AI assistant for your sales team: which tasks to hand over first and which to keep human

BOTrento Team11 October 20269 min read

The split at a glance

One question decides where each task goes: if the assistant gets it wrong, who notices, and what does it cost you? When a person checks the output before it reaches a customer, a mistake costs a minute of editing. When a mistake reaches the customer as a price or a confirmed date, it can cost you the deal or your margin.

Hand over first: four tasks an AI assistant handles well

1. Inbox triage

A shared sales inbox fills up with real enquiries, supplier mail, invoices, newsletters and "just checking in" threads. The assistant can flag what looks like a new enquiry, what is a reply to a quote you sent, and what can wait. This doesn't take judgement. It just takes reading every email, every day, which is exactly what people skip when they're busy.

2. Morning priorities

Most small teams start the day by scrolling. A better start is a short list: the one thing that matters most today, plus a few next steps. That might be an enquiry from yesterday afternoon that's still unanswered, a quote sent last week with no reaction, or a client you promised to call. The assistant builds the list from your mail and data, so nobody has to keep it in their head.

3. First drafts

Writing the same kinds of reply again and again is slow: confirming you've received the brief, asking for missing details, sending the quote cover email. An assistant can draft these from the enquiry text in the customer's language. A person reads the draft, fixes it and sends it. For help with the reply itself, see our short guide on replying to sales enquiries faster.

4. Follow-ups and promises

"I'll get back to you by Friday" is the promise small teams lose track of most often. An assistant can spot these promises in sent mail, remind you before the deadline, and prepare a follow-up draft for quotes that got no reply. If this is where you lose the most, read how to stop forgetting quote follow-ups.

Keep human: where the decision is commercial

Pricing and discounts

Your price reflects things that rarely appear in an email: how busy you are that week, whether this client pays late, whether a smaller job now could lead to a bigger one later. An assistant can put your list prices into a draft. Whether to give a discount, and how much, is your call.

Dates, availability and bookings

An equipment rental company or production crew can't let anyone confirm a date without checking stock, crew and existing bookings. Double-booking a lighting rig or a sound engineer does damage an apology won't fix. The person who can see real availability should give that confirmation. Our pages for equipment rental teams and event and production companies show how this works in practice.

Negotiation and difficult messages

Handling a price objection, a complaint or a scope change needs a person who understands the relationship. AI can suggest wording. It shouldn't decide what you give away.

The send button

Everything that reaches a customer goes out under your company's name. Keep a simple rule: a person approves every outgoing message. The rule also helps you meet transparency and oversight duties under the EU AI Act. Our article on the AI Act for small businesses explains this in practical terms. For your own situation, check with a lawyer.

The grey zone: tasks you share

Some tasks fall in between. Split them within the task: the assistant prepares and a person decides.

• Qualifying an enquiry. The assistant lists what's missing (date, location, number of guests, delivery address). You decide whether the job is worth pursuing.
• Quote drafts. The assistant structures the quote and fills in items from the enquiry. You check quantities, prices and terms.
• Chasing quiet quotes. The assistant reminds you and drafts the follow-up. You decide whether to chase, wait or drop it.
• Attachments. If the brief or technical rider is in a PDF, read it yourself. Many assistants, ours included, work from the email text rather than the contents of attachments.

This is also the main difference between a general chat tool and a managed process. A chat window helps one person write one email. A sales process needs records, reminders and approval across the whole team. The trade-offs are covered in ChatGPT or Claude vs a managed AI process.

How to start: a two-week plan and checklist

Don't hand everything over at once. Test one process on your real enquiries and judge it on the results.

1. Pick one inbox or one data export. Use the shared sales address, or a CSV/XLSX export from Pipedrive, HubSpot or your spreadsheet.
2. Write down the approval rule (template below) and share it with the team before you start.
3. Week 1: triage and priorities only. Each morning, compare the assistant's list with what you would have done anyway. Note anything it missed or got wrong.
4. Week 2: add drafts and follow-ups. Count how many drafts you send with light edits and how many you rewrite from scratch.
5. Decide. Keep the tasks that saved time without causing new mistakes, and drop the rest.

Checklist before you go live:

• Who approves outgoing messages, and who covers when they're on holiday?
• Which topics always go to a senior person (discounts, complaints, large jobs)?
• Where is the data stored, and which AI provider processes the text? See the GDPR checklist for AI sales tools.
• Have you told the team what the assistant does and doesn't do?

An internal rule you can copy:

Our rule for the AI sales assistant
The assistant may sort incoming mail, prepare a daily priority list, draft replies, quotes and follow-ups, and remind us of promises we've made.
The assistant may not send anything, confirm dates or bookings, or set prices and discounts.
A named person reads and approves every message to a customer before it is sent. Discounts and complaints go to [name].

For a longer rollout guide, see how to roll out an AI assistant in a small company's sales.

How BOTrento handles the split

BOTrento Chief of Staff is designed around this division of work. It's a managed process for enquiries, quotes and follow-ups, not an open chat window.

• Watches enquiries and quotes in your mailbox and uploaded data. Mailboxes connect over IMAP. For now, Gmail, Google Workspace and Microsoft business mailboxes (Outlook) connect by forwarding or import. CRM or spreadsheet data comes in as CSV/XLSX.
• Today 1+3: every working-day morning you get one priority plus three next steps, in the app and by email.
• Drafts replies, quotes and follow-ups in English, German, French and Czech. It also picks up promises like "I'll get back to you by Friday" and reminds you about them.
• Never sends anything itself. It doesn't confirm dates or bookings and doesn't replace your CRM or accounting. A person approves every message.
• AI usage is included in the plan, with a monthly limit and a warning before you reach it, so there are no surprise charges. If you prefer, you can use your own Anthropic API key.

Data is stored on servers in the EU (Germany, with the application database in Ireland). Only the text a task needs goes to the language model provider through its commercial API. In our internal pilot on BOTrento's own processes, one week covered 35 enquiries, 18 emails, 24 quotes and 15 actions. You can see the process step by step or try it on your own data. The 14-day free trial needs no card and no API key, and it starts only when you first upload data. Plans are listed on the pricing page.