AI Chatbot: Build or Buy - Process, Providers, Criteria

An AI chatbot can answer questions around the clock and pre-qualify inquiries - but only if it's properly tailored to the business behind it. A poorly built bot that improvises on every unusual question damages trust more than a missing chat window ever could. This guide walks through the process from idea to running bot, the common provider types, and the criteria that separate a solid provider from a nice-looking widget.

When a chatbot is actually worth it

Before shopping for providers, an honest question comes first: does a chatbot solve a real problem? Typical patterns that suggest yes: a team keeps answering the same questions about services, pricing or opening hours. Inquiries come in outside business hours and sit unanswered until the next working day. First contacts need pre-qualifying so only suitable leads reach sales. If none of that applies, a chatbot is often an expensive vanity project - a better FAQ page or a simpler form usually does the job instead. Where that line between chatbot and FAQ page actually sits is covered in AI Chatbots: What They Can Really Do. We cover that calculation in more detail in AI chatbot: cost and benefit.

The process: from idea to a running bot

A workable chatbot takes shape in five steps, not by installing a plugin:

1. Define the use case. What should the bot actually do - answer FAQs, book meetings, qualify leads? The sharper the use case, the better the result. A bot meant to do everything usually does nothing well.
2. Build the knowledge base. The bot is only as good as what it knows. Website copy, FAQs, service descriptions and documents get collected and structured. Outdated or contradictory sources lead straight to wrong answers.
3. Train and configure. The bot is set up against the knowledge base, given a tone that fits the brand, and clear guardrails: what it can say, what it can't, and when it hands off to a human.
4. Embed it. The finished bot is integrated into the website or other channels, as a chat window or connected to booking and CRM systems so inquiries land directly in the right process.
5. Maintain it. A chatbot is not a one-time project. Real conversations reveal gaps in the knowledge base - regular refinement keeps it accurate.

Provider types compared

The right choice depends on how much customisation, integration and accountability matter - and how much effort the team can and wants to take on itself.

Selection criteria that reveal a solid provider

• Does the bot answer exclusively from vetted, proprietary sources instead of improvising from general model knowledge
• Does the bot clearly say when it doesn't know something instead of guessing, and hand off to a person
• Can the bot connect to CRM, calendar or existing tools when that's needed
• Is it transparent which AI model runs in the background and where the data is processed
• Is there a process to keep updating the knowledge base, rather than being left alone after launch

A provider that can answer these points concretely has understood the topic. One that dodges them is usually just selling a chat window with nothing behind it.

Data protection and GDPR

The moment a chatbot talks to real visitors, it's processing personal data. Four points belong in the provider selection from the start: a data processing agreement with every provider involved, clarity on the AI model in use and its server location, a visible notice in the chat window that users are talking to an AI system, and data minimisation in how conversation logs are stored. Sorting out data protection after the fact is more expensive than sorting it out beforehand - cutting corners here usually costs twice.

Concrete steps for the next 90 days

• Days 1-30: collect the five most common recurring questions, define the use case, compare provider types
• Days 31-60: structure the knowledge base, choose a provider with a data processing agreement and a suitable server location, configure the bot
• Days 61-90: launch the bot, review real conversations, refine the knowledge base based on actual questions

Conclusion

Building an AI chatbot isn't primarily a technology project - it's a knowledge project. A business that structures its knowledge base cleanly and sets clear boundaries for the bot ends up with a tool that genuinely helps instead of creating new confusion. For other processes worth automating alongside it, see AI in business: twelve processes. If it's unclear whether broader AI consulting would be a better first step, AI consulting for SMEs offers a starting point.