What is the difference between a chatbot and an AI agent that completes transactions?

The short answer. A chatbot answers a question; an AI agent that completes transactions finishes the job in your systems, or stops and hands it to a person. Helm builds the second kind: at Care Clinics in Cairo, 86 bookings taken on WhatsApp reached reception's sheet with 0 mismatches across the 516 booking fields checked.

Helm's FAQ puts it this way:

A chatbot answers a question. A Helm workflow finishes the job — the slot is held in the real calendar, the deposit is taken, the lead record exists in the CRM with the answer attached — or it stops and hands the work to a person.

What does a chatbot actually do?

It answers. You upload an FAQ or design flows, and the assistant replies from them. The conversation ends at an answer, and whatever has to happen next is left to your team.

A fast answer is not the same as a finished job. Helm counted a bakery chain whose page sent 648 of 703 replies in under five seconds, from an autoresponder with four canned lines. The 55 replies typed by hand took a median 46 minutes. Under one launch post from a national real estate developer, dozens of people commented «details». Most got the same reply, and none of the replies carried a fact.

What does an AI agent finish that a chatbot does not?

The work the reply was about. In a Helm workflow, finished means one of these:

1. The slot is held in the real calendar, from availability the agent can see, and the booking is written where the team works.
2. The deposit is taken, by a link in the thread, under a rule agreed in scoping.
3. The lead record exists in the CRM, with the answer attached.
4. The claim is filed, documented and assigned to the team that owns it.
5. The exception goes to a named person, with full context and a clear next step.

The conversation is the interface. Finishing the job is the product.

How is it different from a helpdesk?

A helpdesk organises work for people to do. A support desk routes a booking request to a person; it does not hold the slot. Helm does the work, inside the channel the customer already used, and escalates by design rather than by default.

Is an agent that acts riskier than a chatbot?

It would be, without rules. Because an agent takes actions, what it may decide alone is set per workflow, in writing, before the build, on three levels: let it act, ask for approval, or bring in a person. The limits are concrete. A clinic agent never confirms a booking it cannot see in the calendar. A dealership agent never computes an instalment. When the agent does not know, it stops: silence and invention are both treated as failures.

How do you tell which one you need?

Ask whether the conversation ends in an answer or in an action. If it ends in an answer, a builder may be enough. If it ends in an action in a system of record, you need the action built and maintained.

Then measure the outcome, not the activity. Replies sent and messages handled are activity. Each of Helm's industry pages publishes one number before the build: booked and attended for a clinic, showroom visits booked for a dealership, applications through to enrolment for a school. The Care Clinics case study shows what a counted outcome looks like.

When is a chatbot the right answer?

When every question ends with «here is our website». A business with a few dozen messages a week, and someone with time to build and maintain flows, is well served by a self-serve chatbot builder. Helm fits high-volume, repeatable work that ends in a transaction, where a wrong answer has a cost. The comparison of the four options and the FAQ go further.