How much does an AI assistant cost a business? What the price is made of and what to watch for

BOTrento Team8 October 2026 · Updated 12 October 20269 min read

What an AI assistant's price is made of

Whether it is a self-service tool or a managed process, you always pay for three things. The only difference is how much of it appears on the invoice.

1. The service

Process setup, connection to the mailbox and the records (CRM or spreadsheet), reply and quote templates, oversight of day-to-day running, error handling and reports. With a self-service chat, someone on your team does this work, so you do not pay the supplier for it, you pay with your own time.

2. Language model usage

Every draft email or quote means a call to the model through an API. You are charged by the volume of text processed. This item grows with the number of enquiries and the length of the source material, for example when the assistant reads long attachments or an entire email thread.

3. Internal time

Preparing materials (price lists, typical quotes, discount rules), training, and above all the ongoing approval of drafts. Approval is necessary in sales, because a human must decide on price, dates and sending. But it is time you need to budget for.

For more on how to estimate these items and track them month by month, see how to estimate AI costs in advance and keep them under control.

Service vs model: why it pays to price them separately

Some offers have a single flat price covering both the service and model usage. It looks simple, but you cannot see how much goes on work and how much on AI, and the supplier may add a markup on usage.

The second model is called BYOK (bring your own key), meaning your own API keys. The company opens an account with a model provider such as Anthropic, OpenAI or Google and enters the key into the supplier's system. Usage is then paid directly to the provider, with no intermediary. You pay the supplier only for the service.

How to create a key and enter it securely is described in the help page on your own API keys.

Hidden costs that get forgotten

When comparing offers, it pays to go through this list. Every item the supplier does not cover shifts to your team.

• Connection to your records. If the assistant does not write to your CRM or spreadsheet, someone has to copy the results by hand.
• Changes to templates and rules. Price lists and terms change. Find out whether updates are included or charged separately.
• Exceeding the limit. What happens when the busy season arrives and there are more enquiries? A surcharge at the end of the month, or a warning in advance and an increase only with your consent?
• Oversight and errors. Who notices that the assistant has stopped reading the mailbox? How quickly does the supplier respond?
• Time spent fixing drafts. A badly configured process produces drafts that have to be rewritten. Approval then takes almost as long as writing from scratch.
• Exit. Will you get an export of your data and outputs and a description of the process? Or will you start again from zero?
• Self-service. With a general-purpose chat you pay for a licence, but your people do the setup, copy the source material and keep track of follow-ups. We cover the difference in ChatGPT or Claude for your business, or a managed AI process.

How to calculate approximate monthly costs

Before you compare price lists, work out your own starting point. All you need is a spreadsheet and an afternoon.

1. Choose one process. For example, "equipment rental enquiry → quote → follow-up". Do not add up everything AI might one day do.
2. Count cases per month. Go through your mailbox for the last month and for the seasonal peak. Count sales cases, not individual emails.
3. Measure current time. For a few cases, note how long the reply, the quote and the reminder take.
4. Ask for a price breakdown. Service separately, model separately, rollout separately. If a supplier will not give a breakdown, that tells you something in itself.
5. Estimate model usage. With BYOK, set a monthly spending limit with the provider and adjust it after the first month based on actual figures.
6. Add internal time. Rollout as a one-off, approvals as an ongoing cost.
7. Set the yardstick for the trial run. What must hold after the 14 days for you to continue? For example, that every enquiry has a draft reply and every quote has a next step scheduled.

Calculation template

Questions for the supplier before you sign

Print these questions or send them by email. The answers will tell you more than the price list.

Hello,
before we decide, could you please answer the following questions:
1. What exactly is included in the service price, and what is charged separately?
2. Do we pay for model usage through you, or directly to the provider with our own API key? Is there a markup?
3. What is the billing unit (email, case, user) and what happens if we exceed it?
4. Does anything reach a customer without human approval?
5. Where do you store data, and who processes the text for the language model?
6. What do you guarantee and what not?
7. What will we receive when the collaboration ends?
Thank you.

Be wary of suppliers who promise business results. Only what the supplier genuinely controls can be verified and secured by contract: a working delivered process and agreed operational indicators, for example that every enquiry receives a draft reply. For what an assistant can realistically do in sales, see AI assistant for business: what it can really do in sales and what it can't.

How BOTrento builds its price

BOTrento Chief of Staff is a managed process for enquiries, quotes and other sales steps. The price is split so that you can see what you are paying for:

• Plan = service. Setup, integration, oversight, escalation and reports. Monthly, excl. VAT: Start €199 (100 processed cases), Business €399 (300 cases), Enterprise €999 (by agreement).
• The unit is a sales case. One case is one sales case actively analysed in the month, with at most five drafts. You get a warning at 80% of capacity, any increase happens only with your consent, and there is no unexpected surcharge.
• You pay for AI directly. You enter your own API keys (Anthropic Claude, OpenAI, Google Gemini, Groq, ElevenLabs) and pay the provider for usage with no markup. Keys are encrypted, only the last four characters are visible, and you can remove them at any time.
• A human approves. Drafts of emails and quotes are approved by a person. Nothing is sent without approval.
• Data. We store it on servers in Germany (EU). The text needed for the task is processed by the model provider you choose with your own key. Commercial API providers do not use the data for training. However, your own EU server calling an external AI API is not a closed computation.

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 upload of customer data (an import or a connected mailbox), and you have 7 days to prepare. All plans, Core, Team, Managed and AI, are included. You add a card only at the end, if you want to continue; without payment, access is blocked until you pay, while data export remains available and we keep your data for 30 days. You can start on the 14-day free trial page. To see how the process works in practice, view the process demo from enquiry to quote. For a sense of scale: in an internal pilot, the process handled 35 enquiries, 18 emails, 24 quotes and 15 actions in 7 days.