Machine Learning for WorkPlace Buzz

WorkplaceBuzz are an occupational psychology company that creates validated wellbeing and engagement tools. Founded by Dr Mark Slaski PhD the company is committed to producing the highest quality psychology products.

The Brief: Create a model that preserves the validation of their annual engagement survey but allows it to be done in a pulse method when participants are surveyed continuously throughout the year via an online app. Using machine learning models we predicted user behaviour from a diverse set of potential inputs.

Requirements: 

• Design a method of distributing the survey to get the most representative data for each pulse

• Preserve the reliability of the data despite not having every participant answer every question per pulse

• Create a normalised scoring method to add more depth to the Reporting Metrics

• Analyse all existing data for trends, correlations and descriptive statistics

• Design and Build Machine Learning Models to Predict missed answers accurately for each combination of variables possible

• Fully Document the entire methodology for developers to use in building the survey app

• Design and build a reporting platform for their engagement tool

Our Solution

Services

Data Collection, Data Science, Data Pipeline, Analytics Design, Analytics Development, Machine Learning, Analytics Consulting

The first part of the project was to talk with the software development team to understand the technology stack and understand the data already held in the app's databases. We did this by meeting the development team in Ahmedabad, India where we held a number of sessions reviewing the current build and discussing what capabilities were available in the technology stack.

Once we returned to the UK we designed a distribution method so that the annual survey was broken down in such a way as to gain a representative dataset each month.

The next stage was to analyse the existing data which required a number of different data sources to be combined into a single dataset containing the answers to around 700,000 questions.

Using IBM SPSS (Statistic Software for Social Sciences) we created a number of descriptive statistics to understand how the different parts of the survey compared to each other, this included Response Distributions, Mean Averages, Standard Deviations, Skewness, Kurtosis, Cronbachs Alpha and other statistical techniques.

Then using this analysis we created a normalised scoring method, with this method clients are able to see how their score relate to the benchmarks at a glance.

The final stage of the process was producing documentation for the development team to use to build out the features.

Client Feedback

"Discovr Analytics' has helped us to modernise our Engagement Survey's infrastructure allowing us to gain much more valuable insights for us and our clients. They worked quickly and effectively to get us set up on their system and provide us with ongoing support for all our survey projects."

M Slaski PhD/ Founder