A chatbot runs a script written in advance; an AI agent reads the question, understands it and pulls the answer from your real data. The first one handles the questions you anticipated, the second handles the ones you did not, which is most of them. The gap shows up directly in how many chats customers abandon.

What separates a chatbot from an AI agent

A classic chatbot runs on three mechanisms: predefined keywords, buttons to tap and a fixed decision tree. As long as the customer types “price” or “delivery”, the machine works fine. The moment they phrase it their own way, something like “how much is that red thing?”, the chatbot falls off its tree and replies that it did not understand.

An AI agent runs on a modern language model instead. It handles a sentence full of typos, slang or two languages mixed together, it follows a thread across several turns, and it holds on to context as it accumulates: a dress, size M, in blue, delivered on Friday. None of those three behaviours can be bolted onto a decision tree by adding more branches.

The split between a chatbot and an AI agent is therefore about nature, not about product generation. A chatbot draws from a list of replies you wrote yourself; an AI agent composes a reply from your data. The first has a ceiling you can predict on day one, the second improves as you widen what it is allowed to consult.

Why a scripted chatbot drives customers away

A scripted chatbot drives customers away because it asks them to restate their need in its own vocabulary. The decision trees deployed between 2018 and 2022 trained people to expect exactly that: around half of customers abandon the conversation within the first two messages, before they have even asked their real question.

The measured gap between the two approaches is wide. Drop-off sits between 40% and 60% with a classic chatbot, against 5% to 15% with an AI agent. Human handover covers 60% of chats on the chatbot side, against 15% to 20% on the agent side. The first number is lost sales, the second is your team’s time.

Setup cost points the same way. A chatbot is paid for in scripts: every possible case has to be imagined, written down, then maintained every time the catalogue changes. An AI agent is paid for in data: you give it the catalogue, the delivery times and the prices, and case coverage follows without rewriting a single dialogue.

What an AI agent does that a chatbot never will

An AI agent takes action where a chatbot can only send a holding message. It reads a screenshot sent by a customer and finds the matching product in your catalogue, it checks availability before promising a delivery date, and it offers a credible alternative when the requested item is out of stock. Each of those calls for a judgement, not a keyword match.

  • read a photo or screenshot sent by the customer and identify the product
  • negotiate within a price range you have set in advance
  • offer an equivalent item when the requested size is out of stock
  • spot a complaint and route it to a human straight away
  • draw on past conversations to sharpen its next answers

Those five capabilities are what separate an AI agent from a chatbot day to day. Recognising a product in a blurry photo comes from reading your actual catalogue rather than from a keyword. Mistakes, meanwhile, get corrected in plain language, the way you would coach a salesperson after a chat went badly, with nothing to reconfigure.

A useful AI agent knows your shop, not just the language

A generic AI agent answers like a consumer assistant: politely, and beside the point. What turns an agent into a seller is access to your operating data, meaning the catalogue with its prices, stock levels and variants, your delivery zones and lead times, the payment methods you accept, your current promotion rules and the history of your customer chats.

Bedones Moderator is built on that principle: the agent reads your catalogue instead of a script, so a price change flows into its answers with no further intervention. A customer asking “do you have yesterday’s dress in blue?” gets the photo, the price and the availability in one message, without tapping through a menu or waiting until morning.

That reliance on data is also the limit worth watching on any AI agent. An agent wired to a wrong catalogue answers wrongly, with exactly the same confidence as an agent wired to a correct one. How fresh your stock levels, prices and lead times are therefore decides, directly, the quality of what your customers read in the chat.

Keeping a human in the loop and measuring the result

An AI agent does not replace your team; it absorbs the repetitive questions, roughly 70% of incoming volume, so that people are free for the work that pays. Large baskets, loyal customers and sensitive complaints stay human conversations, and a properly configured agent hands over as soon as it leaves its area of competence.

That split is also what makes round-the-clock cover with a small team sustainable: the agent works nights and weekends, and the team picks up the flagged cases in the morning. The handover rule behaves like a dial, and the stricter you set it, the closer your human handover rate creeps back to a chatbot’s.

The deciding test takes two weeks. Route 20% of your incoming chats to the AI agent, leave the other 80% untouched, then compare conversion rate, first-response time and the share of chats a human had to take over. By the end of the fortnight, the choice between a chatbot and an AI agent is no longer a matter of opinion.