Mercer streamlines operations with machine learning

Every year, human resources consulting firm Mercer collects more than 30,000 compensation and benefit surveys to help its clients better serve their employees. For years, the process of preparing and cleansing the data required weeks of administrative oversight and back-and-forth communication with clients.

In 2017, Mercer started an initiative to bring those back-office operations under control and provide a better experience to its customers. The Mercer Data Connector, completed last year, reduces the time for survey processing from months to minutes, say Darren Duquette, principal of global business solutions at Mercer, and Rick Koo, Mercer’s digital technology solutions leader.

[ Learn the secrets of highly successful data analytics teams. | Beware the 12 myths of data analytics and the sure-fire ways organizations fail at data analytics. | Get the latest on data analytics by signing up for CIO newsletters. ]

“As we continue on this digital transformation journey, it is important that we continue to view ourselves through the eyes of our clients and deliver user-centric solutions that make it simple and rewarding to do business with Mercer,” says Koo.

The project has earned Mercer a 2019  Digital Edge 50 Award  for  digital innovation.

Overhauling the data pipeline

Duquette and Koo say leaders in the company’s Career Products line identified the opportunity to optimize an already successful business and expand into new markets. In response, representatives from the business, technology and operations came together to brainstorm about the possibilities.

The team found numerous sources of friction in Mercer’s existing data collection process: It was high touch, requiring a lot of back-and-forth between Mercer and its clients to clean the data. There was a high error rate because clients needed to manually enter HRIS data in Excel spreadsheets supplied by Mercer. There was a lot of operational overhead involved in preparing templates for different products and to cleanse the data after clients submitted their files. Normalizing job titles across industry, region, country, client, and jobs to standard benchmarks based on Mercer’s library of 16,000 job titles was a complex task. Moreover, the process simply didn’t meet current business needs, let alone future business needs.

What was needed was a total transformation of the company’s process for ingesting, cleansing, and mapping client data. The project, which also aimed to increase Mercer’s speed-to-market with new products, would need to significantly reduce operational overhead and deliver a superb digital customer experience. The team decided that, to make it easier for clients to do business with Mercer, the resulting platform would need to enable client self-service, provide an intuitive workflow, integrate analytics, and be extensible to meet future business needs.

The outcome of that effort was the Mercer Data Connector, a data collection platform that leverages artificial intelligence (AI) and machine learning (ML) to optimize back-office processes and improve efficiency. It’s built on Apigee’s API gateway to provide microservices between application tiers and domains; LTI Mosaic, a big data platform that provides JIT data ingestion; and numerous open source projects, including Node.JS, Angular, Docker, MongoDB, Jenkins, and Cucumber.

The power of change

Because the solution touches all areas of Mercer’s global compensation survey business, Duquette and Koo say teamwork was essential to success. They underscore the importance of dedicated stakeholder involvement and senior management sponsorship from each core area: product, operations, technology, client service, and markets.

The team utilized an agile approach that started with brainstorming and planning sessions to create a backlog of desired features. Product owners prioritized the backlog based on priority and scope. Duquette and Koo say the process allowed the team to engage each stakeholder community for requirements throughout the process, including frequent product demonstrations and user acceptance testing sessions. They note that one of the most important lessons learned was to be flexible and not force business stakeholders to provide perfect requirements from the beginning.

“Change management is as important as the product and technology,” Duquette says. “Frequent engagement with clients, users, internal stakeholders and sponsors is just as critical in creating a successful new product as each line of code.”

To develop the platform, Mercer used a hybrid Scrum/Kanban methodology using DevOps. Duquette and Koo say they maintained alignment and focus over the eight-month build cycle by continuously revalidating priorities based on strong governance and stakeholder engagement processes.

It required big changes. Duquette and Koo say the team had to change its operating model, tech stack, and people development skills to achieve success, and those changes are ongoing. They provided onsite and self-training for both IT and business employees and brought in partners to help. The changes include the creation of cross-functional teams that draw from multiple functions — IT, strategy, R&D, client support interaction, and operations — to design changes around a set of coordinated specifications.

More on data analytics:

• 12 myths of data analytics debunked
• The secrets of highly successful data analytics teams
• Developing data science skills in-house: Real-world lessons
• 12 data science mistakes to avoid
• 10 hot data analytics trends — and 5 going cold
• 7 sure-fire ways to fail at data analytics
• Data analytics success stories: An inside look
• Demystifying data analytics: How to create business value with data
• Big data and data analytics certifications that will pay off