When an AI chatbot really helps your small business

Almost every website has a chat window these days. Yet customers keep phoning and emailing. The cause is rarely the language technology but the set-up: a chatbot that knows nothing about your diary, orders or prices can only give general answers. Here is when an AI chatbot genuinely takes work off a small business’s hands, and how to judge that yourself.

What an AI chatbot is

An AI chatbot is software that understands questions in plain language and replies in plain language. Old chatbots used fixed buttons and menus, and stalled as soon as a question was not on the menu. Modern chatbots run on a large language model (an LLM) and can cope with a question phrased in a different way.

On its own, that model knows nothing about your business: not your opening hours, not your rates, and certainly not the status of a booking. Everything that makes it useful has to be given to it.

How a chatbot becomes useful

Three elements decide whether a chatbot really helps your customers or just talks back politely.

Your own information as the source

The usual approach is called RAG (retrieval-augmented generation): the chatbot first searches your own texts, such as frequently asked questions, terms, price lists and manuals, and bases its answer on them. Extra training of the model on your own data (fine-tuning) is not needed for most small businesses.

Connections to your systems

A chatbot that can look up an order status, see a free slot in the diary or move a booking takes work off your plate. A chatbot that only has text refers people elsewhere. The difference is not in the language model but in the connection to your diary, till, online shop or customer records.

Limits and a way out to a person

Decide what the chatbot must not do: invent policy, promise discounts, make medical or legal statements. And make sure it recognises when it is unsure. “I’m not sure about that, I’ll pass your question to a colleague” is a good answer, provided that colleague actually receives the question.

Why many chatbots get stuck

Chatbots that fail tend to share the same causes:

• No access to real data: the customer asks where their order is and gets a link to the terms and conditions.
• No business context: the answers sound like a generic AI rather than your business.
• No way out: the customer goes round in circles because handing over to a person has not been set up.
• Only tested on easy questions: the chatbot works in the demo, but not on the real questions from your inbox.

Uses that work for small businesses

We see the best results with well-defined tasks that have a clear outcome:

• Booking appointments: a hair salon or yoga studio lets customers pick, move or cancel a free slot through the chat, directly in the diary.
• Status questions: an online shop or garage answers “where is my order?” or “is my car ready?” with current data.
• Intake in advance: an installer has the chatbot collect the address, boiler type and photos of the fault, so the engineer arrives well prepared.
• Frequently asked questions: a hotel or restaurant answers questions about parking, allergens or check-in times, also outside opening hours.
• Internal questions: a staffing agency lets its people look up procedures and agreements without anyone in the office having to search.

More and more customers prefer to ask such questions on WhatsApp rather than in a chat window on your site. That is what the AI assistant on WhatsApp is for; it uses the same knowledge and connections.

Build it yourself, a separate platform or built into your software

There are roughly three routes:

• A separate chatbot platform: quick to set up and fine for a simple question bot. Connections to your own systems are often limited or take extra work.
• Build your own on an AI service: maximum freedom, but you need technical knowledge and you are responsible for maintenance, security and quality yourself.
• Built into your business software: the chatbot already knows your diary, services and customers because it lives in the same system. There are no connections for you to maintain.

We choose the third route. In our software per industry the AI assistant is a module: it answers questions in your tone, helps with bookings and follow-up, and passes doubtful cases to your team.

What to check before you start

1. Collect your questions: take a month of emails, chats and phone notes and count which questions keep coming back.
2. Start with a defined scope: begin with the three to five questions that come up most often and are easiest to check.
3. Sort out privacy: if customer data passes through the chatbot, the GDPR applies. Have a data processing agreement, tell customers they are talking to AI and keep no more than you need.
4. Keep the source up to date: when prices or opening hours change, the chatbot must know straight away; outdated information causes wrong answers.
5. Measure what it delivers: after launch, check how many questions it handles alone, how often it hands over and where customers drop off.

Costs depend mainly on the number of connections, sources and channels; a separate solution adds language model usage that grows with each conversation. At Vorlyo the AI assistant is part of the software: a one-off investment to set everything up for your business, then a fixed monthly licence for hosting, updates and support.