An AI map for sales. What to set up first and in what order

When you bring AI into sales, the order of steps is more important than the tool you buy. An agent without written rules makes up prices. An agent that cannot pass the client to a person loses the client. So first the base, then the things that bring leads, and only then the things that make salespeople stronger.

This is the second note of the course. The first one shows how my system with agents in the terminal works. It also explains why a chat in the browser does not automate anything. This note is the map. Where AI already works in sales, what to set up in each place, and how to start tomorrow.

My examples come from car dealerships, because I have worked in the car business for 11 years. But the order is the same in any sales team that gets leads and calls.

One rule for the whole course#

First, the AI must know the facts and the rules. Then it can talk to clients. Only when it talks to clients without mistakes can it work with sales calls, trade-in prices and the manager's day plan.

If you change this order, you get a common story. A dealership buys a voice agent. The agent gives yesterday's price. The client comes to the showroom angry. A month later the team remembers AI as the thing that broke everything.

The map. Where to use AI and what to measure#

Each row will become a separate note with steps, ready prompts and one number to track.

The first two weeks are only the base#

Write two pages with prices, stock, offers and five rules for what the agent never does. Put a date and a name on each rule. Add one line that stops made-up answers. "No fact in the base? Tell the client you will check."

Choose a person on duty at once. This is a real person who answers the client when the agent passes the chat or call. In working hours this person answers in 5 minutes.

At this step I give the AI read-only access to email and CRM. The agent can see, but it cannot send or delete anything. In my own system, all scheduled runs still work this way.

Do not go further without these three things.

Weeks 3 to 6. The things that bring leads#

There are three steps here, one after another. A voice agent for incoming calls after working hours. Fast callbacks on fresh leads, while the client still remembers the request. A chat agent on the website and in one messenger, not in all of them at once.

Why start here? From what I see, about 40% of requests to a dealership come after working hours. Today nobody answers them. Here AI does not break anything. It fills an empty place.

Watch one number every week. How many agent bookings came to a visit. Not how many calls the agent took, but how many people came.

Weeks 7 to 10. Salespeople get stronger#

When leads come in, start call analysis. First the AI finds the phrases after which clients leave. Then it checks all calls against one list of questions, and the sales manager reviews them with the team once a week.

Lost calls become scenarios for the trainer. Every day a salesperson spends five minutes on price talks with an AI that does not give up before the fifth reply. Here you also add follow-up emails that the AI writes in the salesperson's voice from a short voice message.

Weeks 11 to 13. Money around the deal#

Now add trade-in estimates from photos, accessories before delivery, and a list of service clients who may want a new car. The manager gets a day plan that the AI makes from email and CRM. Give rights above read-only one at a time and watch what the agent does with them.

Once a month, ask ChatGPT and two other assistants five questions that a buyer asks. Write down if they name your dealership and if they say anything wrong about it.

What I would not start with#

• Calling the whole client base with an ad. Complaints and legal questions come before visits.
• An agent that names a trade-in price or a discount by itself.
• Any automatic sending without a human check in the first two weeks.

If one of these already works, pause it and build the base first.

If you start tomorrow, one day is enough#

1. Write two pages with prices, stock, offers and five things the agent must never do.
2. Make a project in ChatGPT or Claude and put this file there.
3. Ask the project ten real questions from yesterday's leads and mark every mistake.
4. Fix the instructions and repeat.
5. Name the person who answers the client when the agent passes the conversation.

The next note of the course is about this first step. How to write the instructions and build the knowledge base that any sales agent works on.

Questions and answers#

Where do I start with AI in sales?#

With two pages of instructions and a fact base. Put prices, stock, offers and the rules the agent never breaks there. Without this, any agent starts to make things up.

How long does it take?#

The plan in this note is for 90 days. You see the first results with leads in weeks 3 to 6, when the voice agent for incoming calls and fast callbacks work.

Can I start with calls to the whole client base?#

I would not. A mass call campaign with an ad brings complaints and legal questions before it brings visits. Start with callbacks on fresh leads only.

Do I need a programmer?#

No. You can build everything in this plan from ready tools and a project in ChatGPT or Claude. I explain how I work without a programmer in the first note of the course.