Job candidates who decide not to take up a position, after receiving a job offer, can cause headaches for human resource departments. For large scale companies that hire thousands of people every year, this can be a major issue. It can affect everything from talent management plans to revenue.
In this article, we examine how artificial intelligence and machine learning techniques can manage this problem, while improving the employee onboarding experience. We also analyze the challenges that organizations need to address if they want to implement such a solution.
The problems with current recruitment assessments
Many organisations implement a candidate engagement program to engage people after providing an offer to help with their onboarding. Such a program tracks candidates’ engagement with details such as their contact details, communication dates, attendance at company events, previous employment details, travel time from home, and responsiveness. This information helps recruiters provide a more personalized experience for candidates – but hiring managers and recruiters also perform a continuous assessment of the risk that the person won’t finally take up the offer, and use this assessment to manage their hiring pipeline.
However, these cancel hire risk assessments can be unreliable making it difficult for hiring managers to plan for contingency. Factors such as subjective assessments and poor use of data reduce their reliability. In particular we see:
- Subjective judgements are often inaccurate. In many organisations, the assessment of cancel hire risk is performed through “judgement calls” by individual recruiters and hiring managers. The quality of these assessments naturally vary. For instance, a department head can probably make a better judgement call than a new manager because of their prior understanding of the market.
- Assessments ignore past data. Many of these assessments do not leverage knowledge and insights from past data, nor the collective intelligence of recruiters and hiring managers. The usage of assessment data becomes limited to simplistic and past-data oriented dashboards.
These inefficiencies in risk estimation can impact talent management plans.
Artificial intelligence and machine learning have the potential to improve the reliability of these assessments. We have worked with organizations to build machine learning models that use past data and leading indicators to generate a predictive score for each offered candidate.
Such models help drive objective assessments that use an organization’s collective intelligence, rather than relying on a few senior managers. Hiring managers can then use this score to increase engagement activities with candidates to mitigate risk. These scores can also generate an enterprise view of risks that will help central capacity teams.
AI solutions have significant potential, but challenges remain
AI-based solutions can transform the way organisations manage their talent pipelines. However, building such a solution has its own set of challenges that HR teams need to address to realize their potential. In our experience building solutions for our clients, we have seen the following:
- Prior experiences can impact acceptance. First and foremost, gaining acceptance of the idea is not always as easy as you expect. This can be because of the hype about AI and ML. In some cases organisations have tried a “AI/ML solution” in the past and experienced poor results. Such approaches might have been unrealistic and driven more by hype around the technology. It is also possible that earlier solutions failed due to the wrong skill mix. Setting the right expectations is key for success.
- Technology teams often continue to work in isolation from the business. Initiatives that are too technology focused or theoretical tend to fail in delivering a business solution. Teams building the solution often fail to enlist recruiters or HR personnel whose contribution and observations are equally, if not more, important than the algorithms being discussed in the sprint calls.
- Talent acquisition processes have to be revised. For example , many of the leading indicators about hires such as candidate responsiveness, job portal updates and communication dates, that are critical indicators of the outcome, may not be captured in the current process. This implies that processes and spreadsheets have to be revised to capture the data elements that are required for the models.
- The typical challenges of building data based solutions have to be addressed. Examples are missing data, multiple date formats, variable reduction, de-duplication of data, encoding of the categorical data, model selection and cross-validation. Proven techniques usually exist for these tasks, but considerable time often goes into addressing these issues and requires inputs from business users.
We recommend following standard practices like basic descriptive statistics, evaluation of ML models for predictive and cognitive analytics, hypothesis testing, and application of multiple ML techniques. It is then necessary to continuously improve the accuracy of your models by gathering better quality data and learning from historical data.
The more data, the better, for algorithms. This means you may have to spend time collecting and munging the data once the processes and templates are revised.
Solutions are best built using an iterative process rather than treating it as a one-time exercise. Earlier iterations tend to set up a base and accelerate outcomes in future iterations.
AI has significant potential to help organizations manage talent
To summarize, there is immense opportunity for organisations to become better at estimating and managing their new hires by applying machine learning. Recruiters and hiring managers can use these technologies to benefit from the collective intelligence of their team and organization. Doing so means they become better at managing the risk of cancel hires and the resulting business impact.
We published this article with the help of our colleagues, Hrishekesh Shinde , Surender Singh and Aishwaraya Upadhyay.