How Sales Teams Can Use Generative AI: Meeting Prep, Proposal Writing, and CRM Automation

The day before a meeting, I lose the whole evening just hunting down old materials, reworking the proposal and filling in the CRM.

When Sato, who looks after sales planning, said this from the corner of the meeting room, I fell silent for a moment. In conversations about putting generative AI to work, I often hear phrases like “we want to lift sales productivity” or “we’d like to run AI training.” The burden on the ground, however, is far more concrete. Researching the customer, revising the proposal, tidying up what was discussed and logging it in the CRM, then sending the follow-up email. Every one of these is necessary, yet stacked together they quietly drain a salesperson’s time away from talking to customers and into preparation and record-keeping.
Up until six months ago, interest in AI for meeting preparation, proposals and CRM entry had been growing at the company. What stayed vague, though, was “what to hand to the AI, and where the human steps in to check.” One sales manager put it bluntly: “Never mind the quality of the proposals, the way people prepare differs far too much from person to person.”
So over the past three months we ran sales AI training with eighteen members of the sales team, using Kanata. We rehearsed the whole sequence as one flow: researching the prospect ahead of a meeting, competitive analysis, drafting discovery questionnaire, role-play with an AI playing the customer, proposal outlines, post-meeting minutes summaries, follow-up wording, and CRM-ready summaries. On average, the time spent preparing for meetings fell by 60 per cent. That said, this figure is a before-and-after comparison within that one company, and results will vary with the industry, the offering, how the CRM is run, and the sales process.
This article sets out how to connect generative AI training for sales teams directly to the day-to-day work of meeting preparation, proposals and CRM entry. The aim is not to hand sales activity wholesale to an AI. It is to reach a state where preparation quality does not swing wildly whoever is responsible, and where more time goes back into the conversation with the customer. Training alone will not automatically raise your sales capability; only when it is combined with the sales process, information-handling rules and managerial review does it lead to repeatable improvement.

Why sales teams come to need generative AI training

Sales teams carry a great deal of work that sits well with generative AI.
Selling involves plenty of “gathering information,” “organising it to suit the other party,” “putting it into words,” “recording it” and “turning it into the next action.” These overlap neatly with what generative AI does well: summarising, classifying, drafting, comparing and structuring.
In meeting preparation, for instance, the following tasks crop up:

Researching the prospect’s business
Checking industry trends and the competitive landscape
Forming a hypothesis about the customer’s challenges
Drawing up a discovery questionnaire
Digging out similar past proposals
Working out the structure of the proposal
Preparing for likely questions and objections

After the meeting, the salesperson’s work carries on:

Tidying up the meeting notes
Producing a minutes summary
Pulling out the next actions
Writing the follow-up email
Entering the activity history into the CRM
Sharing the situation with the manager
Marshalling the points for the next proposal

These are indispensable to building trust with the customer. Do all of it by hand, though, and a salesperson’s time is heavily eaten up by preparation and record-keeping.
The point of generative AI in sales is not to replace the salesperson’s role. By handing research, summarising, drafting, formatting and entry support to the AI, you return human time to what people should be doing: understanding the customer, forming hypotheses, making decisions and holding the conversation.
On the ground during AI rollouts, I try to think first not only about “what to hand to the AI” but about “where to return the human work.” For a sales team, that means seeing through the customer’s real problem, sensing the temperature of the words in the room, and reading how the other side actually makes decisions.
For that reason, sales AI training that merely teaches how to write prompts is not enough. You need to design which moments in the sales cycle use generative AI, who checks the output, and how it feeds into the CRM and the proposal.

Why generative AI training fails to stick on the ground

You can run generative AI training for a sales team and still find it goes unused in practice.
The main reason is that the training is not joined up with the actual sales work.
Run exercises along the lines of “let’s draft an email,” “let’s summarise this” or “let’s brainstorm some ideas,” and on their own they do not carry over into the next day’s meeting preparation or CRM entry.
What salespeople actually want to know is far more concrete:

For tomorrow’s meeting prep, what should I ask the AI?
How do I shape meeting notes into something fit for CRM entry?
When drafting a first cut of a proposal, how far can I leave it to the AI?
For the follow-up email to the customer, how much of the AI’s wording is it sensible to use?
If the AI plays the customer in a role-play, what conditions should I set?

Training that does not answer these practical questions may go down a storm on the day yet fall out of use within a few weeks.
I have watched this “fizzles out after the session” pattern play out many times. During the session people are amazed. Watching the AI polish prose, summarise and propose questions, they think, “this looks usable.” Come the following week, though, they slide back to their old way of working. The reason is simple: it was never built into their own workflow.
In sales teams, the spread of individual experience also weighs heavily. Top performers have their own playbook even without AI. Newcomers and people who have just moved across, by contrast, do not yet have a playbook for meeting preparation or proposal writing.
To close that gap, it matters to treat generative AI not as “a handy personal tool” but as “a mechanism for sharing the sales team’s playbook.”
When you use a service that lets a team accumulate shared instructions and reference materials, such as Kanata, preparing instruction sets for meeting prep, proposal structure and a CRM-entry summary format in advance makes it easier to reduce the variation in how each person works.
This is less about standardising how AI is used and more about standardising the quality of sales preparation.

The practical areas sales AI training should cover

Generative AI training for sales teams becomes far easier to tie to real work if it covers at least the following five areas.

Meeting preparation AI: understanding the customer and forming hypotheses

Using AI for meeting preparation goes beyond tidying up a profile of the prospect; it extends to forming hypotheses about their challenges.
You might feed in the customer’s industry, company size, public information, the services they already use and the organisational issues you would expect, then ask the AI for output such as:

The challenges the customer is likely to be facing
The questions worth confirming in the first meeting
The points a decision-maker is likely to care about
The worries the people on the ground may feel
The battlegrounds when you are compared with a competitor

The crucial thing here is not to treat the AI’s output as “the answer.” A person confirms the AI’s hypotheses in the meeting. That is the basic way to use meeting preparation AI.
I tend to see AI in sales preparation as something close to a “sparring partner for hypotheses.” It is useful, but it does not know everything about the customer’s internal politics. The subtle reactions from past meetings, a contact’s tone of voice, the power dynamics inside the organisation: these still have to be supplied by a person.
In training, it works well to use cases close to real customers and, as a team, review the AI’s hypotheses for “which can we use,” “which are risky” and “which need confirming.”

Discovery questionnaire s: reducing variation in questioning

In selling, the quality of your questioning shapes the quality of your proposal.
Yet the less experienced the salesperson, the more likely they are to walk into a meeting unsure of what to ask. As a result, information needed for the proposal is missing, and follow-up questions crop up later.
With generative AI you can produce a first draft of a discovery questionnaire tailored to the customer’s profile and the purpose of the meeting.
You might, for example, have the AI organise items such as:

Current operations
Challenges
Existing tools
Decision-makers
Budget expectations
Timing of adoption
Services under comparison
Conditions for success
Concerns

In sales AI training, you go beyond simply generating a list of questions to consider “should this be asked in the first meeting,” “should it wait until the second or later,” and “how do I ask it without making the customer feel cornered.”
What I often tell teams in the field is that a discovery questionnaire is not a questionnaire but a blueprint for a conversation. Too many questions and the meeting turns into an interrogation. Too shallow, and the basis for your proposal grows thin.
After the AI has offered draft questions, the person designs “in what order to ask,” “where to dig deeper” and “what to deliberately leave unasked.” This is where the salesperson’s experience comes into its own.

Proposal AI: easing the burden of starting from a blank page

The value of proposal AI is not in producing a finished version in one go.
Where it pays off most is in drafting the structure, the headings, the key points, the comparison tables and the gist of each slide.
Cut down the time a salesperson spends staring at a blank slide, and get to a customer-shaped skeleton of the proposal sooner. That is the realistic way to use proposal AI.
In training, exercises like these work well:

Building the chapter structure of a proposal from the customer’s challenges
Teasing out the axes for a competitive comparison
Sorting the expected benefits into quantitative and qualitative
Producing several versions of customer-facing explanatory copy
Drawing up review criteria for the manager

In my own work on product development and sales materials, I find the thing that takes the most time is not “writing the prose” but “deciding the structure.” Proposals are no different. In what order do you set out the customer’s challenges, where do you bring in your own strengths, and at what point do you show the cost-benefit case? Once that flow is settled, the work moves along far more readily.
That said, a proposal is a document that goes outside the company. Rather than submitting the AI’s prose as-is, the salesperson and the manager must always check the facts, the figures, the wording and the soundness of the customer understanding.

Minutes summaries and follow-up: speeding up the first move after a meeting

Speed after a meeting feeds directly into the customer’s experience.
Put off the minutes summary, internal sharing, the follow-up email and the CRM entry once a meeting is over, and your response slips. Memory fades, and the warmth of the conversation is lost.
With generative AI you can take meeting notes or a transcript of a recording and organise information such as:

A meeting summary
The customer’s challenges
The words the customer used
Points agreed
Points still to confirm
Next actions
An agenda for the next meeting
A draft follow-up email

The important thing here is to cross-check against the salesperson’s own memory. The AI cannot fully grasp the mood in the room or the customer’s subtle reactions. That is precisely why, after the AI has summarised, a person needs to correct it: “this matters,” “this nuance is off.”
For AI after a meeting, I reckon both “getting it out quickly” and “recording it correctly” matter. A salesperson who follows up promptly tends to feel reassuring from the customer’s side. Rush it and send something inaccurate, however, and you risk damaging trust.
So in training we design things on the assumption that a person will always read back over any AI-produced minutes summary or follow-up wording. The AI does the first pass; the person corrects the temperature and the facts. That division of labour is the realistic one.

CRM-entry AI: turning records into something usable for sales management

CRM entry is heavy going for many salespeople.
CRM stands for Customer Relationship Management: the system for managing customer information and the history of deals. In selling, it is used to record the progress of a deal, the customer’s challenges, the next actions, the likelihood of winning and so on.
Where the quality of CRM entry is poor, sales managers struggle to read the state of their deals accurately. Marketing struggles to analyse which leads are turning into deals and how. Leadership, too, may misjudge the pipeline.
With CRM-entry AI you can take meeting notes and organise items such as:

A summary of the deal
The customer’s challenges
BANT information
Decision-makers and stakeholders
The competitive situation
Likelihood of winning
Next actions
Risks
Matters to raise with the manager

BANT is a sales-management idea taking the initials of Budget, Authority, Need and Timing. There is no need to apply it mechanically to every deal, but it serves as a useful guideline for organising where a deal stands.
In sales AI training it matters to build prompts that match the CRM’s own fields. Rather than each salesperson entering things freely every time, having a common entry format across the team turns the CRM into data you can actually use for sales management and analysis.
I find that treating the CRM as “a box to fill in” only adds to a salesperson’s sense of burden. The CRM is, by rights, the foundation a sales team uses to make its next call. That is exactly why you should use AI to lower the entry burden while, at the same time, bringing the granularity of what gets entered into line.

Steps for designing generative AI training for sales teams

Generative AI training for a sales team is far easier to tie to real work if you design it along the following lines.
What I keep front of mind when designing training is not to over-reach for sophisticated AI use from the outset. On the sales floor, a small win you can use from tomorrow does more for adoption than an elaborate mechanism.

Break the sales process down

First, divide selling into broad stages:

First contact after acquiring a lead
The first meeting
Hearing out the challenges
Preparing the proposal
Writing the proposal
Handling a competitive pitch
Closing
Post-meeting follow-up
CRM entry
Handover after winning

Within these, decide where generative AI is used and where it is not.
Tidying up a company profile, drafting questions, summarising minutes and producing a CRM-ready summary, for instance, are areas that sit well with AI. The final call on pricing, building trust with the customer and settling the terms of a contract, on the other hand, are areas a person should own.
Leave that line blurred and the team is left guessing. One person leans on the AI too much; another barely uses it at all.
In training, make the rules explicit: “before a meeting, use it for company research and drafting questions,” “after a meeting, use it for the minutes summary and the CRM-ready summary,” “anything going to the customer is always checked by a person.”

Build shared prompts

In a sales team, a common template prepared for everyone sticks better than prompts written individually.
You might, for instance, prepare prompts such as:

A meeting-preparation prompt
A hearing-sheet prompt
A proposal-structure prompt
A competitive-analysis prompt
A meeting-notes summary prompt
A CRM-entry prompt
A follow-up-email prompt
A role-play prompt

When you use a service that manages prompts and reference materials at team level, such as Kanata, keeping these instruction sets as a shared asset is handy. Salespeople no longer have to write instructions from scratch every time, and sales managers find it easier to see what playbook the team is using.
I think it matters not to let prompts become “one person’s party trick.” A good prompt sitting only in one salesperson’s hands never becomes an asset for the organisation. Shared as a playbook for meeting preparation or CRM entry, by contrast, it can serve to train newcomers and bring people who have just moved across up to speed.

Get the reference material in order

In sales AI training, what you let the AI refer to also matters.
For use in a sales team, it helps to have the following in order:

Service materials
Proposal templates
Case studies
Frequently asked questions
Competitive comparison materials
Industry-specific talk scripts
A summary of reasons for lost deals
Customer-facing explanatory materials
Sample wording for post-meeting follow-up

Getting this material into a form the sales team can use makes it easier to obtain output that fits your own context.
Customer names, personal data, contract values, undisclosed information and the like, however, call for care in how they are handled. Because sales teams deal with a great deal of confidential material, you need to be clear before the training about “what may be entered,” “what should be masked” and “what must never be entered.” For data-protection considerations when using AI, the UK Information Commissioner’s Office offers useful guidance (ICO guidance on AI and data protection).
In the early stages of sales AI use, I recommend starting with public information and sales materials already shared internally. Rather than handling large volumes of customer-specific deal information straight away, establish the approach within a lower-risk scope. From there, widen the scope as you put information-handling rules in place; that is the safer route.

Use it for role-play

In sales training, using generative AI as a role-play partner is another effective approach.
You might, for instance, give the AI a role such as:

An IT director wary of adoption
A business owner focused on cost-effectiveness
A sales director worried about the load on the team
A marketing lead comparing competing services
A customer contact burned by a past system rollout

With the AI playing the customer, salespeople can rehearse likely questions and how to handle objections.
I find genuine practical value in this AI role-play. Newer and more junior salespeople, in particular, simply do not get enough chances to fail. Fail in a real meeting and it affects the customer relationship; against an AI, you can practise as many times as you like.
That said, real customers do not necessarily react in the tidy way an AI does. There are silences. There are vague answers. Sometimes they decline in a roundabout way, unable to spell out their internal situation.
That is precisely why AI role-play should be positioned not as a substitute for the real thing but as practice in preparation for it.

Have managers review

For sales AI training to stick, the manager’s involvement is indispensable.
The manager reviews the meeting-preparation notes, the proposal structure and the CRM entries the AI has produced, checking against criteria such as:

Is the hypothesis about the customer’s challenge sound?
Has the thrust of the proposal drifted off course?
Is the competitive comparison one-sided?
Are the figures and examples grounded in evidence?
Is the wording fit to send to the customer?
Can the CRM entry actually be used to manage the deal?

Through this review, salespeople learn not just how to use the AI but how to think as a salesperson.
I sometimes feel that what makes or breaks generative AI training is not the trainer but the manager on the ground. The ways of working learned in training get used in everyday deal reviews and pipeline meetings. When a manager asks, “this meeting-prep note was drafted by AI; where did you correct it yourself?”, that kind of conversation is what helps AI use take root on the ground.

Don’t make “time saved” your only measure

Bring generative AI into a sales team and the first benefit you tend to see is time saved.
The time taken on meeting preparation, on producing minutes, on CRM entry and on a first draft of a proposal are all relatively easy to measure.
Judge sales AI training by time saved alone, though, and you can lose sight of the real point.
What truly merits attention is whether the quality of the sales work is rising.

Examples of outcome measures worth checking in sales AI training
Measure	What to check
Meeting preparation time	Has the time spent on pre-meeting research and material preparation come down?
CRM completion rate after meetings	Is the information needed after a meeting actually recorded in the CRM?
Time to send the follow-up	Is the first move after a meeting being made without delay?
Proposal review rework count	Is the structure and content quality of proposals holding steady?
First-to-next meeting conversion	Is the quality of preparation and questioning leading to a next touchpoint?
Quality of meeting notes the manager has checked	Is the information needed to judge the deal being recorded?
The content of follow-up questions from customers	Are you grasping the customer’s interests and concerns?
Accuracy of lost-deal records	Are the records usable for loss analysis and sales improvement?

When I look at the results of sales AI use, I try to look beyond time to “whether the material for decisions has grown.” Get CRM entry in order, for instance, and managers can spot the risks in a deal sooner. Tidy up the meeting notes and marketing can better gauge lead quality. Standardise proposal structure and even a newcomer can face the customer at a consistent level of quality.
The benefit of generative AI shows up in the short term as a cut in working hours. Over the medium to long term, however, the value lies in bringing the whole team’s preparation quality, proposal quality and record quality into line.

What to watch out for when sales teams use generative AI

When a sales team uses generative AI, there are a few things to watch.
First, the handling of customer information.
Meeting notes, proposals, contract terms, contacts’ names, email addresses, budget information and the like all need careful handling. In training, it is important to show, with concrete examples, what information may be entered and what may not.
Personal data, customer-specific confidential information, undisclosed financial information and contract terms should, as a rule, have their handling decided in advance. Where necessary, mask before entering into the AI: swap a customer name for an industry or company size, replace a contact’s name with a job title, express amounts as a range.
Next, do not place too much trust in the AI’s output.
AI can produce plausible-sounding proposals and competitive comparisons. Sometimes, though, the facts are wrong. Competitive information, expected benefits, pricing and anything touching on legal or contractual matters in particular must always be checked by a person. AI use, generative AI included, has to weigh not only the benefits but also the risks: misinformation and disinformation, intellectual property and security. A structured way to think these through is the NIST AI Risk Management Framework (NIST AI RMF).
There is also the matter of not flattening the salesperson’s individuality. Force every proposal and follow-up email into the same mould and the wording can end up at odds with the relationship you have with the customer. Generative AI is suited to drafting and organising, but adjusting the closeness with the customer at the end is a human’s job.
I think a salesperson who can take an AI draft and put it back into their own words is stronger than one who sends the AI’s prose as-is. Customers are not looking only at polished prose. They are watching whether the person genuinely understands their company and is thinking in terms of their particular situation.
Finally, do not let the training be a one-off.
How to use generative AI only beds in once it is built into the work. You need a mechanism for ongoing operation: reviewing prompts monthly, sharing the approaches that worked, and managers checking on AI use during deal reviews.

In summary: sales AI training is a prompt to rethink the sales process

Generative AI training for a sales team is not merely tool training.
Meeting preparation, discovery questionnaires, proposals, competitive analysis, role-play, minutes summaries, follow-up, CRM entry. It is the work of breaking each of these down one by one and deciding what to hand to the AI and what a person should own.

Dividing the work between AI and people in sales AI training
Areas to hand to the AI	Areas people own
The groundwork of research	Understanding the customer
Organising information	Judging which hypotheses hold
A first draft of the prose	The call on the proposal
Tidying up the records	Building the relationship
Entry support	The final check

The essence of generative AI in sales is not simply to make the salesperson’s life easier. It lies in sharing the sales playbook across the whole team, closing the gaps in experience and freeing up more time to face the customer.
I feel the value of generative AI is not only in “working in a person’s place” but also in “reminding people of the work they ought, by rights, to be doing.” In sales, that means listening to the customer’s words, sizing up the challenge, and taking responsibility for the proposal.
When you do bring in a generative AI service, there is no need to change everything at once. It is realistic to start with the work where the benefit is easy to see: meeting preparation, proposal outlines, meeting-notes summaries and CRM-ready summaries.
From there, the team improves the prompts, gets the reference material in order, has managers review, and grows it all into a shared asset for the sales team. That steady accumulation is what leads to generative AI taking root in a sales team.

Q&A: common questions about generative AI training for sales teams

Where should a sales team’s generative AI training begin?

It is realistic to begin with work where the benefit is easy to see and which is not submitted directly to the customer: meeting preparation, summarising meeting notes, CRM-ready summaries and the like. For external deliverables such as proposals and follow-up emails, run things so that the AI produces a draft and a person always checks it.

Is it acceptable to use AI to create proposals?

Using AI for the structure, headings, organising the key points and a draft comparison table is effective. But since it is a document going outside the company, you should avoid using the AI’s prose as-is. Customer information, figures, expected benefits, competitive comparisons, contract terms and so on need to be checked by the salesperson and the manager.

If we adopt CRM-entry AI, will the salesperson’s entry burden disappear?

It will not vanish entirely. The AI can organise candidate entries from meeting notes, but the temperature of the meeting, the customer’s real feelings and the risk judgements still need correcting by the salesperson. Think of CRM-entry AI less as something that does the entry for you and more as an aid that brings the granularity of records into line and lowers the burden of entering them.

Does AI role-play genuinely help with sales training?

There are situations where it helps. It suits newcomers and junior salespeople in particular for practising likely questions, handling objections and the flow of a first meeting. That said, real customers do not necessarily react in the tidy way an AI does. AI role-play should be positioned not as a substitute for the real thing but as preparation practice before a meeting.

What is the single most important thing to watch when a sales team uses generative AI?

The most important thing to watch is the handling of customer and confidential information. Meeting notes, contract terms, contacts’ names, email addresses, budget information and the like need their handling decided before entry. Mask where necessary, and be clear before the training about what information may and may not be entered.