Embed machine learning into an automated business process in under 10 minutes!
How Catalytic Predict Works
1. Select your data set
Use data being collected in an automated process, upload a spreadsheet, or integrate with a system using pre-built connectors.
Example: Use historical examples from an existing manager approval process as the data set.
Minimize the effort to aggregate and clean data sets. When business processes are automated on Catalytic, data from multiple systems can be synchronized and validated, as well as hydrated by input from human experts.
2. Choose what to predict
Train the machine learning model to make predictions for a field of your choice.
Example: Target the "Do you approve?" field for a prediction.
Target fields can include boolean, multiple choice, integer and decimal data types. Catalytic Predict will automatically select the appropriate binary classification, multi-class classification, or regression model to train for a given dataset. Each of these models is based on a logistic regression machine learning algorithm.
3. Identify relevant inputs
Make the prediction based on relevant factors, including both structured data and unstructured text using natural language processing.
Example: Include factors such as the request description, cost, expected ROI, and title of the person making the request. Exclude non-relevant factors like the date of the request and the gender of the person making the request.
Boolean, multiple-choice, integer, decimal, dates and natural language data types can all be used as inputs to a predictive model. Our natural language classification is based on a bag-of-words approach using a logistic regression model. The algorithm automatically determines the set of terms that are most predictive of a given category, and calculates weights to apply to each term. Natural language features are considered along with all other input fields that are provided to the predictive model.
4. Add prediction to automated process
Every prediction includes a confidence rating that can be used to determine whether the decision should be made automatically or routed to a person with a recommendation.
Example: If the confidence rating is over 90% and the cost is less than $10,000, set it to approve automatically. Predictions with lower confidence and higher costs could be routed to the manager to review.
Embed the predictive model into an automated business process in less than 5 minutes and ensure that the predictions are delivered instantly at the time of need. Predictions and confidence ratings generated by predictive models can drive workflow and variable conditions, be presented to people as recommendations, and automatically inserted into records in other systems.
5. Increase accuracy automatically
As the automated process runs and additional data is collected, the model retrains itself, increasing accuracy of predictions and responding to changing business conditions in real-time.
Example: Every manual approval is fed back into the model automatically to increase the confidence rating for future requests that have similar factors.
Predicted model accuracy is calculated when first trained and whenever the model automatically retrains itself. Accuracy of the overall model is always presented transparently. The factors of the model and the underlying data can be modified at any point to increase accuracy or remove bias.