Workforce Optimization Undergoes Radical Change – Ready?

In today’s intensely competitive markets, companies must strive to meet customer expectations during every interaction, and interactions occur through many channels. Our benchmark research into next-generation customer engagement vr_NGCE_Research_12_all_current_channels_for_customer_engagementfinds that customers use up to 17 channels of engagement. Some channels involve assisted service from employees of the company, and some use self-service technologies such as interactive voice response (IVR), websites, mobile apps and social media, also known as digital service. Although the use of self-service is increasing, the research finds that organizations still expect volumes of assisted interactions to grow, albeit more slowly. The research also shows that the employees customers interact with may work in almost any line of business, including marketing, sales, the contact center, finance and human resources. These challenges require organizations to focus on people, processes, information and technology to optimize the performance of the workforce.

To meet customer expectations during assisted interactions, companies must have the right number of skilled employees available, including cases in which the customer begins in a digital channel but switches to assisted service, and handle a number of functions:

  • Route interactions to the employee most likely to satisfy a particular customer and situation.
  • Build work schedules that match agents’ numbers and skills to expected volumes and types of interactions. To do that requires systems that are responsive and flexible enough to handle unexpected events such as sudden surges of interactions or employee illness and absences.
  • Assess the performance of agents and other employees handing interactions, and determine future training and coaching for individual needs.
  • Motivate employees to improve their performance.
  • Give employees access to all information and systems that can help resolve issues, as often as possible at the first attempt.
  • Enable employees to find and collaborate with others who can help resolve issues without having to call the customer back.
  • Assess the overall performance of interaction handling to ensure it remains within operational guidelines while meeting business objectives, and plan process improvements to optimize the customer journey and the agent experience.

To accomplish all these tasks, companies need to use a combination of systems, which vr_NGWO2_06_use_of_agent_workforce_applicationscollectively are known as workforce optimization. Our research into next-generation workforce optimization finds that the three types of systems most commonly in use are call recording (78%), quality management (70%) and workforce management (45%), although coaching (17%) and e-learning (15%) are the most likely to be deployed over the next two years. In each of these categories both conventional systems and newer, more capable ones are available.

Regarding the most popular kind we note that companies have been recording calls for many years but that few use them to full advantage. Typically, companies listen to a small percentage of calls and use this to manually complete agent performance scorecards. More innovative organizations use speech analytics systems that enable them to utilize all calls, automate much of the agent performance assessment process and most importantly link this information with customer feedback. Such companies record all types of interactions and use multiple forms of analytics systems to create an even broader picture of agent performance and customer experience.

Quality management systems are also mature but typically support a manual process of creating and filling out scorecards for different types of interactions. Here again more advanced systems use analytics to automate much of the process, including calculation of agent quality scores.

Workforce management systems typically use historical data about interaction patterns to produce work schedules that optimize available resources and meet operational targets. More advanced systems monitor employee performance against those schedules and include capabilities to optimize short-term agent utilization, for example by filling idle time with training and coaching.

Coaching and e-learning is a less mature category. Conventional products use the output from quality monitoring and analytics to identify coaching and training tasks, which can then be scheduled using workforce management. The most advanced systems can extract data from competed interactions to illustrate areas in need of improvement, or identify the performance of other employees who handle interactions using best practices.

As well as these advances in core workforce optimization applications, there are a number of potential game changers. The first is analytics. As I have already highlighted, analytics is increasingly important  to workforce optimization. Participants in our next-generation customer engagement research said it will have the greatest impact of any application on customer and employee satisfaction. Advanced speech and text analytics, used in combination with structured data analytics, can produce a comprehensive view of interaction handling, employee performance and customer satisfaction. It can spot trends and issues, and predict likely outcomes of future interactions. This information can be used to forecast future resource requirements, suggest process changes, identify training and coaching needs, and automate calculation of metrics focused on both customers and employees, such as first-contact-resolution rates across channels, customer effort scores, customer lifetime value and agent performance. These metrics can become an integral part of gamification techniques that track and reward agents for performance in day-to-day operations and for taking part in training, coaching and game-playing sessions to help hone skills.

Technology integration is having an impact of workforce optimization. Nearly half (48%) of participants in our next-generation workforce optimization research said that it is important for these applications to be integrated. If they are, what have previously been stand-alone processes can flow across application boundaries: For example, customer feedback can link to quality monitoring, analytics can support information-driven processes in multiple applications, and e-learning sessions can be automatically inserted into agent schedules. Integration of applications also supports a closed-loop approach to workforce optimization that uses analytics to assess past performance, identify areas for improvement and monitor the impact of changes.

Other innovations also are having impacts. Many vendors now support access to their systems from mobile devices so that employees can work away from their desk – for instance, supervisors walking the contact center floor – or home. Cloud computing opens up the opportunity for small and midsize companies to access capabilities similar to those designed for large enterprises.

As I said, our research shows that assisted service has and will continue to play a key role in interacting with customers, and these people and processes must be managed. Increasing demands from customers, multiple channels of engagement, greater volumes of interactions and more complex interactions all increase the urgency of deploying the right number of skilled employees to deal with customers. I therefore recommend to organizations that rely on outdated systems, particularly spreadsheets, to manage those tasks evaluate how more advanced, analytics-driven systems can improve the performance of employees and consequently the customer experience.


Richard J. Snow

VP & Research Director, Customer

Follow Me on Twitter  and Connect with me on LinkedIn

RapidMiner Brings Self-Service to Predictive Analytics

Predictive analytics is a rewarding yet challenging subject. In our benchmark research on next-generation predictive analytics at leastvr_NG_Predictive_Analytics_16_why_users_dont_produce_predictive_analyses half the participants reported that predictive analytics allows them to achieve competitive advantage (57%) and create new revenue opportunities (50%). Yet even more participants said that users of predictive analytics don’t have enough skills training to produce their own analyses (79%) and don’t understand the mathematics involved (66%). (In the term “predictive analytics” I include all types of data science, not just one particular type of analysis.)

Various software vendors are taking steps to simplify the use of this technology. RapidMiner is one of them. The company focuses on making its open source predictive analytics faster and easier to use. Its database-independent predictive analytics platform has more than 1,400 customers and averages 20,000 downloads per month. The product, also called RapidMiner, has been deployed more than 100,000 times and has a community of some 250,000 users. The latest version of the platform, Version 7.1 was released in the spring. RapidMiner has been around for almost 10 years, and in that time, the predictive analytics market has grown and changed dramatically in parallel with the big data market. Big data was not part of the original focus of the company, nor was cloud computing, but over time RapidMiner has incorporated capabilities in both areas.

The company also has a distinctive personality embodied by its founder and president, Ingo Mierswa. It is evident in his YouTube video series, “5 Minutes with Ingo”, in which he explains various aspects of predictive analytics. This approach to training potential users makes sense. According to our research, adequate training in predictive analytics concepts and the application of predictive analytics to business problems correlate more highly with satisfaction in using it (93% each) than does product training (85%). These satisfaction rates compare favorably with just 66 percent on average. The RapidMiner training videos are not only entertaining, they can potentially help an organization be more successful in understanding and using predictive analytics.

The RapidMiner product set itself provides several approaches to predictive analytics. RapidMiner Studio is a desktop tool for creating predictive analytic models. It is available for download from the RapidMiner website. Like many other predictive analytics tools, it includes connectors to a variety of data sources and supports data preparation tasks that are often needed before predictive models can be developed. Using drag-and-drop visual design, users create data flows or pipelines of activity moving data from sources, through any necessary transformations and into modeling processes.

RapidMiner Studio has several unique features to guide the user through these processes. In designing the overall pipeline of activity, a feature called Wisdom of Crowds examines what other users have done in similar situations and recommends what the next step (or “operator”) ought to be. Behind the scenes, RapidMiner is using its own technology to help predict the most likely next step. Wisdom of Crowds also provides parameter recommendations to help choose among the myriad of options and parameter settings. As further techniques to assist users, RapidMiner Studio has components to compare multiple models and to select models automatically.

While users can perform the entire predictive analytics process using RapidMiner Studio alone, they also can connect it to RapidMiner Server to support larger data sets and collaboration among multiple users. The Server product has a shared repository for processes, data and connections to other data sources and includes a framework to provide security and version control for the various items in the repository. As an alternative to an on-premises server, RapidMiner Cloud provides the same capabilities as the server product in a hosted environment.

For big data analytics RapidMiner Radoop leverages Hadoop implementations by pushing down the predictive analytics pipelines created in RapidMiner studio. These pipelines execute in the appropriate Hadoop component including MapReduce, Spark, Pig, Hive and Mahout, allowing access to the full data set and taking advantage of the cluster resources for parallel execution of the workloads without the need to code in any of these tools. Spark has become a popular framework for analytics on Hadoop, as evidenced by the Spark Summits, which I wrote about recently. It provides faster execution of analytic processes and a more flexible, expressive framework than MapReduce. Users familiar with Spark (R or MLlib) PySpark, Pig or Hive can write scripts in these packages that can be executed with Radoop. For security and authentication Radoop integrates with Kerberos, Apache Sentry and Apache Ranger.

RapidMiner recognizes the value of visualization in the analytics process and has established technical partnerships and integration with two providers, Qlik and Tableau. RapidMiner Studio can create both Qlik and Tableau data exchange files for visualization of the output of predictive analytics models. Other connections, integrations and extensions are available through the RapidMiner marketplace including Cassandra, MongoDB, SolR and Splunk.

To gain maximum value from predictive analytics, organizations must not only create the models to predict behaviors, they must deploy those models in an operational context to impact business outcomes in real time. According to our research more than one-third (37%) of organizations are applying their models at least on a daily basis. RapidMiner can convert any of its pipeline processes into a Web service so they can be embedded in other business processes and invoked in real time. RapidMiner also supports PMML, which is an industry standard for expressing models and allows embedding of models into databases for real-time scoring of new data records as they are entered into the database.

While RapidMiner has invested in making predictive analytics easier to use and accessible to a wider group of analysts, it is a daunting challenge to make these types of analyses truly self-service. Knowing when to use a particular algorithm and how to set all the various parameters requires deep knowledge of the discipline of predictive analytics. For example, in creating a k-nearest neighbors model, how many people would know what value of “k” to use for the number of nearest neighbors to model? And this is just one relatively simple parameter on one type of algorithm. The Wisdom of Crowds parameter recommendations help, but it’s still not an automated process, and users should realize they will need at least some knowledge of the various algorithms to maximize the effectiveness of their modeling efforts.

I’d also like to see RapidMiner invest more in the model management process. Once a model is created, it immediately starts to become stale for various reasons. Market conditions change. New data is generated. The competitive environment changes. The key questions are how far out of date the model has become and when it should be replaced with a better model.  Models should constantly be re-evaluated. In our predictive analytics research 63 percent of organizations that update their models at least daily reported a significant improvement in their activities and processes, compared with 31 percent of those that update their models less frequently. Any vendor that automates this process could help organizations boost their effectiveness.

Overall RapidMiner has made predictive analytics more accessible to a wider audience via its products and its educational efforts. The company has done this in an entertaining way, which is important to retain the attention of those who are being educated. Predictive analytics is a critical aspect of maximizing the value of data in an organization. Those that are not taking advantage of these types of analytics should be. RapidMiner makes it easier to tackle some of these challenges and may help get any organization over the hump of learning how to build and deploy predictive analytic models.


David Menninger

SVP & Research Director

Follow Me on Twitter @dmenningerVR and Connect with me on LinkedIn.