“How to present predictive information and user interface is also changing. The essence of traditional interfaces is a deep emphasis on clarity (or intuitiveness). One look at the interface tells you what to do. That doesn’t work with likelihoods. Intelligent output requires the user to think before deciding.” - John Sumser
The Limits Of 20th Century Design Ideas in
AI & Predictive Hiring Software in HR
Make sure to read Part 1 of this series Here »
Like most credible companies delivering HR Technology, Manhattan based Greenhouse is working hard to discover the places where added intelligence and predictions can be most useful to their customers. One area they’ve focused on is predicting the Recruiting Departments results. Hiring and finance managers always want to know when the new person will be hired and whether they will show up.
In order to harness the results of predictive tools, Greenhouse instituted a process that involves several repeatable steps:
- Cleary define the predictive value you want to deliver. In this case, the likely date that a new candidate will start work and the aggregate likelihood that Recruiting will meet its KPIs
- Collect and Process the data to make the prediction
- Develop a combination of graphics and text to communicate the data
- Test the results with customers
- Assess customer satisfaction and iterate both the objective and the interface design
Underlying the approach is the very clear understanding that learning is an ongoing process. Today’s best may well become tomorrow’s not good enough. As users become more sophisticated, interfaces will continuously evolve.
John Sumser is a Principal Analyst for HRExaminer.
Greenhouse began with the following hypothesis: We believe Recruiters want informed predictions about when a hire will start in order to gain credibility and set clear expectations with Hiring Managers and Finance teams alike.
In the initial cycle, Greenhouse used bar chart histograms to express the likelihood that a job would be filled. These first interfaces focused on helping to understand how the forecasts were developed so Recruiters could explain their forecasts. Beautiful bar charts indicated the array of possibilities suggested by history.
While accurately portraying the statistical reality, Recruiters were unable to understand what action was required. They did not want all the information; they wanted usable direction. It turns out that complex data often aggravates the anxiety users feel about misinterpreting data.
The Greenhouse team also discovered that project completion meant different things to Recruiters, Hiring Managers, and the Finance Department. Recruiters focus on the hiring decision. Hiring Managers are concerned about the day the new hire reports for work. Finance is interested in cash flow timing. Each group had different, but important, concerns about predictions on when and whether a new employee will start.
In the end, the Greenhouse team opted for an interface that gives action oriented information. Is the project on track? The range of dates in which the offer will be accepted? The final interface replaced detailed probability data with clear and actionable insight.
Each individual prediction in an HR Technology product will be undergoing this sort of testing and retesting.
From the beginnings of the industrial revolution to the end of the first full generation of Enterprise software, interface design focused on one important variable: clarity. Sometimes called intuitive, the basic idea is that a good interface tells you exactly what to do; that the user should have no question about the next step.
Clarity and oversimplification can be horribly ineffective when dealing with the output of intelligent systems. As Marshall McLuhan wrote in his 1967 book, The Medium is the Message, “We look at the present through a rear-view mirror.” This happens time and again with technology but it falls on us to avoid this pitfall. To practice ethical hiring then, or any HR function where critical decisions are aided by intelligent tools, we cannot accept decisions made in a mysterious ‘black box’ of algorithms. We must design our tools to help us practice our AI ethics. That includes a continuous process where you revisit your assumptions, rewrite your questions, validate your data, and measure results aginst not just KPI’s, but your underlying corporate values. There is a path to realize the full potential of AI and intelligent software but it isn’t one made with technology ideas engineered for a different time and reality.
Read The Entire Series
