One of the most common AI applications in sales is lead qualification. Many companies face a huge number of inquiries, including spam and low-potential inquiries. As a result, requests from high-potential customers often get buried among low-priority ones. For one of our clients in the EU, we developed an AI-powered lead qualification system that automatically identifies high-potential prospects and routes them to Key Account Managers.

Challenge

The client needed to build a lead qualification process based on lead potential, so that the most valuable leads would be routed to Key Account Managers — and to do this as quickly as possible without delaying the communication process.

Solution

We implemented an AI-driven lead qualification workflow integrated with Bitrix24. Finding the right qualification strategy required several iterations.

Our first approach was to evaluate companies based on financial indicators such as registered capital, number of employees, and annual revenue. AI was responsible for gathering the data, while the final qualification was performed by a deterministic scoring algorithm.

Soon we discovered that business registries in key European countries didn’t allow AI access, and the cost of accessing them via API was prohibitively expensive. As a result, the model could estimate company capital in fewer than 30% of cases, while employee count and revenue were rarely available. The overall accuracy was therefore insufficient.

At the next stage, we used a prompt based on the scripts that sales managers use for manual lead qualification. This approach delivered higher accuracy, but it also faced a lack of available data. Answers to many of the questions are easy to obtain during a phone call, but impossible to find through online search.

Finally, we exported the existing database of qualified companies and asked the AI to analyze it and create a company scoring strategy using only the limited set of parameters available at the lead generation stage. This approach proved to be both the most accurate and the most cost-effective, as it didn’t require state-of-the-art models and generated a minimal number of output tokens.

Technical Implementation

The solution was deployed on an on-premises Bitrix24 installation using the platform's REST API.

Although the optimized prompt delivered relatively fast responses, AI inference added noticeable latency. To ensure that lead processing remained responsive, we implemented an asynchronous architecture that allowed other automation workflows to continue while lead qualification was performed in the background.