AI assistant for business: what it can really do in sales, what it can't, and how to start with one process

BOTrento Team5 October 2026 · Updated 11 October 20268 min read

What an AI assistant can realistically do in sales

Language models are strong at working with text. They can read unstructured emails, compare them with your reference material and write clear drafts. For a small business's sales, that mainly means the following tasks:

Sorting and extracting enquiries

From an email it pulls out the date, location, scope, requested services or equipment, contact and budget, if the customer gave one. Above all, it flags what is missing from the enquiry. An event company is a good example: a message arrives saying "company party in September, we need sound and lighting". The assistant lists the known parameters, points out that the number of guests, the type of venue and the setup time are missing, and suggests follow-up questions.

Draft replies

Acknowledging an enquiry, clarifying questions, an indicative price or a polite decline. The assistant writes a draft in your tone, and the salesperson only edits and approves it. If you have no templates of your own, start with our process for handling sales enquiries.

Draft quotes

Based on your price list and terms, it assembles line items, quantities and notes. One rule applies: the numbers must come from your price list, not from the model's imagination. We cover this in more detail in how to speed up sales proposals without losing control of your prices.

Follow-up and overview

It tracks where a customer has not replied, suggests a reminder at the right moment and summarises a long thread in a few sentences. That gives the salesperson an overview of what needs doing each morning, without having to piece it together from the inbox.

What an AI assistant can't do on its own (and shouldn't)

The limits are not set by technology alone. They are also about responsibility towards the customer. These things should stay with a person:

• Price and discounts. The assistant can calculate a quote according to the rules. Exceptions, discounts and margin are for the salesperson to decide.
• Confirming dates and availability. Without a link to live bookings of equipment and people, the assistant has no way of knowing whether a date is free. Even with such a link, a person should check the confirmation.
• Commercial judgement. Which job to take, which to decline and which customer deserves more time depends on context you will not find in an email.
• Negotiation and relationships. A call with a long-standing client or the handling of a complaint belongs to people.
• Facts it doesn't have. When a model lacks information, it often fills the gap with a plausible-sounding answer. That is why it must work only from your material and admit what it does not know.

The working rule is simple: the assistant prepares, a person decides and sends.

Risks and how to keep them under control

For legal questions such as GDPR or obligations under the EU AI Act, we recommend discussing your specific setup with a lawyer. A general checklist is a good start, but it does not replace a legal assessment of your situation.

How to start: one process, one measured trial run

The most common mistake is wanting "AI for everything" at once. The result is vague expectations and nothing you can measure. It is more reliable to start with a single process:

1. Choose one type of business case. Pick one that repeats and can be described, such as a sound equipment rental enquiry, a corporate event enquiry or a service booking.
2. Describe the current procedure. Where the enquiry arrives, who owns it, what is recorded in the spreadsheet or CRM (HubSpot, Pipedrive or similar) and when follow-up goes out.
3. Prepare the source material. Price list, terms, a few good replies and quotes, and a list of frequent questions.
4. Set approval rules. Who approves replies, who approves quotes and what must never be sent, for example a discount beyond the price list.
5. Define process KPIs. For example time to first reply, the share of enquiries with a clear next step, or the number of quotes that passed without major corrections. Don't measure revenue; too many other things influence it. If you need a set of numbers, see 6 KPIs to track every week.
6. Evaluate and decide. When the trial run ends, compare the KPIs with the situation before it. Then decide whether to extend, adjust or end the process.

Trial run brief template

Process: enquiries of [case type] from mailbox [address]
Record: [spreadsheet / CRM]
The assistant does: extracts details, flags missing information, drafts a reply and a quote based on the price list, reminds about follow-up
The assistant does not: send without approval, change prices, confirm dates
Approved by: [name and role]
We measure: time to first reply, share of enquiries with a next step, number of corrections to drafts
Evaluation: [date]

How BOTrento Chief of Staff approaches it

BOTrento Chief of Staff is not another self-service chat. It is a managed process for enquiries, quotes and other sales steps. It is introduced exactly as outlined above, through a 14-day free trial. The 14 days start with your first successful upload of customer data (import or mailbox connection), and you have 7 days to prepare.

• Human approval: nothing is sent until you approve it. Email and quote drafts are always reviewed by a person.
• Your own API keys: you enter the keys of your model provider (e.g. Anthropic Claude, OpenAI or Google Gemini) and pay for AI directly to them, with no markup. Keys are encrypted, only the last 4 characters are shown, and you can remove them at any time.
• Data: we store it on servers in Germany (EU). The text needed for the task is processed by the model provider you choose. Commercial API providers do not use data for training. However, an EU server that calls an external AI API is not a closed computation, and we say so openly.
• Oversight: monitoring runs around the clock, and we respond to incidents within 4 hours on working days.
• Guarantee: we take over and fix defects in the delivered process and guarantee the agreed process KPIs. We do not promise revenue or a number of orders.

For a sense of scale: in an internal pilot, 35 enquiries, 18 emails, 24 quotes and 15 actions went through the process in 7 days. To see what such a process looks like step by step, view the Chief of Staff process demo. You can start with the 14-day free trial.