AI sales agent: how it works and how to set it up

An AI sales agent on the levelpartner.ai platform runs the first sales conversation with every new lead: it replies within a minute, asks your qualifying questions, decides whether the lead fits and books qualified prospects straight into a salesperson's calendar. Leads that do not fit get a polite, useful answer instead of a meeting. This article explains how to give the agent a qualification framework it can apply, where it sits in your pipeline, and how to set it up so your salespeople start each day with meetings already in the diary.

Selling versus answering

A support agent answers questions. A sales agent has a goal: move the right leads to a meeting and keep the wrong ones out of the diary. That goal changes how you configure it. You write qualification criteria, not just FAQs, and you decide what the agent does with everyone who does not pass.

For an established firm with stalled revenue, the payoff is focus. Salespeople at firms of this size can lose a large part of the week to calls with people who were never going to buy. An agent that filters before the meeting hands that time back to proposals and follow-up.

Where conversational AI for sales fits in the pipeline

Conversational AI for sales earns its place in one spot: the gap between a prospect raising a hand and a salesperson replying. Picture your pipeline as new lead, contacted, qualified, meeting booked, proposal, won. The agent works the first four stages. It never sends a proposal, negotiates or discusses discounts.

That boundary is what makes it safe to run on its own. A buyer who wants to talk scope, price or risk is passed to a person, and the agent says so plainly rather than improvising an answer.

Writing qualification the agent can apply

Pick three to five criteria, each answerable in one question. Budget band, timing, decision role and fit with your service are the usual four. For each, write what counts as a pass, what counts as "not yet" and what counts as not a fit. Avoid criteria that need judgement the agent cannot make, such as "seems serious".

• Agencies: monthly budget range and whether the lead has worked with an outside team before.
• Consultants: the business problem, company size and who signs off.
• Mortgage brokers: purchase timeline, deposit status and whether an agreement in principle exists.
• Real estate: buying or selling, location, timescale and whether they have a property to sell.

Ask the questions in the order a good salesperson would, one per message. A prospect who receives five questions at once feels processed and often stops replying.

Before you start

• A pipeline in Opportunities with a stage for qualified leads and one for leads not ready yet.
• Your qualification criteria written down, with pass, nurture and not-a-fit answers for each.
• Each salesperson's calendar set up on the platform, with round robin or direct assignment agreed.
• Custom contact fields for every answer the agent collects, so salespeople can read them before the call.
• A nurture route for leads who fail on timing but may buy later.
• Lead sources already arriving on the platform: forms, ad leads, website chat or inbound SMS.
• A short description of what you sell and who it is not for.

How to set it up

Build the agent with the prompt-based route, choosing the appointment booking starting point:

1. Go to AI Agents > Conversation AI and click + Create Bot.
2. Select Prompt Based Bot.
3. Choose Appointment Booking.
4. Continue into the bot editor.
5. Configure Bot Settings.
6. Configure Bot Training.
7. Configure Bot Goals using prompts, model settings and actions.

In Bot Settings and the prompt, write the brief: your offer, your ideal buyer, the people you turn away and the criteria above. In Bot Training, add your service pages and the answers your best salesperson gives to common questions.

Then set the goals that turn each conversation into pipeline data and a booked meeting:

1. In Bot Goals, open Information Collection and select Name, Email and Phone.
2. Add a custom question for each qualifying criterion.
3. For each question, open Update Contact Field and choose the field.
4. Turn on Skip if Already Filled for details you may already hold.
5. Turn on Appointment Booking once the sales calendar is configured.
6. Select Trigger Workflow to start your qualified-lead automation.
7. Set up Email Notification Settings so the sales lead is alerted.
8. Click Save and send sample messages in the test panel.

The workflow the agent triggers is where you move the opportunity to the qualified stage and send unqualified contacts into nurture instead of discarding them. Start the agent in Suggestive mode so a salesperson approves its first replies, then move to Auto-Pilot once the drafts need few corrections.

What changes in the numbers

The first change you will see is speed to first reply. The second is how many qualified meetings each salesperson gets per week. The third is a cleaner pipeline, because answers sit in fields rather than in someone's notebook.

What it replaces

It replaces the discovery call that should have been an email, the shared inbox someone checks twice a day, the spreadsheet of leads to call back and the junior rep whose job is to ring every lead with the same questions. It does not replace the salesperson who runs the real meeting, handles objections and writes the proposal.

Time to run it

Writing the criteria and testing them takes most of a day. Afterwards, plan on about an hour a week: the sales lead reads a sample of conversations, reviews leads the agent rejected and checks that qualified really meant qualified. Once a month, compare booked meetings against won deals. If the agent books meetings that never close, the criteria are too loose.

When not to use it

Very high-value, low-volume sales, such as large consulting engagements, rarely benefit from automated qualification, because the prospect expects a person from the start. Use the agent only to book the first call there. Skip it too if you get a handful of leads a week and someone can reply within minutes. If you want AI to help your salespeople rather than talk to leads, read about the AI sales assistant. For the booking step on its own, see the AI appointment setter.

Troubleshooting

Good leads are rejected. A criterion is too strict or the question is misread. Read the transcript, reword the question and widen the passing answer.

The agent asks everything at once. The instruction to ask one question per message is missing or buried. Put it near the top of the prompt.

Qualified leads do not appear in the pipeline. The triggered workflow is missing or points to the wrong pipeline. Check it and test with a dummy lead.

Prospects book but do not turn up. Add confirmation and reminder messages to the booking calendar, and have the agent restate the purpose of the meeting when it confirms.

The agent sounds pushy. Remove urgency phrases from the prompt and tell it to offer a meeting once, then answer questions.

Salespeople ignore agent-booked meetings. They do not trust the qualification. Show the captured fields on the opportunity and agree the criteria with them.

Frequently asked questions

Can an AI sales agent close deals?

For simple, low-priced services it can share a payment link. For most firms of this size it should qualify and book, and a person should close. Closing needs judgement about scope and price.

What about leads that fail qualification?

Give them something useful: a guide, a smaller offer or a referral. Tag them for a slower nurture, because some will be ready in six months. AI follow up explains how to stay in touch.

Will prospects know they are talking to AI?

They should, so say it in the first message. Buyers care far more about a quick, useful answer than about who typed it. In California it is also the law: a bot used online to encourage a sale must not hide that it is a bot.

Does it work on SMS and the phone?

The same agent can answer web chat, SMS, WhatsApp and social messages from one inbox; AI text messaging covers the SMS side. On the phone, qualification works the same way through a voice agent, and the conversational AI guide explains how voice and text agents relate.

The journey this guide describes