Ask a salesperson what slows them down and you rarely hear “selling”. You hear CRM updates, preparing for meetings, writing the same follow-up email for the fifth time, chasing a colleague for the latest price list.
Where sales time goes
- 70% of their time, sales reps say, goes on tasks other than selling
Picture a building materials distributor with six field reps. After each customer visit the notes go into a notebook. The CRM gets updated on Friday afternoon from memory, if at all. Follow-up emails go out a few days late. The sales manager finds out about the pipeline in the Monday meeting.
AI can take over a good part of that paperwork. It can’t build trust with a buyer, and it shouldn’t pretend to. This post goes through six places where AI helps a sales team, what stays with people, the data you need and how to roll it out so reps actually use it.
Six places AI helps a sales team
1. Research before a call
Before a meeting, a rep wants to know what the company does, how big it is, what has changed recently, who they’re meeting and what your company has already had to do with them. An assistant can put that on one page. It draws on the company’s website and news, public registers (in Slovakia, the business register and published financial statements) and your own data: CRM history, past quotes, open invoices, service tickets.
The internal part is often worth more than the public part. “They complained about late deliveries in March” changes a meeting more than the year the company was founded.
- You: Prep me for tomorrow’s 10:00 at the builders’ merchant in Trnava.
- Tool call: crm.account(id: 1187) · crm.activities(last: 12 months)
- Result: 3 orders this year, 5 last year · last visit 14 Feb · complaint 2 Mar: late delivery
- Tool call: invoices.open(customer: 1187)
- Result: 1 invoice overdue by 9 days · €4,380
- AI: Orders are down from 5 to 3. They complained about a late delivery in March, so check it was resolved before you offer anything new. One invoice is 9 days overdue. Their website announced a second branch in Žilina in September (link).
Two rules. Every fact in a brief should link to its source, and the rep should treat anything without a source as a guess. Models are fluent at filling gaps with plausible details; we explain why, and how to design around it, in why AI makes things up.
2. Meeting notes and CRM updates
For most teams this is the biggest time saver. Online meetings are recorded and transcribed, with the other side’s consent. Field reps dictate a two-minute voice memo in the car after a visit. The AI turns the transcript into a short summary, the CRM fields you care about, tasks and a follow-up draft. The rep checks and confirms.
From meeting to CRM
- Meeting or visit Online call or a voice memo in the car
- Transcript Speech turned into text
- AI drafts the update Summary, CRM fields, tasks, follow-up email
- Rep checks and edits Names, numbers and promises
- CRM updated, email sent Next step with a date
Decide on the fields first. If your CRM has 60 fields, pick the eight or ten that matter for forecasting and next steps: the next step and its date, decision makers, budget, timeline, competitors mentioned, products discussed. The AI fills those reliably. It won’t fix a CRM nobody agrees on.
3. Follow-up emails
The follow-up after a meeting is where deals quietly slow down. With the transcript in hand, the AI drafts the recap: what was agreed, what you’ll send, what you need from them and the date of the next contact. The rep edits it and sends it the same day.
The rule at the start is simple: AI drafts, a person sends. An email that promises a discount nobody agreed to is hard to take back.
4. Lead qualification and routing
Inbound enquiries arrive as free text: “We need about 200 m² of facade insulation for a project in Nitra, starting in April.” The AI reads it, pulls out what you need (company, product, quantity, location, timing), checks whether the company is already a customer and drafts a reply. Routing by region or product stays a plain rule, like the lead-routing example in our post on workflows, AI workflows and agents. Inbound phone calls can get the same treatment from a voice agent that asks a few questions and books a call with the right rep.
Be careful with AI scores. A lead marked “72” with no reason teaches reps to ignore the number. A short reason (“existing customer, large quantity, starts in six weeks”) is more useful than any score.
5. Proposal and quote drafts
AI is good at the words in a proposal: a summary of the customer’s situation, a description of the approach, a reference project adapted to their industry. It’s bad at prices. Quantities, prices, discounts and delivery terms should come from your price list, ERP or quoting tool, calculated by ordinary software. The AI can put them into the document. It shouldn’t produce them.
6. Pipeline summaries and questions
Every Monday the manager wants to know which deals moved, which stalled and which next steps are overdue. Most of that is a database query, not AI. The AI step is writing it up in plain language and answering follow-up questions, such as “which deals over €20,000 have had no activity for 30 days?”, without someone building a report first.
We built this kind of plain-language question answering over live data for Shovels. Their customers in construction and building materials use permit and contractor data for market research and sales, and can now ask questions about it in plain language.
Put together, a rep’s day after a customer visit changes like this:
After a customer visit
Today
- Notes in a notebook
- CRM updated on Friday from memory
- Follow-up email written in the evening
- Manager asks for status in the Monday meeting
With AI drafts
- A two-minute voice memo in the car
- CRM fields and tasks drafted, the rep confirms
- Follow-up draft ready the same day
- Manager reads a summary of what moved
What to keep human
- The relationship. Buyers want to know who they’re dealing with. A visit or a call can’t be delegated to a model.
- Negotiation, prices and discounts.
- Anything sent in the rep’s name, at least until you trust the drafts. Later, routine confirmations may go out on their own.
- Judgement on key accounts. Which deal to fight for and which to let go.
- Honest personal touches. An email that mentions the buyer’s latest post and is otherwise generic fools nobody.
Your CRM data decides the quality
An assistant that reads your CRM is only as good as the CRM. Duplicate companies, contacts who left two years ago, deals stuck in “negotiation” since last year: the AI will repeat all of it confidently. Before you roll anything out:
- Agree on the eight to ten fields that must be filled for every deal.
- Merge duplicates. AI can suggest likely matches; a person confirms them.
- Close or archive deals that are clearly dead.
- Decide who owns data quality, by name.
There’s good news too. Once the AI drafts CRM updates from meetings, the data tends to stay cleaner, because reps no longer have to type it in.
Recordings and consent
Tell the other side you’re recording or transcribing, and why. Many meeting tools announce it automatically; a field rep’s voice memo about a visit is a different case from recording the customer. Under GDPR you need a lawful basis and a retention period for both. Keep transcripts in tools your company controls, not in a rep’s personal AI account.
Buy or build?
Most CRMs, including HubSpot, Salesforce and Pipedrive, now have AI features, and plenty of meeting recorders write summaries. If your sales data lives in the CRM and your process is standard, start there.
A custom setup pays off when the useful information lives elsewhere: in the ERP, stock levels, price lists, invoices or service tickets. It also makes sense when you want one flow across several systems, for example from the voice memo to the CRM, the quote and the follow-up. We look at that decision in more detail in build or buy.
Getting reps to use it
Sales tools have a long history of being bought and then ignored. A few things make the difference:
- Save the rep time first and the manager second. If the first thing reps notice is more reporting, they’ll stop recording meetings.
- Don’t use transcripts to police people. If reps suspect every call will be replayed in their review, they’ll find reasons not to record. Agree up front who can read transcripts and why.
- Let reps shape the drafts. The best follow-up template comes from your best rep’s emails, not from a tool’s default. Collect a few good examples and give them to the model as the pattern to follow.
- Show results in the team meeting. A colleague who got an afternoon a week back convinces the others faster than any announcement.
How to roll it out
Start with two or three reps who want it, and one use. Meeting notes to CRM is usually the best first step, because each rep feels the benefit on the first day. Run it for four to six weeks. Then add follow-up drafts, then research briefs. Leave scoring, and anything that writes to customers on its own, for later.
What to measure:
- Time from a meeting to the CRM update
- Share of meetings with a next step and a date logged
- Follow-ups sent within 24 hours
- Time reps spend on admin, estimated weekly (rough is fine)
- Adoption: who still uses it after the first month
- Later on, win rate and the length of the sales cycle. They move slowly and depend on many things, so don’t judge a pilot by them alone.
Before you roll it out
- One process to start with Meeting notes to CRM is a good first choice
- Two or three reps who want it Volunteers, not the most sceptical person on the team
- The CRM fields that must be filled Eight to ten, agreed with the team
- A consent line for recordings And how long transcripts are kept
- A rule on what AI may send At first, nothing reaches a customer without a person
- A baseline Time to CRM update, follow-up speed, admin hours
- A date to decide Widen, change or stop after four to six weeks
If your reps spend more time on admin than with customers, tell us how your sales process runs today. We’ll suggest where AI would save the most time and where a simple rule would do the job.
