“We want a chatbot on the website” is one of the most common requests we hear. Behind the same phrase, though, there are very different needs, with different costs and different results. If you don’t clarify from the start what you actually need, it is easy to pay for something that does not solve your problem.
Three things we call a “chatbot”
1. The simple FAQ assistant. It answers a limited set of routine questions: opening hours, payment methods, shipping times. The answers are more or less predefined. It is cheap and reliable, but it knows nothing beyond what you have given it.
2. The knowledge assistant. It answers from your own documents and data: catalogues, manuals, procedures, terms, technical sheets. It finds the relevant part and phrases the answer, ideally citing the source so that someone can check it.
3. The assistant that acts. It does not only answer, it does things: checks an order’s status, books an appointment, opens a support ticket, updates a system. To do that, it has to be connected to your systems, with proper access rights and logging.
When the simple one is enough
If the questions you receive are few and repetitive, a simple assistant, or even a good FAQ page, does the job. There is no need to pay for something more complex.
When you need a knowledge assistant
When the information customers or partners look for is spread across many documents and changes often. In that case we often suggest starting internally: your own staff use it first, where a mistake is easy to spot, and it opens to customers only once it has proved its value.
When it needs to connect to your systems
When the answer depends on live data: “where is my order?”, “do you have this size in stock?”, “is there a free slot on Thursday?”. Here the assistant has to read from your online store, ERP or booking system. It is the most useful form, and also the one that needs the most careful design.
Mistakes we often see
- An assistant without the right data. A language model with no access to your information will give answers that sound right but are not.
- No way out to a human. A customer who doesn’t get an answer must be able to talk to someone, without retyping the whole problem.
- No measurement. If you don’t know how many questions were answered correctly and how many ended up with a person, you don’t know whether the assistant is worth the money.
- Customer-facing first. Your customers’ first experience is not the best place to test something new.
How to decide
Answer four questions:
- What are the 20 most frequent questions you receive, and how many can be answered without live data?
- Where does the information for the answers live today, and how often does it change?
- Who will review the answers and keep the content up to date?
- How will you measure whether it is worth it: fewer phone calls, faster answers, less staff time?
Your answers almost always show which of the three forms you need, and whether you need a chatbot at all or something else. If you would like to work through it together, we always start with this analysis, before any technology. See how we approach the move to AI.