your data. If the bot will make automated decisions about customers, check with your DPO whether Article 22 UK GDPR or a Data Protection Impact Assessment applies.
Which is better: a no-code chatbot builder or a custom-built chatbot?
A no-code builder is better for simple FAQ deflection with a short job list and no live integrations. A custom build is better once the bot needs to connect to your systems, follow specific guardrails, or run on several channels. Many businesses start with no-code and move to a custom build when they hit its limits.