The short answer
- An internal assistant connects to your systems through their APIs or a read-only database connection; it can look, never change.
- Managers ask in plain language and get an answer, a table, a spreadsheet or a chart in the same conversation.
- Access is limited to an allow-list of staff and every question and query is logged.
- Nothing is used to train AI models.
What an internal assistant is
It is the same kind of AI assistant that answers your customers, pointed inward. Instead of a knowledge base of prices and policies, it is connected to the systems where your numbers live: ERP, POS, accounting, booking, CRM, Google Sheets, a custom API or your database. A manager asks "today's sales per branch" and gets a table; "October profit and loss as a spreadsheet" and gets the file.
What managers actually ask
- Today's or this week's sales per branch, per product or per channel.
- Profit and loss for a month, as a table or a spreadsheet.
- Which products are low in stock, and which are not moving.
- Bookings or appointments this week, and who is fully booked.
- Which customers or students have not paid, and how much is outstanding.
- A comparison: this month against last month, this branch against that one, with a chart.
How it connects to your systems
Systems with an API are connected through that API with authentication; the assistant can also read your API documentation and run read requests and aggregations on its own. Google Sheets are read directly. For a database, a read-only connection is set up with your IT team during implementation. The assistant queries only the data a question needs and works out totals, comparisons and rankings on the spot.
Read-only, staff-only, logged
Three rules make it safe. The connections are read-only: the assistant can look, never change. Only the people on an allow-list that you control can use it, through a staff-only WhatsApp line, the dashboard chat or private links; everyone else is declined. And every question and every query it runs is logged, so you can see who asked what and when. Nothing is used to train AI models.
Where it runs
In the cloud, or installed on your own servers with the database connection never leaving your network. Either way the AI model is reached through an encrypted API connection and receives only what a question needs.
How to get started
- List the five questions you ask most often and which system holds each answer.
- Decide who may use the assistant: owners, branch managers, finance.
- Give our team API access or a read-only database user for those systems.
- Test the five questions together before rolling it out to the team.