Chatbot or AI assistant: which one does your business actually need?

A chatbot follows the tree you drew: an assistant reads the question you actually asked, and the honest choice starts with how varied your inbox is.

Vendors use the two words as if they were synonyms, and buyers pay for one while expecting the other. A company installs what the sales page called an assistant and discovers a phone menu wearing a chat window: pre-arranged buttons, canned replies, and a polite refusal to understand anything phrased off-script. The reverse disappointment is rarer but worse: a buyer who needed a predictable script receives an open-ended system and does not know what to do with it. The confusion is expensive in both directions, and it starts with vocabulary.

A scripted chatbot, described honestly, is a decision tree with a friendly face. Its answers are written in advance; its paths were drawn by someone who tried to guess the questions. Within its tree it is excellent: fast, cheap, predictable, incapable of embarrassing the brand because it never says anything new. The moment a question falls outside the branches it breaks, and users feel the exit ramps: the repeated I did not understand, the loop back to the menu, the handover to an inbox nobody watches.

An AI assistant is a different machine. It reads the question the customer actually asked rather than the one the flowchart anticipated, it answers from your own documents and rules, it holds a conversation across languages and half-sentences, and it carries phrasing nobody predicted. That capability has a price beyond licensing: it needs grounding so it does not improvise, guardrails so it never makes final promises, escalation to a human for what it should not hold, and honesty about being a machine. The assistant is stronger and more demanding; that is the whole trade.

Which one a business needs is not a fashion question; it is a traffic question. Look at the questions your customers actually send and count two things: how many arrive, and how differently they arrive. A few questions, asked thousands of times in the same words, are script territory, and buying intelligence for them is waste. A widening variety, asked in mixed languages and unexpected shapes, kills scripts fast, because every new branch you draw is one more thing to maintain and the tree is never finished.

The commonest failure we see is a mismatch bought with confidence: the company with narrow, repeated questions paying assistant prices to answer the same question in the same way, and the company with wide, varied conversations jamming them into a script and watching clients abandon the chat. Both bought a category instead of diagnosing a case. The diagnosis costs an afternoon of reading real customer messages; the wrong purchase costs a year of pretending it works.

Where we sit: we run an assistant on our own funnel, grounded in our own documents, escalating to named humans, and saying plainly that it is AI, so our position is a description of our own arrangement rather than a vendor's promise. We will tell a client just as readily that their case is a script's case and they should not pay for more. Guidance, not guarantees: the category names are the vendor's words; the questions in your inbox are yours.