Azure OpenAI Data Analytics Bot Helps Darnytsia Drive Ad-hoc BI Insights

Business Challenge 

As one of Ukraine’s leading pharmaceutical enterprises, Darnytsia has to deal with vast volumes of data and prepare BI reports regarding the company’s operations, supply and demand planning, sales analytics, competitive intelligence, and more. The company’s corporate data analytics department struggled to find the time and resources to address the ever-growing demand for BI insights from other business divisions. 

For this reason, our client started searching for an intelligent solution that would: 

• Democratize access to data analytics across the company  
• Rapidly address a large pool of internal data analytics requests  
• Optimize the time, resources, and costs required to prepare BI reports  

Initially, Darnytsia considered the implementation of Microsoft Power BI Copilot. Yet, the organization-wide adoption of Power BI Copilot at that time turned out to be less cost-efficient for the client, as the requested analytics reports varied in type, complexity, and timeframes. Thus, it was essential to find a solution with the perfect cost and capabilities ratio that would deliver swift ad-hoc BI insights for Darnytsia.   

Solution & Business Value 

We developed a Data Analytics Bot powered by the Azure OpenAI service, which serves as a smart and intuitive corporate GenAI assistant, providing swift on-demand analytics to any company employee. As a result, the solution has brought the following benefits for Darnytsia:   

• Unprecedented speed-to-insight – the data analytics delivery time was reduced from weeks to mere seconds  
• Optimized time and efforts of the analytics team, who can now focus on valuable strategic tasks instead of covering an array of internal ad-hoc requests  
• Significantly reduced costs for creating and maintaining BI reports, along with a cost-efficient pay-as-you-go pricing model of the solution  
• Democratized access to data analytics – there’s no need for advanced Power BI or SQL knowledge – any employee can use the Bot and request the analytics they need via simple text prompts  
• Near real-time insights at the client’s fingertips – the solution was deployed as a corporate chatbot that is available in desktop and mobile versions of Microsoft Teams  
• Language-agnostic Analytics Bot that perfectly understands human intent and responds accordingly  
• Reinforced data security and privacy, as Azure OpenAI does not apply customer data for model training and improvement  

We deployed the Data Analytics Bot for Darnytsia in just 2 weeks, and currently our team provides ongoing support services to help the Bot learn, minimize errors, and improve its overall performance. 

Being satisfied with the project outcomes, Darnytsia decided to take a step forward and requested engineering a mobile version of the Analytics Bot with speech recognition capabilities. We continue the project with Darnytsia to integrate voice-based commands and seamless verbal responses to the solution, aimed at enabling convenient on-the-go access to ad-hoc insights for the client. 

Technical Details  

The project began with a preliminary assessment of the client’s IT infrastructure and databases. Consequently, we engineered the Data Analytics Bot, which is based on the retrieval-augmented generation (RAG) framework and built on the Azure OpenAI service. 

When the chat session starts, users can enter their text prompts that may vary in complexity – from simple “trends of paracetamol sales in 2023” to more complex, such as “compare the sales rates of Darnytsia’s key products against five major competitors”.  

Once the prompt is submitted, the Bot instantly detects the language to tailor the response accordingly. Next, it generates an SQL query to access the client’s database and provides a link to an Excel report with the relevant analytics, which is followed by a concise text-based summary of the findings. After the analytics are delivered, the Bot asks whether the data needs to be visualized. Users can enter prompts such as “Graph” or “Chart” and the Bot will instantly provide a link to the visual representation of the insights.  

The Data Analytics Bot is stored in the client’s Microsoft Azure tenant, and comprises multiple GPTs, each being responsible for a specific stage of the data analytics flow – from language detection to data visualization. 

The conversational capabilities of the Bot are limited to the context of the client’s databases, meaning that it only provides answers to analytics-related user prompts, and cannot engage in discussions on other topics. If the supplied text prompt is vague, the Bot will provide an extensive reply and ask for further clarifications from the user. In case of data-related errors and inconsistencies, the developers can check and resolve the issues, as the Bot tracks all of the chat session information, such as time, session/user ID, SQL queries, message index, and the generated reports.   

The solution can be rapidly deployed and scaled across businesses of any size – the implementation time for small-scale databases requires just 2 days, and up to 2 weeks for complex data warehouses (DWHs).   

Furthermore, the Data Analytics Bot features multiple deployment options and a range of customizations. In addition to a Microsoft Teams-based chat, the solution can be integrated with Slack, Google Chat, Viber, Telegram, and other messaging apps. The list of custom extensions includes the addition of new databases, an increase in the analytics scope, new styles of visual diagrams, as well as the addition of corporate logos, branding, watermarks, etc.  

Technologies & Tools 

• Azure OpenAI 
• Azure Bot Service 
• Azure SQL Database 
• Azure App Services 
• Microsoft Teams