One of my clients operates in the Education technology industry, providing a multitude of courses for competitive exams in India. They sought a solution to increase revenue per customer without adversely affecting customer feedback, prompting them to engage our services for an end-to-end AI-based solution.
Our agreed-upon solution for the first phase involves providing detailed course recommendations to customer service executives during subscriber calls. The call is transcribed in real-time, and recommendations dynamically adjust based on the conversation.
The precise AI-based recommendations aim to enhance upselling, thereby increasing revenue. Although there are additional project phases, such as price adjustment and quality improvement, they are beyond the scope of this post for now.
During a presentation, the client\’s CTO showcased the extensive data available to my team for this problem. The data encompassed CRM data, geography, demographics, political information, school results, financial status, and more.
In response,I conducted a thorough inspection.I decided not to utilize nearly 40% of the provided data, leading to the client\’s disappointment. Despite the client\’s expectation of requesting more data, I justified the reduction, emphasizing specific criteria for data selection in AI models. Some of the criteria I used was:
Source: Trust in the data requires knowledge of its source for each data type.
Completeness: Data should not contain excessive missing values. While some missing values can be recalculated, not all can be accounted for.
Continuous Availability: AI models relying on the assumption of continuous, up-to-date data must ensure its ongoing availability.
Avoid Overengineering: Visualization and experimentation are crucial to determine the necessity of a particular data source. If the incremental change in results from adding less relevant data is negligible, it may not be worthwhile.
Share your experience if you have come across a similar situation.
*image by Freepik
![\"\"](\"https://www.mysticai.io//wp-content/uploads/2024/01/16-1.jpeg\")