AI-powered data imputation boosts sales performance
Sigmoid leveraged a neural network-based deep learning library to predict the missing data values with up to 98% accuracy to help a semiconductor manufacturer boost its sales performance.
Business Challenges
The customer is a leading semiconductor manufacturer that produces chips for high-growth markets such as mobility, automotive, computing, consumer, and industrial IoT. Close to 500K new data records for fields like chip description, cost, chip size, etc., were added daily into the product master data table for cost estimation, and revenue forecasting. However, this master table had many missing values and data quality issues owing to gaps in data management. Filling in these missing values manually by more than 15 global teams lead to high man hour-costs and significant errors.
Sigmoid Solution
Sigmoid adopted a classification-based approach and used Amazon’s neural network-based deep learning library, Datawig, to create prediction models. The deep learning library helped predict the values of missing string data (categorical, numerical and textual data) with high accuracy. The prediction model was productionized and optimized on AWS Cloud and stored on S3. SageMaker allowed for productionizing classification-based ML models, defining training, and prediction pipelines, inference jobs, and publishing model artifacts and versions published. Only approved models were plugged into the prediction pipeline, whereas rejected models were trained again. Email notifications for clients were enabled to notify the confidence value of the predicted input data.
Business Impact
Sigmoid’s solution saved time, cost, reduced manual efforts and brought a significant improvement in revenue forecasting. The new solution was 90% more efficient than the earlier charted method of manual data correction.
17% increased
closure of sales opportunities
6M+ new data records
analyzed accurately in real time
<0.5 standard deviation
for numeric (cost) columns