Project Sophia, take two :)
- Ana Inés Urrutia
- hace 7 minutos
- 5 Min. de lectura
At ColorCloud, I talked about Project Sophia. It was ambitious. Also? Kind of messy. The tool looked cool on paper, but I didn’t have proper access. Now that I’ve actually had help digging in, here’s the real version — no guessing, no fluff.
Project Sophia — now called Business Research Agents — is Microsoft’s new AI thing built into Dynamics 365 - but also accessible with external data (see examples below). It’s not just another Agent that summarizes stuff; it’s more like a space where you can ask real business questions and get answers that actually make sense.
How are our sales looking this quarter compared to the same time last year — by region and product line?
Normally, to answer this question we need a mix of CRM reports, Excel exports, Power BI dashboards, and Teams threads. With the agent, it’s one question and one answer — context included (assuming you know how to ask it right. Prompting still matters — a lot.)
Sophia got a new name (hello, Microsoft marketing). It’s now live as part of a broader suite of Business Research Agents. The first one out is the Sales Research Agent, embedded in Dynamics 365 Sales.
What makes Business Research Agents cool - IMHO + example:
Feature | Translation | Example |
Immersive experiences | The interface adapts to what you’re exploring. | Show me hiring trends by department |
Cross-domain thinking | You can ask something that touches HR, Finance, Ops — it pulls the pieces together. | Are delays in onboarding affecting project timelines? |
Understands business concepts | It recognises terms like “headcount,” “attrition,” “pipeline” without needing extra context. | How many people left in Q1 and what were the common reasons? |
Acts like a colleague | It suggests things you didn’t ask but might want to know — like a smart colleague would. | Here’s the data, but also — notice this spike in turnover in one region. |
Knows you well | It understands your structure and naming. You don’t need to explain what “Region North” means. | How many people joined last quarter per region? |
Business Research Agent vs. Power BI vs. Copilot/LLMs
If you’ve used tools like Power BI or Copilot (or even just ChatGPT with your company data 👮 ), you might wonder where BRA fits in - honestly the naming is not great. Aren’t they all just ways to ask questions and get answers? Kind of — but not really. Here’s how they differ in terms of what they’re good at, how they work, and what they expect from you as the user.
Feature / Capability | Business Research Agent | Power BI | Copilot / GPT-based Assistants |
Input style | Natural language (questions) | Drag/drop dashboards, DAX, visual interactions | Natural language (chat style prompts) |
Output style | Narrative answers + insights + data context | Visual dashboards, charts | Answers, explanations, generative text |
Data source | Deep enterprise data estate, real-time connections | Static or semi-live datasets | None (unless manually connected) |
Domain intelligence | Built-in business process logic (HR, finance, etc.) | Minimal — depends on how the report is designed | General knowledge, not process-aware |
Cross domain analysis | Yes – Finance + Sales + HR, etc. | Not native – requires custom data models | Not native – lacks business schema |
Enterprise context | Understands your org structure + data quirks | Not inherently – requires setup | None – unless manually trained/fed |
Collab ready | Designed to support business leaders' decision flows | Can be shared, but static | Collaborative for writing |
Speed | Very high — ask one question, get strategic answer | Medium — requires navigation + filters | High, but often generic and non org specific |
Setup required | Medium — needs access to company data sources | High — data modeling, visuals, publishing | Low — out-of-the-box |
Ideal user | Business leaders, decision-makers | Analysts, data-literate users | General knowledge workers |
Example | “Which sales region underperformed last quarter and why?” | “Show me total revenue by region” | “How do I write a performance review?” |
I still couldn’t get my own environment to work. (Yep. Still not freaking working, all setup in every place is set the right way, everything seems to be ok but then the agent is not cooperating.). But I annoyed people so much that they finally gave in to help me. Shout out to Chris Hansen | LinkedIn for the support and patience.
We tested two different examples, one very simple and the other one was a little bit more complex. You'll get it in a minute.
Example 1: Membership data.
Membership about what? We don't know, we generated it with AI and started testing randomly. Chris has already worked with it, so he was leading :) You can see below the spreadsheet...
We did some cool research on how this works, and asked questions. We didn't get very creative with the questions, but I'm honestly surprised with the answers and some of the guidelines that provides, for example opportunities for upselling. It was better than I expected.
Example 2: Dynamics365 HR employee - dummy - data
Through Data Management in Dynamics 365 Finance and Operations I extracted the file called Employee V2, lots of information. The file has a lot of fields, see below a screenshot of it.
We asked a random completely unrelated question and it turned out well, so I guess it did an interpretation of what we wanted to know. If you pay attention you will notice that we tried to fix the day count (for time of employment) from days to years and it didn't work well. I guess this is part of the learning curve of the product... Regardless there are some cool stuff happening and without any build effort on BI or anything like it. It understands the data without me explaining and also generates the research providing useful insights.
As you can see in both examples takes a simple Excel file and reads through the data, analyzing some context and providing actionable insights. I like that very much. As you can also tell after the second video is that some queries may not respond as good as we expected, we don't know if that is a consequence of the prompt or the solution. We will have to continue testing to find out, and keep an eye on
Business Research Agents might actually change how we explore information — especially in the messy places where data’s all over the place, answers are vague, and you don’t have time to dig through it all.
I’ll keep testing. But I wanted to get this out now — especially for those who sat through my session and deserved better. Some AI tools overpromise. This one? It’s starting to deliver.
Microsoft official resources:
And again, thanks to Chris Hansen - who helped a lot :)