AI reporting and connecting to company data
For the people who assemble a report from several files and systems every month, and for the technical person who has to connect it all. The most technical module, and the one a trainer without a developer’s background usually cannot show.
One sentence that sorts the topic out
AI sees whatever the account it runs on sees. Tidy up access first, connect second.
An assistant connected to a database or CRM gets no new permissions. It uses the ones its account has. If that is an administrator account, the assistant can do anything. So in this module everything that reads company data runs on a separate, read-only account, and participants learn to read and verify what AI wrote before anyone trusts it.
What we practise
Plain-language questions to your data
“How much did we sell in region X versus last quarter?” How to ask so the answer can be checked, and how to ask for the route to the result, not just the result.
SQL and SOQL written by AI
AI writes the query, the participant reads it: where the data comes from, which filters, which joins. A check on a small sample. You do not need to write SQL; you need to be able to read it.
Formulas and pivot tables
Excel and Google Sheets: a formula described in words, written by AI, tested on three rows. A pivot table from a description, not from clicking.
Power BI and Looker Studio
A measure, a calculated column, a chart from a description. What AI prepares and what you still have to click. On your tool, if you have one.
Connecting through MCP
An assistant (Claude, ChatGPT, Copilot) connected to a database or CRM through an MCP server, on a separate read-only account. I show it on my own server, then on a copy of your data if you wish.
A recurring report
The data pulls itself, AI adds the commentary, the result lands in email on Monday morning. What a user can do, what the technical person can do, and what needs a developer.
Verifying results
A control total, a comparison with a report you already have, the same question asked a different way. The habits without which an AI report is pretty and wrong.
Permissions
How to set up a read-only account, which fields to exclude (personal data, salaries), how to log who asked what.
Two versions of the module
For users
For people who assemble reports and ask questions of data: managers, analysts, accounting, sales.
- Plain-language questions and reading the answers.
- Formulas and pivot tables with AI.
- Reading an SQL query AI wrote, without writing it yourself.
- Using an assistant connected to data that someone has already set up.
- Verifying results.
For the technical person
For the person in the company who has to connect and maintain it: IT, an administrator, an analyst with database access.
- Setting up an MCP server and the connection to the database or CRM.
- A read-only account, field scope, a query log.
- A recurring report: data pull, AI commentary, delivery. What to build and what to buy.
- How to receive queries from AI: performance, limits, personal data.
- What to do when a user gets a wrong result.
Example exercises
The report that takes three hours
A participant brings the report they build by hand every month. We break it into sources and steps, then check which steps AI can do with a formula and which need a connection to the data.
Read this query
AI writes an SQL query for a business question. The group translates it into sentences: from where, what, which conditions. Then we deliberately introduce an error into the question and see whether anyone notices.
A live question to the CRM
An assistant connected to a copy of the CRM through MCP answers questions from the room. We check every answer in the system.
Monday morning
For the technical person: a draft recurring report from data pull to email. A working prototype at the end of a workshop on company data.
Who it is for
- A company where someone assembles a report from several files and systems by hand every month.
- A manager who wants to ask questions of the data without waiting for an analyst.
- A technical person who has to connect AI to a database or CRM and does not want to do it on an administrator account.
- A company with Salesforce that wants CRM reports without exporting to Excel.
What you take away
- A set of prompts for data questions, formulas and reading queries.
- A checklist for verifying a report from AI.
- Your worst monthly report broken down into steps, with an assessment of what can be automated.
- For the technical person: a connection design (account, scope, log) and a prototype recurring report, if the workshop ran on your data.
- A certificate, 30 days of questions by email, a session after a month.
- If you wish, a quote for implementing what came out of the workshop.
Questions about this module
Do participants need to know SQL?
No. In the user version we teach reading the queries AI writes, not writing them. The technical version assumes the person already has database access and basic orientation.
Do you connect to our real database?
Not in a standard workshop: I demonstrate on my own MCP server and sample data. A workshop on your data is prepared separately: a copy or a read-only account, sensitive fields excluded, a confidentiality agreement.
We have Power BI / Looker Studio / only Excel.
I adapt. The principles stay the same; only the tool where the chart gets made changes.
Will the recurring report be built during the training?
A prototype, yes, if it is a workshop on company data with your technical person. A production version with error handling and monitoring is usually implementation after training.
We have Salesforce. Does this work with SOQL?
Yes. I show SOQL queries, Salesforce reports and connecting an assistant to the CRM. Before you connect anything, I recommend the free permissions audit.
After the workshop you have a list of things to build. If you want, I build them.
A recurring report, AI connected to your CRM or database, email and documents. A separate site about AI implementation.
Ask about training for your team
Tell me the department and the headcount. I reply within one business day with questions or straight away with a proposed programme. Quoted after a call, depending on format and headcount.