Data science transformation best practices

This video shows you how to give a brief overview of Pega best practices for data science transformation, which is an essential part of the adoption of Pega Customer Decision Hub™. Pega developed their best practices based on deep product knowledge and years of delivery experience with clients across the globe. Pega's best practices are an amalgamation of all the best bits that work well and help clients make the best of their Pega Customer Decision Hub AI capabilities.

Generally, Data Science Transformation represents a significant shift towards the comprehensive adoption of Machine Learning and Data Science practices. The primary objective is to enhance and optimize the customer experience across various touchpoints and interactions with the company, based on a deep understanding of the interplay between decisioning and data science in the context of customer interaction optimization.

 

By integrating these advanced practices, every customer interaction becomes more personalized, efficient, and effective.

Now, let us talk about what Data Science Transformation specifically means within the context of Pega Customer Decision Hub. Firstly, it involves the adoption of Pega Customer Decision Hub Machine Learning capabilities, which include both adaptive and predictive models. These capabilities enable anticipation of customer needs and behaviors more accurately. Secondly, there is the continuous optimization of predictions. This means constantly refining models to ensure they provide the most accurate and up-to-date insights. Thirdly, the enablement of Data Science resources is crucial. Data scientists need the tools, data, and infrastructure to perform their work effectively. Lastly, establishing governance for machine learning operations, (MLOPS), is essential. This involves setting up frameworks and policies to manage and oversee the deployment and operation of machine learning models, ensuring they are reliable, secure, and compliant with regulations.

Before diving into the details, make sure you understand these key concepts and terms.

Artificial intelligence, or AI, describes the system where machines learn from experience, adjusting to new inputs and potentially performing tasks previously done by humans. More specifically, it is a field of computer science dedicated to simulating intelligent behavior in computers.

Machine Learning

Machine Learning is a field of AI involving algorithms that enable computer systems to iteratively learn from and then make decisions, inferences, or predictions based on data. These algorithms build a model from training data to perform a specific task on new data without being explicitly programmed to do so.

Predictive Model

A Predictive Model calculates the probability of an outcome based on a set of input or predictor data. In Pega Platform™, you can create predictive models in Prediction Studio by applying its machine learning capabilities, as well as import or connect to models built in third-party tools.

Adaptive Model

A predictive model that is self-learning and which Adaptive Decision Manager updates continuously as the system captures new input data and outcomes. Adaptive Models calculate the propensities in Pega Customer Decision Hub in real time to obtain the relevance of each available action to a customer when determining the next-best-action.

Propensity

Propensity is the likelihood of a customer responding positively to an action. This is also calculated by AI. For instance, a click on an offer banner or acceptance of an offer in the contact center are considered positive behaviors.

Why is this topic important?

By using Machine Learning, organizations can unlock the full potential of Pega Customer Decision Hub. This enables the delivery of exceptional customer experiences that not only meet, but exceed customer expectations.

Data is a central asset in Pega Customer Decision Hub, and Data Science Transformation enables the extraction of valuable insights from this data. These insights can then be converted into actionable strategies through the use of Adaptive and Predictive Models.

Pega Adaptive Models play a crucial role here. They predict customer propensities, making every customer interaction more relevant and personalized. These models are fully automated, self-learning, and do not require pre-training.

The adoption of Adaptive Models is key to the successful implementation of Pega Customer Decision Hub. These models are preconfigured for immediate use out of the box and learn quickly, ensuring a swift and efficient integration.

Furthermore, clients can use their historical data to train Pega Predictive Models. They can also import in-house pre-trained models, such as those used for churn prediction, or integrate Pega Customer Decision Hub with real-time AI services like Azure SageMaker or Google AI. These scores can then be used as predictors in Adaptive Models.

To give context to the importance of adaptive models, remember that Pega Customer Decision Hub is described as 'the always-on brain.' This brain sits at the center of all channels and applications, collecting data from all of them.

For instance, what is the customer doing on the website, did they call, did they open an email? CDH combines this knowledge with all the historical information in the customer's profile or interaction history to determine their propensity to respond to potential conversations.

Based on this information, looking at each customer in each unique moment, CDH determines if there is anything to add value, and what is the next best action?

This might be linking to an education article, a service task, or a nudge to complete a process – but whatever it is, it is figured out in real-time using the models discussed today, and then delivered back out across any channel, inbound or outbound, and all of this happens in less than 200ms.

The customer then reacts to the next best action. Their reaction, whether positive or negative, becomes a new piece of information for the brain to use, to make the next decision for them or other customers. This feedback loop is exactly why Pega Customer Decision Hub is always on and always learning.

 

Take information about the customer and the actions as the input, predict for each action and treatment the likelihood that a customer will like the action, prioritize the candidates, and present the customer with the next best action. The response to this next best action then feeds back into the brain and becomes a set of powerful predictors for future interactions.

 

The key element of CDH's 'brain' is its adaptive modeling capability, which predicts the likelihood that a customer will like the action.

Why use adaptive models? Here is an overview of adaptive models and Adaptive Decision Manager.

 

The Adaptive Decision Manager automatically generates adaptive models – one for each treatment and action pair. They play a key part in the arbitration formula that generates the next-best-action by predicting the likelihood of clicks/accepts in real time.

 

The models continuously learn customer preferences from every additional interaction and are a closed-loop system that automates the machine learning (ML) model creation, deployment, and monitoring process.

The traditional modeling approach requires models to be trained with data, and the performance decays over time, so they need to be retrained and turned regularly. This process is typically time-consuming, needs frequent redeployment, and is hard or even impossible to maintain at scale.

In contrast, adaptive models optimize themselves on data that becomes more available over time. This hands-off, automatic model factory approach allows prediction of every action and treatment – hundreds or many more, and changing frequently.

This approach enables zero time-to-market for analytics-driven propositions, allows the enterprise to automatically react to frequent changes in customer behavior, and optimizes the system for successful outcomes.

Now that we understand the basics of adaptive models, let's take a deep dive into the operational aspect. Some stakeholders, especially data science experts, require further enablement to adopt Adaptive Decision Management in Pega Customer Decision Hub.

For those people, you can direct them to Pega Academy or cover the following topics with them directly:

- How adaptive models work

- 'Cold start' for models

- Predictors

- Monitoring of Adaptive Models

- Monitoring in Prediction Studio

- Action Performance Tracker

- Impact Analyzer

- Ethical Bias Check

- ADM Notifications

- PDS Tools

- MLOps

Given an adequate understanding of what Pega Adaptive Decision Manager.

(ADM) is, key operational stakeholders need to understand how to set up the adaptive modeling capability to be ready for business as usual (BAU) after go-live.

 

Even though the adaptive models are automatically created by ADM, there is still work for the team to do within an MLP, or project.

 

Following the Pega Express methodology, start with the Discover stage. During this stage, you identify resources to support this area of work, align on objectives and decisioning context, and understand the potential of data as predictors.

Then, during the Prepare stage, LDAs need to prepare Prediction Studio and start early-stage enablement.

The Build stage is where the adaptive predictions are configured, alongside other aspects of the Pega Customer Decision Hub configuration. LDAs need to define the candidate predictors and ensure that the outcomes are correctly configured to match the outcomes/responses from the channels in which customers interact.

In terms of business transformation, a business transformation consultant works with the client from end to end for an MLP. During the early stages, the work revolves around understanding the current situation and identifying gaps to be closed. Share the Pega best practice BAU operating model for adaptive modeling, and together with the client define a working operational architecture around the monitoring, reporting, and improvement of the models, to ensure sustainable success. You should also support enablement of the resources that are identified, to be responsible for the BAU of adaptive models too.

Once implementation is complete and you're moving into BAU, what does the role of a Pega data scientist look like?

This can be a part-time role. Normally, you can expect to assign 50%-100% of a full time resource depending on the project phase. Although referred to as a 'data scientist', the role can be filled by a data scientist, data analyst, business analyst, or LDA with AI/Modeling expertise.

To be effective in this role, the individual needs to have an intermediate level of knowledge in data science and analytics. They should be familiar with Pega AI, especially adaptive models, and capable of using Pega Data Scientist Tools, whether based on R or Python. Additionally, strong analytical skills are essential. The ability to extract meaningful insights from data and reports and convert these insights into actionable improvement ideas is a must.

Typical activities that are involved in this role include model monitoring and reporting, which is done using out-of-the-box tools as well as Pega Data Scientist Tools. The role also involves analyzing and diagnosing issues, and more importantly, identifying opportunities for improvement. Pega has also developed best practices for these processes, and more details can be obtained from various sources, such as the Pega Community website.

Collaboration is also important, because this person will work closely with other team members on model configuration, experimentation, and defining new requirements. Finally, they will liaise with the data team to ensure the availability and quality of predictors.

For an efficient operating model, the data scientist is typically assigned directly to the Pega Customer Decision Hub Execution team and works closely with them on an ongoing basis.

 

The model monitoring, reporting, and diagnostics use the ADM, the data mart, the out-of-the-box tools and charts in Prediction Studio, supplemented by external tools developed outside of Pega. One example of such a tool is the Pega Data Science Tools which is R or Python based, and can generate more comprehensive reports for large sets of models.

 

For more information about Customer Decision Hub's out-of-the-box tools, refer to Pega Academy or Pega Documentation.

For details on the Pega Data Scientist Tools, refer to the GitHub page linked to on the screen.

Pega has also developed best practices content on BAU activities for model monitoring, analytics, and optimization, which can be used to expandyour conversations with the clients. These cover questions such as:

- How frequently should the model reports be reviewed?

- How do I read the reports in Prediction Studio?

- What should I do if I see a drop in Clickthrough Rate in the last 2 weeks on a particular Issue/Group/Action? Do we need new predictors?

It is important to highlight that all of the BAU activities carried out by this data scientist, such as monitoring and optimizing the models, ultimately serve one purpose – to realize the overall business objective. For example, is the business intent to increase relevancy and clickthrough rate, or is it to increase conversion? Depending on the answer to this question, the adaptive models can be configured and adjusted to fulfill different purposes.

This is an overview of the end-to-end BAU process. The BAU tasks for a data scientist fall under 'Review & Optimize', as marked on screen.

Alongside the next-best-action Analysts or LDAs, Data Scientists use Prediction Studio in Pega Customer Decision Hub to review and monitor the performance of predictions and adaptive models. Reporting back to key stakeholders feeds back into the cycle seen earlier, because potential improvements and adjustments are identified this way and raised as future change requests.

They can also identify opportunities to improve and refine the marketing strategy and those opportunities can be fed back to key stakeholders who can then restart the cycle and request a future change request. This stage finally closes the loop.

 

The data scientist role needs to complete the Pega Customer Decision Hub Foundation mission and be certified as a Pega Certified Data Scientist.

 

Refer your client to Pega Role Hub for the latest information.

You have reached the end of this video. What did it show you?

• Overview of Pega best practice for data science transformation.
• Definition and importance of data science transformation in Pega Customer Decision Hub.
• Key terms and concepts related to AI, machine learning, predictive models, and adaptive models.
• Role and responsibilities of a Pega data scientist in BAU.
• Operational aspects and setup of adaptive modeling capability.
• Monitoring, reporting, and diagnostics in Pega Customer Decision Hub.
• Academy training requirements for data scientists.