
4.1. Workforce Scheduling and Activity Forecasting.
HR scheduling and customer traffic forecasting is increasingly based on integrated data. Kronos for example, can predict customer traffic and dynamically change the labor hours of employees and then reduce costs. Another platform RetailNext performs future predictions of retail in-store behavior, based on historical data and statistical modelling. Tara is another tool capable of automatically creating development tasks and timelines for software projects, based on millions of previous software projects data. Additionally, it automatically matches these tasks with pre-screened external contractors or with internal developers.
Tara automatically matches these tasks with pre-screened external contractors or with internal developers.
Automatic call distributors based on skills, are systems that match callers and agents based on predefined skills of available staff. The caller provides some information that is used to direct the call to the first available target agent that can handle the call.
4.2. Attrition, Engagement and Turnover Prediction
Companies are continuously using profiling and predictive tools to measure employee engagement or reactions to changes in a company. For example, Peakon is an engagement platform used by many UK companies that are moving from annual surveys to continuous data gathering and evaluation approach. Peakon includes performance benchmarks to produce an engagement score that is connected to targets of the business. It also makes use of ML to identify how demographics are related to employee engagement scores. My Happy Force is another mobile app that gathers data on how employees feel and analyse their opinions, worries and motivations.
Predictive analytics are also extensively used in turnover predictions.
Predictive analytics are also extensively used in turnover predictions. For example, IBM released a dataset with fictional data to motivate research on attrition predictions and models to calculate the risk of an employee quitting. Other data, provided by Happy Force records, attempts to study the feasibility of predicting turnover, no matter if the worker quits or was fired, based on data of mood, comments and interactions of workers using the app.
4.3. Profiling, Performance Evaluation, Optimization and Forecasting.
Performance evaluation is perhaps the primary objective that motivates workplace monitoring. The measurement and evaluation of performance is categorized in three ways.
- Reports and metrics are direct observation of data, e.g. cost per hire.
- Key Performance Indicators are measures of performance against business objectives.
- Analytics are intended to identify which factors impact on performance. Examples are People Analytics and HR Analytics.
Analytics always have a wider view that typically involves the whole company.
Organizational Network analysis
Organizational network analysis refers to the discovery and optimization of the social network of a company. This network describes the actual relations that drive the company in parallel to the formal organization and roles. Its target is to measure the social value at individual, team and organization levels by building the social graph of an organization. In that graph, people are represented by nodes who serve as essential connections to exchange ideas and information.
Deloitte points out three types of nodes
- Central node are people that seem to know every co-worker, share lots of information and influence groups.
- Knowledge brokers refer to people who create links between groups helping information to spread.
- Peripheral are workers overlooked and unconnected to the rest of the company.
Central nodes are people worthy of identification and proper management. Peripheral workers on the other hand are labelled as risk of exit and if talented can be hard to replace, since the people with this profile are less likely to share their knowledge.
ONA can be used for multiple purposes. For example, TrustSphere, a global company providing People Analytics tools, can analyse the network evolution to characterize a typical set of communication patterns so that deviations from normal behavior signal fraud events such as information leakage. The Municipality of Odense in Denmark applied the ONA by Innovisor to reconfigure the organizational design or infrastructures so that they are closer to the real web of employees.
The social graph is built by using several information sources such as metadata of e-mails, shared calendars, file sharing systems and productivity tools. The figure below illustrates social graphs produced by Worklytics, a product used by Telefonica, WeWork and others.
Conclusion
In this insightful report several key trends, regarding the use of AI and ML tools have emerged, which are the following:
- Surveillance and control system adoption increases with the size of the company.
- The more common a form of surveillance is, the more accepted by the workforce it will be.
- Many tools to automate the hiring funnel are used in companies with high-volume hiring and high turnover, particularly in low wage employment. The tool that is mostly used in such companies is Tribepad.
- In assessing the automation of the hiring process, the funnel metaphor helps to understand hiring as a set of stages that will filter candidates.
- With automation, an emphasis is seen on not only what candidates are able to do, but who they are or are likely to be.
- Recommendations systems have become prominent.
However, the discussion of the impact of these technologies is not straightforward and several gray areas exist. As research in the rapid datafication of the workplace continues, researchers at the Data Justice Project hope to shed further light on this debate. Those grey areas are the following:
- Research has shown that the use of multiple source data incorporates biases that can worsen structural inequalities.
- Most of the emotional detection tools assume that all humans feel the same six basic emotions, something which is a largely outdated scientific premise.
- Misuse of ML models can occur if the model and training data is not properly documented or understood by the development team.