Building high performance Customer Service applications

Familiarize yourself with performance and debugging concepts by reviewing the following resources:

• Analyzing application performance - https://academy.pega.com/challenge/assessing-application-performance/v4/in/38221
• Debugging and performance - https://academy.pega.com/module/debugging-and-performance/v1
• Predictive Diagnostic Cloud - https://docs.pega.com/bundle/predictive-diagnostic-cloud/page/predictive-diagnostic-cloud/reference-guide/how-to-start-using-pega-predictive-diagnostic-cloud.html

Depending on the version of Pega Customer Service that you are implementing, you may have anywhere from 2 to 4 work pools out of the box (OOTB). When setting up your implementation, ensure that you have at least one work pool for the interaction classes, and one work pool for the service intent cases, each pointing to a different table.

By default, the interaction classes work pools and service intent case work pools point to a single table. Although a single table will work for small implementations, it will cause a bottleneck for larger implementations.

For extremely high-volume implementations, consider creating work pools that mirror your core entities, or interaction driver categories that spread the load across more tables. The more tables that are present, the lower the IO contention is at high volume. For example:

• BigBank-Servicing-Work-Acct (work pool for account-related cases)
• BigBank-Servicing-Work-Gen (work pool for general service-related cases)
• BigBank-Servicing-Work-Interaction (work pool for customer interactions)

For report definitions and dashboards that don't have to be instantly up to date, consider using a reporting data source instead of the primary data source. If your applications are hosted in a Pega Cloud Services environment, the use of read replicas is also a factor.

How the user experience, and in turn the user interface, is designed and built can have a large impact on performance, both actual and perceived. Determine what information is important to display to users. Running and loading an interaction and the 360° customer composite requires anywhere from 6 to 30 interfaces.

• Don't display something until you actually need it. For example, when searching for a customer, only display enough information to select the customer, not to verify them. Wait until verification to display the additional verification information.
• Ensure that sections are collapsed and defer loading if possible.
• When showing a list of items, enable progressive loading and set a reasonable page size, such as 15 items per page.
• When building show/hide logic, try to use expressions instead of When rules because expressions execute on the client side instead of on the server side. As a result, expressions will perform faster and reduce server-side load as you scale.

A clean clipboard on agent requestors, reasonable overall requestor clipboard size, and efficient and correct load strategies on data pages can mean the difference between an unstable, slow Pega Customer Service application and one with fast, consistent performance. A well-designed application should provide continuously responsive performance for agents, regardless of their shift duration, and it should scale when there are many users per node.

• When possible, make data pages requestor level to avoid constant reload. System of record data, data used across cases, and data shared across interactions and service intents are all good candidates to be made requestor level.
• Do not use the Reload once per interaction option. This option reloads the data page on every HTTP interaction, instead of once during the customer interaction. For example, supposed a screen has four sections that are defer loaded, and all are rendered on the same data page. Each defer load is a different HTTP interaction, and thus the data page will be reloaded four times. If you need a fine-grain reload strategy, consider using the if older than reload, or the do not reload when, but forcing the condition to evaluate to true.
• Always select Clear pages after non-use when this option is available (on read-only requestor level data pages).

• Upon wrapping up an interaction, Pega Customer Service will invoke the CPMPreClose activity, which contains two different methods of removing requestor-level data pages: CPMRemoveRequestorDataPages, which loops through D_Interaction_RequestorDataPages and removes them explicitly by signature; and the CPMCleanInteractionPages data transform. Unless you have touched a page from the portal thread which does not get removed when an interaction is closed, or you did not select the Clear pages after non-use check box on your requestor-level data pages, you will need to add your implementation build data pages to CPMCleanInteractionPagesExtension. Also note that if you have changed the parameter signature of any of the OOTB data pages, you may also have to add this update to the extension.
• Periodically, check during the development and QA cycle that all data pages are being cleared upon the wrapping up of interactions and service cases.

Monitor requestors periodically for average size to help identify potential issues with loading too many things on the clipboard, pages not being cleared out, and users who do not close and wrap up interactions once finished.

• The following screenshot is from Admin Studio > Users > Requestors > Load requestor size.
• Key items to watch are the average page size of your requestors and the largest page sizes. It is common to have some requestors in a Pega Customer Service implementation go as high as 20-25mb in size, based on having multiple interactions and service intents open, especially when agents are handling multiple active chat sessions. On average, the requestor size should typically be 12mb or less across all system users.

• If you notice that your requestors are often larger than this range, then use the Clipboard tool to perform clipboard analysis. The following figure shows a requestor that is larger than the target threshold. By digging deeper we can see that the D_OpenServiceCases data page is not only large, but also does not appear to be getting cleared out from interaction to interaction.

Most Pega Customer Service implementations do not store data locally but integrate to external systems of record. When integrating with external systems of record, follow these guidelines to ensure high performance.

Upon receiving connectors, try to independently run load against those connectors using industry available tooling, such as JMeter, to understand their performance profile under load. Interfaces should ideally respond in less than 500 milliseconds. Anything that is on average over 1 second should be remediated for performance, independently of integration into your application.

Use aggregate data sources in data pages to reduce the number of moving pieces required for integration, both for saves and reads. Implementations often use complex and hard to follow activities for performing integration. With the advent of aggregate source data pages, this is no longer required. Directly reference connectors and their associated mapping, and if required, add a data transform after your connector sources, to perform any reconciliation or error handling after calling multiple interfaces.