.png)
This paper outlines how to setup and run a continuous learning model environment to increase customer retention. We explain the mechanisms behind the set up of the continuous learning platform, preparation of the data sets, and creation of a “listening” mechanism that optimizes machine models continuously in order to identify behavior changes and subsequent churn predictions in real-time.
Our customer churn solution starts with a suggested initial data dictionary, which will be tailored and modified by your team. Using TAZI Profiler you can evaluate the quality of your data by determining feature relevances and other statistics, the Profiler also provides automated data processing so that the initial model is created rapidly, and it is accurate and successful. Automated feature preprocessing and feature selection reduces overfitting, new feature generation based on transformations or combinations of other features enable the initial models to be successful. Usually, days of data scientist time are spent writing and maintaining these pre-processing scripts. Note that you can keep the Profiler running continuously to report and validate your data in production.
Initial continuous learning models are then created based on known outcomes. The models’ machine learning and business KPI performance can be verifed through TAZI business dashboards. The model explanations in TAZI Models & Insights allow the subject matter experts to understand, rapidly validate and own the models.
The models listen to incoming inputs and create real-time predictions. Predictions with reasons are provided instantly to the business user. In parallel, the models are continuously monitored for data and model drift to ensure accuracy and confidence in the model’s performance.
Both the data and business teams receive real-time alerts on model performance as well as data drift if thresholds are exceeded.
If performance starts dropping, both data and business teams have the ability to inspect changes in data through the Profiler and dashboard. Changes in models can also be expected through easy-to-understand model explanations in Live and Models & Insights. These inspections help data and business users to identify the root cause of the problem and correct it. The problem may be with incorrect updates on a data field. Also, there may be too little data to learn at the beginning of a change in patterns. The business can update the existing model parameters very quickly or provide human-in-the-loop feedback when new patterns emerge and they are not yet represented enough in the new data. They can also go back to a previous working model in time and take it into production instantly.
TAZI's underlying continuous learning models can listen to dynamic inputs and are updated automatically to increase their predictive accuracy.
In the case of retention optimization, the system can listen for the following dynamic inputs: competitor pricing, economic, political, environmental data, payment method, number of and content of in and outbound calls, emails, messages, the credit score (if allowed by the state), customer review score, claim information. In addition, driver, vehicle, location, agency, payment method, policy, and all product information can be input consumed by the models. Features' values, as well as their historical values, changes, or combinations, can also be consumed. When data changes dynamically, automated continuous learning enables models to learn from new data immediately and provide improved performance over traditional batch learning models.
TAZI's underlying continuous machine learning automatically ignores unimportant features as your customer behavior evolves. Model complexity also changes automatically to prevent overfitting. With batch model training and feature generation, feature selections and model complexity decisions need to be reworked for each new model training. With TAZI, this work occurs automatically with automated monitoring and alerting to ensure quality decisions are provided.
Another key difference with TAZI’s continuous learning process is that models can learn from the decisions and outcomes provided by existing batch-based models running in production. Then as these batch-based models require updating, the continuously updated TAZI model could be promoted into production after monitoring its performance within the dashboards.
When this option is deployed, instead of live continuous learning, you can have a continuous learning backup model(s) and replace the deployed model with the continuous learning model after quick checks on the continuous model performance and quality. This would result in a much faster reaction to change with much less effort than with batch machine learning.
Imagine a situation where the new model is ready to replace the old version in a fraction of the time and effort of the traditional batch methods. Imagine how much more productive your data science team will become solving new problems vs maintaining and updating stale models.
Models created or updated with continuous learning are also more stable than batch learning models that are created from scratch each time. This is primarily due to the infinite number of possible machine learning solutions for a given dataset which often results in two consecutive models created using batch learning to be not necessarily similar. For example, one batch ML model may use change in competitor pricing and location of the customer, while the next month’s model trained from scratch with the added data, might use not these but completely unrelated features such as the number of outbound calls and customer review score. Unstable batch learning solutions confuse the business and hinder them from taking continuous, consistent and useful actions with machine learning.
TAZI allows the formation of different user groups and also roles. People in the same groups can share data, configurations, models and results so that communication between data and business teams becomes much easier. The cross-functional teams can exchange and implement ideas easily. Quick creation, monitoring and deployment of models allow fast-fail. Users can experiment with and deploy new models with different inputs, KPIs, or data. When needed they can also create models with different additional purposes easily. AI is a dynamic journey for any organization. Continuous learning AI models, supporting dynamic business processes and business people who own and control both processes make the AI useful for the whole organization, in a much shorter time and continuously.
Artificial intelligence (AI) is a source of both huge excitement and apprehension, transforming enterprise operations today. It is more intelligent as it unlocks new sources of value creation and becomes a critical driver of competitive advantage by helping companies achieve new levels of performance at greater scale, growth, and speed than ever before, making it the biggest commercial opportunity in today’s fast-changing economy.
TAZI is a leading global Automated Machine Learning product/solutions provider with offices in San Francisco. TAZI is a Gartner Cool Vendor in Core AI Technologies (May 2019) and is considered as "The Next Generation of Automated Machine Learning” by Data Science Central.
Founded in 2015, TAZI has a single mission which is to help businesses to directly benefit from Automated Machine Learning by using TAZI as a superpower, shaping the future of their organizations while realizing direct benefits like cost reduction, increasing efficiency, enhanced (dynamic) business insight, new business (uncovered), and business automation.
Through its understandable continuous machine learning from data and humans, TAZI is supporting companies in banking, insurance, retail, and telco industries in making smarter, more intelligent business decisions.
TAZI solutions are based on a most compelling architecture that combines the experiences of 23 patents granted in AI and real-time systems, proven at different global implementations.
Some unique differentiators of TAZI products are:
