Business: Sales · Lesson N.vnd.2

Connecting AI to your sales context

Generic AI gives you a generic email; with your sales context (CRM, ICP, history, and proposals that closed) it prepares material that looks handmade. This lesson shows what to connect and how to do it safely and within LGPD.

Examples for

You ask AI to write an important email and it hands back a polite, forgettable text, the kind that fits any situation and convinces no one. The problem isn't that the model is weak, it's that you asked a cook who's never seen your kitchen for a finished dish: without the conversation history, without knowing who's on the other side, without your way of writing, all it can do is guess at the average. The person who receives it smells the robot and ignores it. The moment you connect what already exists, the history, the context, your tone, the same model jumps up a level and the text starts sounding like it was written calmly, by you.

Let me tell you something that bothers me. You open AI, ask for a sales email, and get back that lukewarm, faceless text. Then you think: "this machine doesn't understand my business". And you're right, it doesn't understand. But the problem isn't that it's dumb. It's that you're asking a gourmet dish from a cook who's never set foot in your kitchen, never seen your pantry, never tasted what your customers like. Give the same cook the context and they become a chef.

The core idea of this lesson. Sales with AI without context is spam; sales with AI with context is relevance at scale. The generic model gives you a generic email. When you connect your sales context (the CRM, the ICP, the account history, the proposals that closed, your way of selling), AI stops guessing and starts preparing material that looks handmade. The gain isn't writing faster, it's writing right, for the right person, in the right tone, without turning into a manual.

01Why generic AI gives you generic output

Remember lesson 2.2, about rules and context? The logic here is the same, just applied to sales. AI knows nothing about your world until you tell it. It doesn't know your customer, doesn't know who closes deals with you, has no idea how you write when you want to win an account. Without this, it does the only thing it can: it takes the average of everything it's seen on the internet and hands you the safest, blandest text possible.

Think of an industrial kitchen. The cook is excellent, sharp technique, fast. But if they've never seen your menu, never tasted your dishes, they'll cook by the average: rice, beans, steak, fries. Correct, edible, forgettable. What turns that same cook into your house's chef isn't more talent, it's the context of your kitchen: the pantry, the recipes that worked, the palate of whoever sits at your table.

Sales context is exactly that pantry. And it already exists, scattered through your CRM, your emails, the proposals stored in some folder. The work isn't creating context from scratch, it's connecting what you already have.

02What to connect: the four pantries

There are four sources that turn to gold when you connect them to AI. You don't need to connect all of them at once, but the more open pantries, the more the dish comes out with your face on it.

The first is the ICP and personas: who the right customer is for you, what pain they have, what role they hold, what makes them buy. Without this, AI talks to everyone, which is the same as talking to no one. The second is the account's interaction history: what's already been discussed, what's already been proposed, where the negotiation stalled. That's what pulls the email out of "Dear customer" and into "regarding what we agreed in April". The third is the proposals that closed: the texts, the arguments, and the offers that actually turned into contracts. That's your winning sales playbook, and AI can learn the pattern of what works with you. The fourth is the company's tone: your way of writing, formal or direct, the house's vocabulary. It's what makes the material sound like you, not like just any site.

ICP and personas account history proposals that closed company tone AI material with your face

03The caution: customer data and LGPD

Now hold your excitement for a second, because this is where the risk lives. Customer data isn't yours to go spreading around. When you connect the CRM to AI, you're handling real people's names, phone numbers, emails, negotiations, and sometimes sensitive information. And that has an owner and a law: LGPD exists precisely to say what can and can't be done.

The line is simple to understand, even if it takes work to maintain. What's allowed: using your internal context to prepare your material, inside tools you control that respect a data-processing agreement. What's not allowed: dumping a prospect's sensitive information into an open tool without knowing where it goes, exposing one customer's data to serve another, or processing personal data without the legal basis and consent the operation requires. Think of sales context as a locked drawer: you use what's inside, but you don't leave the drawer open in the hallway.

The practical rule for day to day: before pasting anything into AI, ask "if this data leaked, would I have a problem with the customer or with the law?". If the answer is yes, it doesn't go out without a secure channel. Minimizing is the name of the game: bring AI only the context necessary for the task, not one field more.

Can your internal context, on a secure channel only the data necessary for the task legal basis and consent in order Can't sensitive data in an open tool exposing one customer to serve another processing personal data without legal basis

04Connecting safely

The good news is you don't have to choose between relevance and security. You can have both, and the way to have both is connecting through a governed path. Remember what you saw in The Stack and in the AI Gateway, and in lesson 4.4 on governance? It's exactly that applied to sales: instead of every salesperson pasting customer data into any tool, context passes through a single, controlled point, which decides what goes in, what goes out, and keeps a record of everything.

Think of a company's front desk. People don't come in through the window: they pass reception, identify themselves, and there's a record of who came in and at what time. The gateway is that front desk for your sales data. It lets AI use context without the information going out uncontrolled, and it gives you visibility into where each piece of data ended up. Security here isn't the brake that stops you from using AI in sales; it's what lets you use it with peace of mind.

And there's an economic detail that closes the reasoning. Without this governed path, the risk of a leak or an LGPD fine is a hidden cost that only shows up when it explodes. With the governed path, that cost becomes predictable and small. You trade a big, uncertain risk for a controlled, cheap process. It's the same logic as any insurance: pay a little to avoid paying a lot all at once.

05The map: which memory for your sales data

There's a technical decision that changes the outcome, and it shows up in the "Which memory for your data" map. Not every piece of sales data connects the same way. A loose document, like an old email or a proposal PDF, you throw into a similarity search and AI retrieves the matching passage. That's RAG, great for running text.

But the CRM is a different animal. CRM is structured data: fields, values, statuses, dates, funnel stage. When you ask "which accounts have been stuck for more than thirty days at the proposal stage with a ticket above X?", you don't want a similar-sounding passage, you want reasoning over the data, a filter, a count. Treating structured CRM data as if it were loose text is like asking the cook to guess the recipe from the smell instead of reading the spec sheet. It goes wrong.

The practical rule: for text and documents, similarity search does the job. For the CRM, you want AI reasoning over structured data, consulting the right field, not just retrieving a similar-sounding paragraph. Knowing this difference is what separates whoever really connects sales with AI from whoever just makes AI parrot old emails.

Do it now

Do it yourself

Take a real, open account from your pipeline (your real task works well) and build, on paper or in a notes app, the context package you'd give AI to prepare that account's next email. Do it in three blocks:

  1. THE FOUR PANTRIES: list what you have from each source for that account. ICP/persona (who they are, what pain), history (what's already been discussed and where it stopped), one of your proposals that closed in a similar situation, and a sentence describing your company's tone.
  1. THE LGPD FILTER: look at block 1's list and mark what's personal or sensitive customer data. For each marked item, decide: goes out through a secure channel, goes out anonymized, or doesn't go out. If you don't have a secure channel today, note that as a gap to fix.
  1. THE GENERIC TEST: write in one sentence how the email would look WITHOUT that context (the generic one) and in one sentence what changes WITH it. If the two sentences are the same, you haven't really connected context yet; go back to block 1 and open one more pantry.

Practice

1. Why does AI, without your sales context, tend to give back a generic, faceless email?

2. You're going to connect the CRM to AI to prepare sales materials. Which posture aligns with LGPD and customer data security?

3. When connecting the CRM, why does treating it as structured data make a difference compared to treating it as a loose document?

Fair enough? The takeaway is direct: AI isn't generic by nature, it's generic for lack of context. The moment you open the right pantries, the ICP, the history, the proposals that closed, and your tone, the material starts to look handmade, at scale. And you do this without giving up security: context through a governed channel, customer data handled with LGPD's lock, CRM read as structured data and not as an old-email parrot. Sales with AI without context is noise; with context, it's relevance that sells. Now it's time to open the first pantry.

For the board

On the generic outputwith no pantry, the cook cooks by the average. The problem is not the model, it is what you did not give it.
On the pantriescustomer profile, history, past proposals and tone. With all four, the same model becomes a chef.
On the CRMcustomer data calls for a legal basis, a secure channel and the bare minimum. A drawer with a lock, not one left open in the corridor.
What did you think of this page?
Would you recommend this page to someone on your team?