AI assistant for business: how to roll it out in a small company's sales and what to expect

BOTrento Team9 October 2026 · Updated 12 October 20269 min read

Where an AI assistant really saves time in sales

An AI assistant helps most with work that is repetitive, text-based and easy to check. For enquiries and quotes, that usually means five areas.

Sorting and completing enquiries

The assistant pulls the key details out of an email and flags anything missing. Take an event company: an enquiry for sound equipment at a corporate event gives the date, the venue and a rough number of guests, but says nothing about crew access, power supply or whether the event is outdoors. The assistant flags these gaps and suggests follow-up questions.

Drafting the first reply

Using your template replies and terms, it prepares a draft email. The salesperson edits and sends it rather than writing from scratch.

Material for the quote

The assistant puts together a proposed list of items based on your price list and similar past jobs. A person still checks and confirms the price and any discounts.

Follow-up and next step

For every open case it tracks when you should get back in touch and drafts a reminder. That way, fewer quotes slip just because nobody had time to chase them.

Updating your records

Case status, contact and next step go into your CRM (such as HubSpot or Pipedrive) or a spreadsheet. The owner can see where each enquiry stands without asking around the office.

What not to expect from an AI assistant

A lot of disappointment comes from expecting too much. Plan on the basis that the assistant:

• does not decide on price, discounts or dates. It can suggest them, but a person is responsible for them.
• does not know equipment availability or staff capacity unless that information is in records it can access.
• will not fix a chaotic process. If you haven't agreed who handles an enquiry and what stage it is at, the assistant will just make the chaos run faster.
• is not a sales result in itself. It cuts routine work and makes the pipeline easier to see, but whether you win a deal still depends on the customer and your offer.
• should not send anything on its own. The customer is dealing with your company, not with a model.

For a fuller breakdown of which tasks an assistant can and cannot handle, see what an AI assistant can really do in sales.

Self-service chat or a managed process

Most companies start with a salesperson pasting an enquiry into a chat and asking for a reply. For one person, that is enough. Once a whole team and dozens of cases a month are involved, the difference becomes clear:

For when a chat is enough and when a managed process makes sense, see ChatGPT or Claude for your business, or a managed AI process.

How to introduce an AI assistant: a 7-step plan

1. Choose one process. For example, "equipment rental enquiry → first reply → quote". Don't start with every type of job at once.
2. Map how things work today. Note where enquiries arrive, who handles them, where they are recorded and where they most often get stuck. Our article on handling event enquiries in peak season may help.
3. Set your metrics. Keep to a few: time to first reply, number of enquiries with an agreed next step, number of open quotes with no follow-up. Measure them before you start.
4. Prepare your materials. Gather your price list, template replies, terms and conditions, and a few sample quotes you consider good. The service quote template can serve as a reference.
5. Set up approval. Agree who approves drafts, what they can change and what must go to the owner (non-standard price, date, complaint).
6. Run the trial for a fixed period. A month is usually long enough to see both typical and less typical cases.
7. Evaluate and decide. Compare the metrics with your starting point, review the rejected drafts and decide whether to widen the scope, adjust the rules or stop.

Trial run brief template

A brief keeps your team and your supplier on the same page. One page is enough:

Process: enquiries for [job type] from the mailbox [address]
Records: [CRM / spreadsheet], fields: contact, stage, next step, deadline
What the assistant does: sorts enquiries, suggests follow-up questions, drafts the first reply and the quote, tracks follow-ups
Who approves: [name], deputy: [name]
Escalation to the owner: discount beyond the price list, non-standard date, complaint
Metrics: time to first reply, share of cases with a next step, quotes without a follow-up
Evaluation: [date], participants: [names]

Pre-launch checklist

• We have chosen one type of sales case.
• We know our current figures for each metric.
• Materials (price list, templates, terms) are up to date.
• The approver and their deputy have time set aside.
• We know which data we do not send to the model.
• We have agreed a date for the evaluation.

How BOTrento Chief of Staff handles it

BOTrento Chief of Staff is a managed process for enquiries, quotes and other sales steps. You don't roll it out across the whole company at once. You start with a 14-day free trial, with no payment card and no API key of your own. The 14 days begin with your first successful data upload (import or mailbox connection), and you get 7 days to prepare.

• Human approval: nothing is sent until one of your people approves it. A person always approves email and quote drafts.
• Your own API keys: you enter your own language model keys (for example Anthropic Claude, OpenAI or Google Gemini) and pay the provider directly for AI usage, with no markup. The plan covers the service itself: setup, integration, oversight, escalation and reports. Keys are encrypted, only the last 4 characters are visible and you can remove them at any time. See the help page on your own API keys.
• Quick start: we reply within 24 hours on working days, prepare a demo on your data within 7 days, and deployment takes days. You can see how the process works in practice in the process demo.
• Oversight: monitoring runs 24/7, and we respond to incidents on working days within 4 hours.
• A clear guarantee: it covers taking over and fixing defects in the delivered process, plus the agreed process KPIs. It does not cover business results.
• No lock-in: if you leave, you get an export of your data and outputs, a description of the processes and an agreed date for removing access and copies.

To give a sense of scale, in our internal pilot the process handled 35 enquiries, 18 emails, 24 quotes and 15 actions in 7 days. That was an internal test, not a client reference. You add a payment card only at the end, if you want to continue. Without payment, access is blocked after the trial until you pay, but data export stays available and we keep your data for 30 days. The plans are on the free trial page.

Data, cost and responsibility: what to ask a supplier

Before you sign with any AI assistant supplier, get clear answers to these questions:

• Where is data stored and who processes it? BOTrento stores data on servers in Germany (EU). The text needed for a task is processed via API by the model provider you choose with your own key. Commercial API providers do not use this data for training. Still, an EU server that calls an external AI API is not a closed computation, so it's sensible to describe it that way in your internal documentation.
• What should not go to the model? Agree rules for sensitive data and check them against GDPR. The overview of what an assistant should and should not handle offers a practical starting point. For specific cases, consult a lawyer.
• What makes up the price? Separate the service fee, model usage and the time your people spend on approvals. How much an AI assistant costs a business shows how to work it out, and AI costs in your business explains how to keep them under control.
• What happens if you exceed capacity? Ask whether you are warned in advance and whether any increase needs your consent.
• Who is responsible for an error? Clarify what the supplier guarantees and what stays with you.