---
title: "AI for Sales"
date: 2026-10-20
draft: false
tags: [ai-training, sales, team-learning]
cover: cover.webp
coverAlt: "A young woman with a high ponytail in a cyan blazer sits alone at a wooden desk, checking a printed sheet line by line with a ruler against an open ring binder, pen in hand. A light armchair stands behind her."
metaTitle: "AI for Sales"
description: "AI training for sales teams: prepare calls, check AI drafts against the product card, and hand off deals when conditions change. A full walkthrough."
sources:
  - "https://handbook.gitlab.com/handbook/sales/playbook/discovery/"
  - "https://handbook.gitlab.com/handbook/about/handbook-usage/"
  - "https://creativecommons.org/licenses/by-sa/4.0/"
---

AI training for sales teams works on one rule: AI writes the draft, the seller checks every promise. This article is not about selling a course or a workshop. It walks through a full business conversation cycle - from the buyer's email to the finished reply and the deal handoff. All data are instructional, composed specifically for teaching.

## What Does AI Do in Sales - and Where Does the Human Decide?

AI handles the preparatory work: it builds a list of clarifying questions, sketches a meeting plan, and suggests reply options for common requests. The seller's job is to read every option, cut promises that are not in the product card, and decide what to send the buyer.

AI does not call the buyer, run negotiations, or get a budget cleared with the CFO. It does not turn "approximately" into "definitely."

How much faster a seller works after this kind of walkthrough has not been measured.

## What Do You Start with Before the First Question?

You start with two things: the buyer's email and your product card. The walkthrough uses the following teaching input, composed by the author.

The buyer sent this email:

> We have four warehouses. The purchasing officer collects branch orders into a spreadsheet. Sometimes we have to recheck items before sending to the supplier. We want to prepare orders faster. We need integration with our accounting system, a fixed price, and launch next week. The purchasing manager makes the decision; the CFO approves the budget. We can show a de-identified example of an order.

The seller keeps a product card (composed by the author, not a real offer). Four features:

- the tool collects rows from a file in an agreed CSV format (CSV is a plain text file that stores data as a table, readable by Excel and most accounting systems);
- it flags rows where an item code or quantity is missing;
- it prepares a draft consolidated order - a combined order built from all warehouse requests;
- a staff member reviews the result and manually exports the CSV.

What the card does not describe: direct integrations with other systems, auto-ordering, a fixed price, a specific launch date, or reductions in errors or time. The import format is agreed before the start. Duplicate handling, catalog matching, and data-reconciliation order are all open questions.

That is the entire input. Everything else the seller either draws out in conversation or honestly names as unknown.

Instructional prompt for preparation (composed by the author, not executed):

> Request to check a reply to a buyer's email. Input data are the author's teaching example, not a real client.
>
> Buyer's email: "We have four warehouses. The purchasing officer collects branch orders into a spreadsheet. Sometimes we have to recheck items before sending to the supplier. We want to prepare orders faster. We need integration with our accounting system, a fixed price, and launch next week. The purchasing manager makes the decision; the CFO approves the budget. We can show a de-identified example of an order."
>
> Product card: collects rows from an agreed CSV format; flags missing item codes or quantities; creates a draft consolidated order; staff member reviews and manually exports CSV. Not confirmed: direct integration, automatic ordering, fixed price, launch date, time or error reduction, duplicate handling, catalog matching, order of data approval.
>
> Do not name the accounting software. Do not promise a price or a date that depends on a trial period or format agreement. The next step is a proposal, not an already-agreed action.
>
> Four results needed: 1) clarifying questions for the buyer; 2) mapping of requests to confirmed features; 3) a draft reply that states plainly that price, date, and integration are unknown; 4) a handoff variant for a colleague if the buyer rejects manual CSV export.

![An open metal sample case: fabric swatches and wood finish samples with blank tags sit in foam slots, and a sheet of coloured strokes is clipped inside the lid. A hand lifts out a cyan fabric swatch.](/blog/ai-training-sales-team/fig-exercise.webp)

## How Do You Prepare for the Sales Call?

Before each call, prepare four blocks: what you know, the outcome you want, what is missing, and open questions. This four-block approach is described in the [GitLab Sales Playbook](https://handbook.gitlab.com/handbook/sales/playbook/discovery/) (licensed [CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/)). We do not copy their template directly - we take the underlying logic: first understand where the buyer is now, then match that against what you actually offer. Everything else in this article rests on our own teaching fact pack and example conversation.

Ask AI to fill in the blocks from the email text, then verify and add to them yourself.

1. **Known.** Four warehouses, a purchasing officer, a spreadsheet, item rechecking. A request for integration, a fixed price, and launch next week. Two people: the purchasing manager and the CFO. The buyer offered to show a de-identified example of an order - that is an offer, not a confirmed commitment to send one.

2. **Desired outcome.** Understand what exactly hurts in the current process. Clarify what the buyer means by "integration." Agree on the specific next step.

3. **Missing.** Who else is involved besides the purchasing officer. How often orders come in and how many line items each has. What "faster" means: ten fewer minutes per order or an hour fewer per shift. Whether they already have CSV export from their accounting system.

4. **Open questions.** What "next week" means: a technical constraint or pressure from above. What happens if integration turns out to be outside the product's capabilities.

AI will draft thirty questions. Pick seven or eight that actually move the deal forward. Set the rest aside.

A checklist is not a script - it is a reminder. The seller does not read it aloud. When the conversation drifts, the checklist helps return to what is still unresolved.

Filled-in meeting plan (instructional; not observed in a real conversation):

1. **Greeting.** Name the topic; confirm the call is about the email.
2. **Current process.** Who collects orders, how often, how many items, what gets rechecked, what happens on an error.
3. **What "faster" means.** Specific pain: time per order, number of errors, load on the purchasing officer.
4. **What "integration" means.** How data leaves the accounting system. Is there CSV export.
5. **What "next week" means.** Hard deadline or a preference.
6. **What "fixed price" means.** Why fixation is needed: budget already approved, or just a reference point.
7. **Next step.** De-identified example of an order. Format agreement. When we come back with a figure.

The next step is a proposal the seller puts forward, not an agreed outcome.

## How Do You Clarify What "Integration", "Price", and "Deadline" Mean?

Start by learning the buyer's current process - before you discuss integration, price, or timeline.

**Uncovering the current process**

The buyer wrote "sometimes we have to recheck" - that could mean two errors a month or chaos at every warehouse. Do not turn a complaint into a diagnosis right away.

In the conversation the seller asks concretely:

- Who currently collects orders and how long it takes.
- What exactly the purchasing officer rechecks: item availability, quantity, price, article number.
- What happens when an item is missing or a quantity does not match.
- Whether they already have export from the accounting system to CSV or Excel.

AI prepared this question block in advance. In the conversation the seller does not read the list - he listens and probes. If the purchasing officer says "sometimes they mix up the warehouse," the seller asks: "How often? How many orders a week?" A number matters more than a feeling.

**Clarifying the integration request**

"Integration with the accounting system" means one of three things:

1. A direct API connection - automatic export and two-way synchronization. An API (application programming interface) is a way for two programs to exchange data directly, without human involvement.
2. A one-time CSV import that a staff member triggers manually.
3. Simply reading the same table format the buyer already uses.

The product card covers option 3 and partly option 2. For option 1, the seller says neither "possible" nor "impossible" - he asks: "How does data leave your system? Is there an export to file?"

On the fixed price, the seller is direct: in the teaching card, price is not listed at all, so naming it is not an option. "I cannot name a fixed price at this stage. I can offer to clarify volume with the colleague responsible for pricing."

On launch next week, the seller does not promise and does not refuse. He asks: "What happens if we do not make it next week? Is there a hard deadline from above, or would that just be convenient?" Sometimes "next week" means "we want to start as soon as possible," not "otherwise we will not sign."

The next step is a proposal: get the de-identified example and details about the accounting system. It is not yet agreed.

**If the buyer says: "Without integration, this is useless to me"**

The buyer might say plainly: "If you will not connect to our system, I do not understand what I would do with this."

The seller clarifies: "How does the purchasing officer currently get data from the system? Does he export a file, copy rows, retype them?" If there is at least a file export, the tool covers part of the manual work: row collection, field validation, draft preparation. If data does not exist in machine-readable form at all - meaning the computer cannot process it without manual entry - the seller says honestly: "We need at least a file. Without it there is nothing to automate."

The objection is not resolved by a promise. It is resolved by clarifying facts.

## How Do You Check the AI Draft and Send the Buyer a Reply?

AI composes a draft mapping. The seller checks every row against the product card.

| Buyer's request | What the product has | How to phrase it |
|---|---|---|
| Integration with accounting system | Import from agreed CSV, validation of empty fields | "We work with a file in an agreed format. Direct connection is not in the card; I will check with the team." |
| Faster order preparation | Row collection, missing-field flags, order draft | "The tool collects rows, flags missing fields, and prepares a draft. We have not measured a specific time reduction. We propose timing the path from receiving the file to a finished draft - as a check, not as a guarantee of savings." |
| Fixed price | Price not in the card | "I cannot name a fixed price. I will offer to clarify volume with the colleague who handles pricing. No condition guarantees a figure." |
| Launch next week | Date not in the card | "I cannot confirm a date. I cannot promise launch next week. I will ask a technical colleague whether it is feasible and on what timeline." |

Not one row becomes a promise. CSV capability does not become integration. A deadline request does not become an agreed launch. Reviewing the input does not promise an evaluation result.

**What a bad AI draft looks like.** Without oversight, AI might suggest:

> "We will connect your tool to the accounting system via API, launch next week, fixed price 48,000 rubles, error reduction of 60%."

Checking each point against the card:

- API integration - not in the card; only CSV import exists. Error.
- Launch next week - date confirmed by no source. Error.
- Fixed price 48,000 rubles - invented; the card has no price. Error.
- Error reduction of 60% - no measurement exists; the buyer only mentioned "sometimes rechecks." Error.

The seller does not send this text. He crosses out every promise and rewrites.

**Finished reply to the buyer** (instructional text):

> Thank you for the email and for your willingness to show an example order.
>
> **Tool.** We collect rows from a file in an agreed format, flag missing article numbers and quantities, and prepare a draft consolidated order. A staff member reviews the result and exports the finished CSV themselves. Direct connection to your accounting system is not described in our card. If you have an export from your system to file, we work with that file.
>
> **Price.** I cannot name a fixed price at this stage. I will offer to clarify volume with the colleague responsible for pricing.
>
> **Timeline.** I cannot confirm a date. I will ask a technical colleague whether it is feasible and on what timeline.
>
> **Next step.** Send a de-identified example of an order and tell me what the current export from your system looks like. Tell me how important the manual file transfer is in your process. We will check whether the format fits and come back with specifics.

The reply has no promises beyond the card and no invented numbers.

![A yellow armchair in a showroom. One hand removes a card with scribbles struck through in red, while the other places a new card with a cyan check mark into a clear stand.](/blog/ai-training-sales-team/fig-review.webp)

## When and How Do You Hand Off the Deal?

The seller hands off the deal when a material condition of the conversation changes.

In the teaching example, the purchasing manager says the staff member will not export the file manually. That changes a condition: the product at the current stage requires manual export.

The seller does not promise integration. He tells the buyer: "I need to check with the team whether an extension is possible. I will come back with an answer." Then he writes to his colleague.

Teaching handoff text:

> Colleague, I need your view. A buyer with four warehouses refuses manual CSV export. Our tool at the current stage requires a staff member to export the file themselves. Request: assess whether auto-export (automatic file output without manual action by a staff member) is possible, or whether this is a separate project. Their accounting system is not named; format unknown. I have not promised the client anything except that I will come back with an answer. The buyer said they can show a de-identified example of an order; that is an offer, not a confirmed commitment to send one.

In the handoff the seller names seven things:

- **To whom.** A colleague from the technical team.
- **What.** A request to assess whether auto-export is possible for the CSV tool.
- **Context.** Four warehouses, purchasing officer, rejection of the manual step. Accounting system not named, format unknown.
- **What is unknown.** Direct connection, auto-ordering, timeline, price, effect.
- **What to assess.** Whether a path to auto-export exists or whether this is a separate project.
- **What not to promise the buyer** before the team responds.
- **Proposed next step.** The buyer shows a de-identified example of an order; the seller clarifies the accounting system.

AI helped compose the handoff text. The seller checked that there is no word like "guarantee" and no specific date. Everything was composed for the instructional walkthrough.

## FAQ: AI Training for Sales Teams

### Can I use this walkthrough as a template for a real meeting?

You can take the structure as a starting point: input, four-block preparation, process discovery, request clarification, capability mapping, draft check, and handoff. Specific wording depends on your product and situation. Replace the teaching card with your own, update the list of unconfirmed features, and adapt the questions to your product. The core rule stays the same: AI writes the draft, you check every point against the card.

### What if the buyer insists on a fixed price before a trial run?

If the exact amount is not in your product card, you cannot name it. Do not invent a calculation and do not offer discounts. Say plainly: "The amount is not set. I will pass the question to the colleague responsible for pricing." Do not phrase a trial period or a delivery volume as a condition under which a price will appear - that creates an expectation you cannot meet. A trial period and delivery volume are not pricing policy and do not substitute for one.

### Does the buyer need to see the conversation plan?

No. The checklist works on the seller's side. What matters to the buyer is that the conversation moves toward a concrete next step. They do not need the list itself. The seller uses the plan as a prompt: when the conversation drifts, the plan shows what is still unresolved. There is no need to read it aloud - it is a tool for the seller, not a script for the buyer.

### What if AI suggests an answer that sounds convincing but is not confirmed?

A convincing tone does not make the text accurate. Check every point of the reply against your product card: if a point is in the source, it goes in the answer; if it is not, it stays out. That is exactly why the walkthrough shows an intentionally wrong draft - it is a model of what to catch before you hit send. The product card is the only source of permitted claims; a persuasive AI tone does not replace that check.

### How often should a deal be handed off to colleagues?

Hand it off when a specific fact that the conversation rested on changes. Not when it is inconvenient. Not when the buyer is pressing. The buyer refused the manual step - hand it off. The buyer named a new budget that changes the scope - hand it off. Their process turned out to differ from what the card describes - hand it off. A seller who tries to answer everything alone risks promising something that is not in the card.

### How does the teaching example differ from a real conversation?

All data in the walkthrough were composed specifically: there is no real buyer, no real call, and no real launch. The goal is to show how the cycle works and where the seller makes decisions. In a real conversation, details appear that shift the priority of questions - a different file format, an unfamiliar accounting system, or an unexpected rejection of a step. The structure of the cycle stays the same.

### Does the seller need to answer every question during the conversation?

No. If you do not have an answer right now, say so honestly: "I will check and come back." Answers invented on the spot tend to become promises that are not in the product card. The walkthrough shows exactly that moment: the seller does not know the answer about auto-export and says so plainly. A clear "I will check" works better than an uncertain "I think so."

## Next Step: AI Training for Your Sales Team

This walkthrough shows one thing: AI in sales is useful for drafts, not because it knows the answer. To keep it from creating promises in your place, you need three things - a product card, a list of unconfirmed features, and the habit of checking every point before you send. Checking and deciding is the seller's job.

If you want to work through a similar instructional case in sales, write to [hello@majento.ai](mailto:hello@majento.ai) or [Telegram](https://t.me/shimaoz). We will look at what worked and what needs rewriting.
