Follow me down into the rabbit hole, where I take a look at a new AI-based project called Project Sophia. Project Sophia looks to become a perfect companion for Business Central, if it works? Check out the video:
In this video, Erik explores Project Sophia — Microsoft’s new AI-first business application — and demonstrates how it can be used to analyze data exported from Business Central. While still in preview and limited to US-based tenants, the concept behind Project Sophia offers a fascinating glimpse into how AI-powered data exploration might integrate with Business Central in the future.
What is Project Sophia?
Project Sophia is described on Microsoft Learn as a new generation of AI-first business application that helps users with complex, cross-functional tasks. It offers a new way to discover, visualize, and interact with business data.
The key features include:
- AI-powered research canvas — Create exploratory journeys and understand business data with AI-generated charts and insights without manually mining data or creating visuals
- Large language model powered blueprints — The app provides quick insights and recommendations, suggesting next steps
- Iterative exploration — Start with a question, explore different options, and continue iterating until you reach the best possible outcome
To try it out, you can visit projectsophia.microsoft.com, though it’s important to note that Project Sophia is only available with a US-based Office 365 tenant during the preview period.
Feeding Business Central Data into Project Sophia
Before recording the video, Erik prepared data by going into Business Central (using the Cronus demo company) and exporting several datasets:
- Chart of Accounts (Excel)
- General Ledger Entries (Excel)
- Items (Excel)
- Vendors (Excel)
- Customers (Excel)
- Customer/Item Sales Statistics report (PDF)
These files were then uploaded into Project Sophia’s workspace for analysis. The platform accepts multiple file uploads and allows you to specify the business domain — in this case, Finance.
Exploring Items Data
After uploading the items Excel file, Project Sophia immediately went to work — “charging your data’s journey,” “planning the next step,” “crafting insights,” and “laying groundwork for action.” The AI generated several visualizations and insights automatically:
- Proportion of items in inventory where substitutions exist
- Relationship between unit cost and unit price
- An overview describing the Excel file contents
- Identification of the top three inventory items with the highest unit price
- Suggested next steps like “generate email draft” or “analyze the correlation between unit cost and unit price”
Clicking on a suggested next step generates a new canvas with additional analysis — the concept being that you build an exploratory journey step by step, with each canvas building on the previous one.
Analyzing the Sales Statistics PDF
Erik then uploaded the Customer/Item Sales Statistics report as a PDF. Project Sophia was able to parse the PDF and generate insights including:
- Top selling items
- Total profit and profit percentage breakdowns
- A chart showing total pieces sold for each product
However, Erik noticed some confusion in the AI’s interpretation — it occasionally mixed up the direction of sales. For example, it said a customer “sold 43 lamps” when they actually purchased 43 lamps. This is a good reminder that AI-generated insights always need human verification.
When asked “who purchased the Conference Bundle,” the AI correctly identified that it was purchased by two customers — A. Datum Corporation (18 units) and Trade Research (26 units) — totaling 44 units. Erik noted the math actually checked out, which is worth verifying since large language models are notoriously unreliable with arithmetic.
Multi-File Analysis: Who Owes Us Money?
For the most interesting test, Erik created a new workspace and uploaded multiple files at once: the Chart of Accounts, General Ledger Entries, and Customers. He then asked a simple but powerful question: “Who owes us money?”
Project Sophia generated:
- A visualization of Balance Due (LCY) for each customer
- Net change in Chart of Accounts over time
- Identification of the top three customers with the highest balance: School of Fine Art, R. Cloud, and Alpine Ski House
Erik confirmed these results were accurate. He then asked about the overall financial health of Cronus, and the AI provided a reasonable assessment — noting the absence of recent transactions in certain accounts and outstanding balances from specific customers, while suggesting further analysis steps.
The Bigger Vision
While the current workflow requires manually exporting data from Business Central and uploading files, Erik highlighted the exciting possibility: What if Project Sophia had direct access to Business Central via an OData feed?
The concept would work like this:
- You provide the AI with a model describing all the data available in Business Central — item information, value entries, customer tables, and so on
- You ask a question in natural language
- The AI determines which data sources it needs to answer your question
- It pulls down that information directly from Business Central
- It generates blueprints and canvas visualizations with insights
Erik also suggested that to get the most out of this kind of technology, it might be beneficial to build specialized reports that dump data in a way that describes what we know about it — helping the language model process and interpret Business Central data more effectively.
Conclusion
While Project Sophia in its current preview incarnation may not be fully production-ready, the concept is genuinely exciting for the Business Central ecosystem. The ability to ask natural language questions about your business data and receive AI-generated visualizations, insights, and recommended next steps represents a significant shift in how users might interact with ERP data in the future. If you have a US-based Office 365 tenant, it’s worth visiting projectsophia.microsoft.com to experiment with it yourself — and perhaps start by reading the documentation first, which Erik cheerfully admitted he skipped.
