Nitrio: ML-powered Intent Extraction Platform

Nitrio is an artificial intelligence company that empowers sales & marketing teams through their state-of-the-art, data-driven NLP solution for sales optimization.

Challenge

Nitrio required a modern and robust ML-based intent extraction platform due to their existing natural language processing platform relying on manual rules and heuristics-based models.

This caused both infrastructure and team performance bottlenecks and inefficiencies, and did not accommodate the company’s growth.

Additionally, the current platform was not providing the needed level of accuracy for sentiment analysis of representative-to-lead messages, resulting in a significant amount of messages being outsourced to a third party for manual analysis.

The issues with the existing infrastructure were further compounded by its tight coupling between services, increasing their dependencies and leading to data quality and data consistency issues.

Nitrio’s platform was designed to help analyze incoming rep-to-lead messages to identify their intent and collect data about the performance of each sales representative. This data was used to create data-driven strategies for Nitrio’s clients. The platform had to process many different types of emails and messages, making the accuracy of sentiment analysis a priority. If the system was unable to accurately determine the message’s intent with 95% accuracy, the task would be handed over to a third-party provider, resulting in increased service costs. This reliance on manual processes led to a number of problems, including bottlenecks, scalability issues and stifled growth.

In order to take their product to the next level, Nitrio approached Provectus to design and create an automated machine learning-based intent extraction platform for sales optimization.

Solution

Provectus designed and constructed a machine learning platform for intent extraction with high accuracy.

Manually developed regular expressions were supplemented to guarantee the accuracy of analyzing incoming rep-to-lead correspondence and messages. Rather than having to maintain over 4,000 individual regular expressions, the single model was deployed, maintaining the same F1 score.

Crowdsourced data annotation and an Active Learning workflow replaced the need for developing and sustaining regular expressions.

The ML platform was created using advanced neural networks and natural language processing built on Tensorflow. Automation of data annotation, training and evaluation within a deep neural network was incorporated. The platform was located on a separate EC2 instance and operated with hydrosphere.io. Amazon SQS managed all inbound and outbound messages from the ML platform, reducing the complexity and overhead associated with such transactions.

Amazon CloudWatch was employed to provide continuous monitoring, allowing Nitrio to access all necessary logs, metrics and events.

Outcome

Nitrio implemented an ML-powered intent extraction platform, which enabled them to markedly improve their daily throughput by 50%, reduce manual labor by a factor of 5, and decrease operational costs by up to 20%. It also led to a heightened accuracy in intent analysis, allowing the company to provide more effective sales strategies for their clients and offer guidance to their sales representatives. As a result, Nitrio’s ability to attract enterprise clients has grown significantly.

The ML platform also helped to overcome scalability issues, since the team no longer had to rely so heavily on third-party sentiment analysis services. Moreover, it eliminated structural bottlenecks.

The platform’s accuracy enabled the sales representatives to focus on communication with leads, instead of spending time manually processing emails to find out if a prospect was interested in their product. This enabled Nitrio to reach out to more clients, improving their prospects.