Chatbot
What is a chatbot and why do most of them irritate?
A chatbot follows a script. It works as long as the customer stays inside that script, and customers rarely stay inside a script.
In short
A chatbot is software that runs a conversation according to predefined rules: the customer picks from buttons or types something matched on a keyword, and the system returns the answer belonging to that branch. The boundary with an AI agent lies in what happens with a question not in the script. A chatbot falls back to a default line, a language model still tries to answer.
Why button menus so often miss
A chatbot with buttons forces the customer to translate their question into your categories. Someone with a damaged parcel sees Returns, Delivery and Payment and has to guess which of the three fits. Pick wrong and you get a branch that doesn't apply, then they start over or leave.
That's the source of the irritation people associate with chatbots. It's rarely that the answer is wrong; it's that the customer notices they're not allowed to ask their question the way they have it.
Where a script is the right tool
For an action with fixed steps a script beats a language model. Registering a return needs an order number, a reason and a confirmation, in that order. You don't want creativity there, you want the same thing to happen every time with correct data.
The split that works in practice: let the language model understand the question and let the script carry out the action. The customer types in their own words, and once it's clear this is a return, the fixed sequence of steps starts.
What a chatbot needs from you before it returns anything
A script is only as good as its maintenance. Change your return period and the old answer stays until someone updates it. With a small number of branches that's manageable; with a few dozen there's always a branch with stale information within six months.
So agree beforehand who maintains the script and when. With a chatbot that's not a side issue: it's the only reason a chatbot still holds up after a year.
Chatbot or AI agent?
| Chatbot | AI agent | |
|---|---|---|
| Understands | Keywords and buttons | The question in the customer's own words |
| On an unknown question | Default line or handover | Attempts an answer from sources |
| Maintenance | Update each branch separately | Update the source, the answer follows |
| Predictability | High, it says what's in it | Lower, so boundaries and review are needed |
| Risk | Customer gets stuck and drops off | An answer right in tone but wrong in substance |
For fixed actions with a mandatory order a script is better. For questions customers ask in their own words, a language model with access to your own sources is better. Most working setups use both, with the model at the front and the script for the action.
Where it goes wrong
Pretending a person is there
A bot with a first name and a photo that only admits after four turns that it's software costs you trust at exactly the wrong moment. Say it at the start; customers rarely object when the answer is right.
No route to a person
A conversation where the customer hears the same thing four times with no escape is why people say the word chatbot with reluctance. Put a visible route to an agent in every step.
Never rereading the script
Set a six-monthly check on every branch containing an amount, a period or an opening time. Those are the branches that go stale without anyone noticing.
Only measuring how many conversations the bot handled
A bot that ends the conversation hasn't necessarily helped. Measure how many customers then email or call anyway about the same topic, because that's the number that says whether it works.
What to do this week
Run ten conversations with your own chatbot yourself, using questions the way customers ask them. Take those questions verbatim from your mailbox, including typos and half sentences. Count how often the bot misses. Where it does, check whether it's the script or the fact that customers are asking something you've never written down anywhere.
What this looks like in a system
In Cuego the agent is a language model reading your own sources: knowledge base, product information and the order data of the customer currently in conversation. For actions with fixed steps, such as registering a return or changing an address, it starts a workflow with a defined sequence. You set per topic what it may do alone and where it hands over, and you can see how often agents adjust an answer before you move that line.
AI customer service vs chatbotRelated terms
AI agent
An AI agent understands a question in plain language, looks up information in your systems and carries out the action behind it. That's something other than a chatbot with choice buttons.
Read onWorkflow
A workflow is a sequence of steps that runs by itself once something happens. The skill is choosing which steps deserve it.
Read onSelf-service
Self-service is letting the customer arrange themselves what would otherwise be a conversation. It only works when it takes less effort than firing off an email.
Read on