# AI chatbot vs AI sales agent: look at the workflow, not the label

"Chatbot" and "AI agent" are used differently by different vendors. The name alone does not tell you whether a product can capture leads, book appointments or contact customers later.

For a business website, the useful question is simpler: what happens after a visitor asks something?

## Start with answer quality

A rule-based bot can be effective for a narrow set of predictable questions. A knowledge-based assistant can handle more natural wording, but still needs accurate source information.

Test your own content. Ask about two similarly named services, a price with conditions and a detail that is not published. The assistant should distinguish them and explain missing information rather than inventing an answer.

## Follow the enquiry beyond the answer

Check whether the product can:

- Ask a relevant clarification.
- Capture contact details when the visitor wants a next step.
- Keep the original question and conversation with the lead.
- Let staff manage ownership and status.
- Provide a clear human handoff.
- Retain source information from tagged campaigns.

A product called a chatbot may offer these features. A product called an agent may not. Evaluate the working flow rather than assuming a category-wide checklist.

## An evaluation scorecard

Take this into every demo and ask the vendor to demonstrate rather than describe. The right-hand column is what a satisfactory answer looks like — anything vaguer deserves a follow-up question.

**Answers**

| Ask them to show | A good answer |
|---|---|
| A price that has conditions attached | The conditions appear with the price, not separately |
| Two similarly named products or services | They are distinguished without you clarifying |
| Something absent from the content | It says what needs checking; it does not improvise |
| A one-word message like "cost?" | It asks a useful clarifying question |
| The same question in a second language or mixed script | Either it handles it, or it says plainly that it does not |

**Capture**

| Ask them to show | A good answer |
|---|---|
| The lead record after a test conversation | Question, context, contact detail and source all present |
| What happens when someone declines to share details | The conversation still ends helpfully |
| Campaign source from a tagged link | It arrives on the record automatically |
| Duplicate enquiries from one person | You can see it is the same person |

**Actions — the section where claims usually inflate**

| Ask them to show | A good answer |
|---|---|
| "Book a demo" or "book an appointment" | They demonstrate whether a live slot is reserved or a request is captured |
| Order or account lookup | An actual working integration, or an honest no |
| Follow-up messaging | Whether messages reach *your team* or *your customer* |
| WhatsApp | Which of the four possible routes this is |

**Failure behaviour**

| Ask them to show | A good answer |
|---|---|
| The knowledge source being wrong or outdated | It does not confidently repeat the error as fact |
| A request outside its capability | It declines clearly and offers a real next step |
| A visitor becoming frustrated or abusive | It stays useful and offers a human |

If a vendor answers the actions section with "yes" four times without showing you anything, that is your result.

## Separate requests from completed actions

An assistant saying "I can request a demo" is not the same as reserving a calendar slot. Collecting an order question is not looking up an order.

Ask the vendor to demonstrate the actual integration and confirmation path for any action you need. Test failures as well as successes.

Similarly, team reminders are not automatic outbound customer follow-ups, and a WhatsApp link is not a native WhatsApp AI responder.

## What Serviti provides

Serviti is an [AI chatbot for business websites](/ai-chatbot-for-website) that answers from approved business content and captures enquiries for the lead CRM. You can embed it, share its hosted link or use a QR code.

Growth adds human WhatsApp handoff and real-time lead alerts. Scale adds Lead Intelligence and supported integrations. AI HelpDesk is a separate add-on for support requests.

Native WhatsApp AI Agent and Instagram DM AI Agent are separate paid add-ons for Growth and Scale, excluded from the trial. They answer incoming messages inside your connected WhatsApp Business number or Instagram account using approved business content. These native replies are different from WhatsApp alerts to your team, human handoff and links that open a website chat. Account connection and messaging requirements apply.

Serviti can capture a preferred visit or callback time for staff. It does not claim automatic live booking or unsolicited outbound customer-nurture sequences.

## Run a small acceptance test

Use genuine customer questions before publishing the widget. Include an incomplete message, a follow-up referring to an earlier question and a request outside the assistant's capabilities.

Check what the customer sees and what the team receives. Verify the captured details rather than judging only the quality of the prose.

## Choose for the job you actually have

If your problem is missed enquiries, focus on useful answers, reliable capture and clear staff follow-up. More autonomous features are not automatically better.

A useful way to decide: write down the last ten enquiries you lost and why. If most were lost because nobody replied in time, you need answers and capture. If most were lost because the reply was wrong, you need better source content, not more automation. If most were lost after a good conversation, your gap is follow-up and ownership — and no widget fixes that.

## Common questions

**Is an "AI agent" more capable than an "AI chatbot"?**
Not reliably. The terms are marketing labels applied inconsistently across the industry. Some products called chatbots capture leads, manage ownership and hand off to staff; some products called agents only answer questions. Judge the demonstrated workflow.

**How do we know if it will make things up?**
Test it against a gap. Ask about something deliberately absent from your content and see whether it says the information needs checking or produces a confident-sounding answer. Do this before launch and again after every significant content change.

**Does it need to book appointments to be worth having?**
For most small businesses, no. Capturing a preferred time for staff to confirm solves the same problem with far less that can go wrong. Live booking only pays off when you have a reliable calendar system and enough volume that manual confirmation is the bottleneck.

**Will it replace someone on our team?**
It should not be bought on that basis. It handles the repetitive always-on part — known questions, capture, context — so staff spend their time on conversations needing judgement. If it is sold to you as a headcount replacement, ask what happens to the enquiries it cannot handle.

**What is the most common mistake in choosing one?**
Buying on demo quality. A demo is a controlled conversation with clean content. Insist on testing with your own content, your own awkward questions, and your own phone.

Read next: [where enquiries go missing](/blog/missed-customer-enquiries-cost-small-business) and [a follow-up routine your team can maintain](/blog/automated-lead-follow-up-small-business).