By Peter Weddle on the TAtech Blog
The recent TAtech Leadership Summit on AI & Machine Learning in Talent Acquisition featured a session on how best to ensure a new talent technology product advances an employer’s commitment to diversity, equity and inclusion (DE&I). It was presented by Gerri Allamby, the Associate Compliance & Diversity Manager at Unilever. She argues that the assessment of a product’s impact on compliance should begin at the outset of its acquisition rather than during its implementation, which is the more typical approach.
Evaluating new talent technology products – especially those using AI and machine learning – should cover a range of factors, including:
• Product design,
• Algorithm description and transparency,
• Training data, and
• Model validation and performance.
In the area of algorithm description and transparency, for example, an employer should ask the solution provider:
• How does the algorithm work? What are the specific methods applied? What factors are measured and how is each factor weighted?
• If the product measures a subject’s skill or fit for a position, what types of characteristics or traits does it evaluate?
Or, in the important area of training data:
• How has the algorithm been trained? What was the source of the training data and how was it collected?
• Are there any inherent biases in the training set? What steps have been taken to adjust the training data to improve model performance or reduce bias?
The assessment should also cover the security of data used during the product’s operation. As data are input into the product or generated by it for the recruiting team:
• What are the data retention protocols for the product? How long are data stored and can it be destroyed upon request?
• If the organization is audited, can the data it used and/or developed be accessed and, if so, can it be provided on a topic-specific rather than aggregate basis?
Similarly, the pre-acquisition evaluation should also address the increasingly important area of data privacy. When the product is implemented and used, for example:
• How is consent obtained to collect and work with personal data and/or biometric identifiers?
• Are the data generated by the product retained by the vendor or a third party and, if so, under what rules?
In addition, the examination of the product’s use should also focus on its accessibility and accommodation for those with disabilities. Does it, for example, provide a disclaimer or offer a way by which people with visual, mental, hearing or other disabilities can request a reasonable accommodation?
Finally, given the rise in labor mobility and remote working, it is important to determine if the product can be used in countries outside the US. If that’s the case, the assessment should carefully review how it navigates the various regulatory regimes of those other countries. And, of course, in today’s highly litigious environment, the assessment should explore what kind of assistance the vendor will provide in the event of a lawsuit. Will it, for example, provide support should a discrimination claim be filed and/or will it indemnify the organization against such claims?
Yes, a pre-acquisition compliance assessment adds another level of complexity to the introduction of new talent technology products. However, this “compliance first” approach also avoids the inefficiency and potential substandard outcomes of having to bolt DE&I considerations onto a product later during implementation. Simply put, it’s best to get it right from the outset rather than having to fix it after the fact.
Food for Thought,
Peter Weddle is the author or editor of over two dozen books and a former columnist for The Wall Street Journal. He is also the founder and CEO of TAtech: The Association for Talent Acquisition Solutions. You can check out his latest books on Amazon or in the TAtech Bookstore.