Both are called a "bot", but they work differently, and choosing wrong gets expensive: either customers get annoyed by rigid buttons, or you pay for intelligence where a simple form would have done.
The scripted bot
This is a pre-drawn tree: buttons, branches, ready-made answers. The bot does not "understand"; it walks the person along a route.
- Pros: predictable, cheap to maintain, and it cannot say anything it shouldn't.
- Cons: any question outside the tree stumps it, and the tree keeps growing and is hard to change.
The AI assistant
It understands free text, answers from your materials, and can ask clarifying questions and carry a conversation. You do not need to draw every branch.
- Pros: handles non-standard phrasing, updates quickly along with the knowledge base, and takes more routine off your team.
- Cons: it needs limits and checks, otherwise confident inaccuracies are possible, and quality depends on the materials.
Comparison
| Scripted bot | AI assistant | |
|---|---|---|
| How it communicates | Buttons and fixed branches | Free text |
| What it does with an unexpected question | Gets lost | Answers from the knowledge base or brings in a person |
| How to update it | Redraw the tree | Update the knowledge base |
| Risk of a wrong answer | Almost zero | Present, reduced by limits |
| Best for | Simple choices: booking, order status | Consultations, leads, complex questions |
A combination works best
In practice we most often build a hybrid. Where precision matters (order status, booking a time slot, payment), a clear script runs. Where the customer phrases things their own way (consultation, product selection, clarification), AI steps in. And critical decisions are always confirmed by a person.
How to choose
- Customer questions are repetitive and there aren't many: a scripted bot is enough.
- Phrasing varies and the catalog or terms are large: you need an AI assistant.
- The cost of a mistake is high: a hybrid with human confirmation.
FAQ
Which costs more to maintain: a bot or an assistant?
A scripted bot is simpler, but the tree has to be redone with every change. An assistant is updated through the knowledge base, but needs quality checks. The total cost depends on the task and is calculated in the teardown.
Can an AI assistant make a mistake?
Yes, just like a person. The risk is reduced by restricting answers to the knowledge base, by checks, and by passing complex cases to a manager.