Accelerating analytics in the increasingly dynamic retail market was a key challenge the client faced. The company’s traditional approach didn’t meet modern requirements for rapid response to market changes and competitors’ actions.

The company’s strategic development was built on incomplete or outdated data. Figures and reports didn’t always allow for assessing the true performance of business units and specific sales managers, and SKU statistics were difficult to translate into concrete recommendations and actions. Decisions were often made less based on the information gathered by analysts than on collective brainstorming and pressure from influential employees.

Having immersed ourselves in the customer’s processes and analyzed the logic of choosing a development strategy, we proposed introducing an AI tool that combines the functions of the analytics department.

To do this, we have developed an analytical system based on LLM, into which current reports, plan-fact documents, price lists and financial indicators of the enterprise are loaded.

Then, the NLP-enabled dialog system allows an employee to ask the system a question in any form and within 30 seconds receive a structured answer based on a comprehensive in-depth analysis of the relationships between downloaded documents with links to specific numbers and indicators.

Detailed analytics using AI allows you to:

• assess the sources of lost profit according to plan-fact;
• identify the most and least effective units and managers;
• understand the geography and dependence of sales on SKU;
• identify problematic SKUs and sales channels;
• calculate profitability and ROI in general and separately by channel.

The AI ​​analyst will also offer options for strategic development, point out weaknesses and justify the need for certain measures.

Test runs on historical data demonstrated the high effectiveness of the developed solution. In situations where a historical decision proved beneficial for the company, the AI ​​assistant made the same or a very similar decision, and its predictions matched the actual results. Conversely, in cases of decisions that turned out to be unsuccessful in reality, the AI, using the same data, recommended a strategy that could have led to a different, more beneficial outcome for the company.

Proof of Concept testing is currently ongoing.