Live chat vs chatbot vs AI concierge: what a B2B site should run

Human live chat is best when you have a person available and few visitors. A scripted chatbot suits simple, repeated questions with fixed answers. An AI concierge, an agent that knows your product, qualifies the visitor and books a meeting, fits a B2B site where questions are varied and arrive at night. Whatever you pick, a person must be able to take over, and the chat should land in the same inbox as your other channels.

The three names are used loosely, and vendors blur them on purpose. This post defines them by what they do, so you can decide by your own traffic and not by a label.

What are the three options?

Human live chat. A widget connects the visitor to a person on your team. It is warm and accurate, and it stops when the person goes home.

A scripted chatbot. A decision tree: the visitor picks a button, the bot shows the next question, and eventually it shows an answer or a form. It works for fixed questions such as opening hours or order status. It breaks when the visitor types something the tree did not expect.

An AI concierge. An agent that writes each reply from the knowledge and goals you give it. It can answer an unscripted question, ask the qualifying questions you set, handle an objection and offer times for a meeting. It also needs rules, because a model that answers anything will sometimes answer wrongly.

How do they compare?

Where does each one break?

• Human live chat breaks on time. A B2B site gets visitors in other time zones, at night and on weekends. If an inquiry waits, the person has usually moved on. On our own LinkedIn outbound, 48% of prospect messages arrived at night or on weekends, and 74% of those were answered before the next morning. A website is no different.
• A scripted chatbot breaks on variety. Every new question is a new branch to write, and the visitor who does not fit a branch leaves. Teams often stop maintaining the script after a few months.
• An AI concierge breaks on knowledge and rules. If the product knowledge is thin, it guesses. If it has no rule for what it may not say, such as discounts or legal promises, it can say too much. That is a setup problem, and it is why the knowledge base and the rules matter more than the model.

What does an AI concierge do on a B2B site?

In Sellinger, the live chat widget is installed with one script tag, and a Shopify connection is available. The agent answers product questions, qualifies visitors against your goals and books meetings. Two modes exist: Sales Mode for new visitors and Support Mode for existing customers. Returning visitors are recognised by cookie, and every chat is stored in the same inbox as the LinkedIn, email and WhatsApp threads for that person, so a later follow up knows what was said in the chat.

When the AI cannot resolve something, it can pass the conversation to your team with the visitor's history, pages viewed and the transcript. The live chat use case page has the steps, and plans start at $19 a month with a 30 day free trial and no card.

What does a fast answer look like in practice?

We have no case study of a B2B website widget with published results, and we do not claim any. The closest evidence is our own hotel. At Hotel Pirin the reservations agent answered guest inquiries on email and Messenger with a median first reply of 41 seconds over 54 inquiries, and 122 of 123 agent answers reached the guest within two minutes. The agent did not take bookings. People did. That is a fair picture of the division of labour: the agent answers fast and asks for what a person needs, and the person closes.

See the Hotel Pirin case study for the whole story, including that the hotel is owned by a Sellinger co founder.

How should you choose?

Answer four questions.

1. When do your visitors arrive? If most arrive during your working hours and you have someone free, human chat may be enough. If many arrive outside, you need something that answers while you are away.
2. How varied are the questions? If 80% are the same five, a scripted bot is cheap and fine. If every visitor asks something different, use an agent.
3. What is a lost visitor worth? For a B2B site with a high value deal, an unanswered question has a real cost. For a low value site, it may not.
4. Who takes over? Pick a person, and decide when. Write it down: pricing beyond the published plans, legal questions, anything the agent flags as unclear.

You can also combine them. Many sites run the agent after hours and a human during the day, with the same transcript in both cases.

What does each look like on a quiet Saturday night?

This is a made up scenario, not a customer. A visitor from another time zone lands on a B2B pricing page and types: "Do you integrate with HubSpot, and can I try it without a card?"

• Human live chat. Nobody is online. The widget collects an email address and the visitor gets a reply on Monday, if they are still interested.
• Scripted chatbot. The bot shows buttons: Pricing, Demo, Support. The question is about an integration and a trial in one sentence, so no branch fits. The visitor is sent to a contact form.
• AI concierge. It answers from the knowledge base: HubSpot is a live two way integration, and messaging modules have a 30 day free trial with no card. It then asks one qualifying question and offers times for a call from the connected calendar.

The difference is not cleverness. It is that the third option can answer a question nobody scripted, with facts you gave it, while your team is asleep. The first option is still the best choice when a person is there, which is why many sites combine them.

What should you ask a vendor?

• Can the visitor reach a person, and does that person see the whole transcript?
• Where do the chats go? Is it the same inbox as your other channels?
• What does the agent say when it does not know?
• Can I set a list of things it must never say?
• Can it offer meeting times from my calendar?
• How do I read every conversation, not a sample?
• What does it cost at my traffic, and what is the trial?

What should you set up first?

• A short opening message. Something specific and honest, not "How can I help?". There is a live chat opening message template.
• The facts. Prices, policies, opening hours and objection answers in the knowledge base, so the agent never guesses.
• A handoff rule. Which questions go to a person and how they are notified.
• A review habit. Read the first thirty chats. You will find the questions you did not think of.

A checklist

• [ ] I know when my visitors arrive and who answers then.
• [ ] I chose by traffic and question variety, not by label.
• [ ] The agent has real product facts and a written list of things it must not say.
• [ ] A person can take over and gets the transcript.
• [ ] Chats land in the same inbox as my other channels.
• [ ] I read the first thirty conversations.