AI in Talent Acquisition – Revolution or Evolution?

Across the different verticals and industries that figure as part of our daily ‘concern’, the ever advancing drumbeat of Artificial Intelligence (AI) has been, and continues to be relentless. From a month old European AI start-up raising a €105 million seed round to working with scaling businesses assessing what AI tools could best support their growth and agencies contending with how they can best leverage AI on behalf of their clients and if or when is it appropriate to do so. 

AI can’t be ‘uninvented’. Pandora’s box is open. It is here to stay and will only develop in its prominence, uses and reliability. As such it has opened up a really interesting new set of conversations over the past few months, both internally at Sapio and with our clients, about how best to employ AI (pun intended) in Talent Acquisition and Recruitment. But where to start? 

There are a myriad of AI tools Talent Acquisition functions can look at implementing, from; internal network mining to resume and application screening, chatbots to automate candidate interaction, facilitate diary management and interview scheduling, video and virtual interviewing platforms both of which can provide additional deep insights around tone of voice, body language etc. There are compliance tools that offer depth and speed for things like background, reference or employment checks. There are contract execution and support tools. There are workforce analytics platforms that identify correlations between your top performers in order to better drive hiring decisions. 

The fact is, there are just too many options out there to wade into the mire without some semblance of a plan. Otherwise, you risk implementing AI for the sake of AI, or being at the mercy of whomever’s Google Ad spend. Besides, assessing each option or application based purely on its upside without attention paid to the potential downsides, would be fairly short sighted as risks tend to centre around key areas like candidate experience, the possible lack of employer branding control, and latent threat of amplifying or introducing biases. 

With that in mind, we thought it might be helpful to share some top line takeaways from our own research and conversations with other businesses about how to best approach the implementation of AI. 

Break down your process

Before we even think about AI itself, it’s important to first break down our respective workflows into their component elements so as we can best evaluate where the application of AI might be most helpful, and then how best to pass on those increased efficiencies to our customers. In our case our clients and candidates. For us that looks something like; Sourcing, Outreach, Engagement, Evaluation, Logistics, Offering, Closing, Onboarding, Developing and Feedback. 

Qualify ‘Successful Outcome

Once you have a clear understanding of the specific elements where you might seek to affect change, it is important to consider what a successful outcome for each component is, in order to understand viability and assign priorities. For us, our natural mindset to continuous improvement is driven by the principle of marginal gains; an approach by which individual marginal improvements in the process have a compounded effect on the overall result. With this in mind, we considered which elements are most vulnerable to time inefficiencies, ineffectual resource allocation, and which aspects could have the most impact on customer experience.

Define your priorities

For any business, understanding where to invest time, effort and money is good sense, but for scaling businesses where growth is often not linear but exponential, it is paramount. Defining the priority area(s) in which you wish to roll out AI implementation, whatever metric you chose to base that on, will ensure the investment made and impact expected are highly correlated. For us, filtering each component element through the lens of marginal gains makes the prioritisation process very easy and structured. As per the below scoring matrix, anything that scores 3 or higher is an immediate priority. 

Score Value 
2Ineffectual Resource Allocation
3Indirect Impact on Customer Experience
4Director Impact on Customer Experience 

Identify the requirement and research appropriate tools

Identifying the overarching requirement of each component prior to researching available tools is a must. Whether that be AI to Automate, AI to Personalise, AI to draw out Data & Insights, AI to mitigate bias and subjectivity, AI for compliance etc. Otherwise you risk fitting the problem at hand to the tool you find, rather than the tool to the problem at hand. 

Researching appropriate tools and applications is a fine art, and requires time, patience and a fair amount of reading. Testing tools themselves can be hard due to the volume and specificity of data required so patience, and a continued methodical approach will hopefully ensure a fruitful resolution. During this process, we shouldn’t neglect reaching out to our network to understand how peers are tackling the problem, or what solutions they may have already found. 

Data, Data, Data

AI requires, relies and thrives on data and, the more the better. Volume may be harder in the initial stages of implementation but an area where you can immediately affect positive progress is in its quality and accessibility, and it’s never too late to start. For us, in the early days we debated around volume or specificity of data, but as we’ve learnt we were entirely missing the point. Volume can be made specific, turning specificity into volume is much harder. By approaching implementation of AI methodically (as per above) you will be building a clearer picture of specific objectives and desired outcomes which will allow you to better understand which data points are most relevant. From there, and in a similar flow to the very nature of this article, you can Define, Qualify, Gather, Clean and Standardise, Annotate, Format, and Deploy.

Continuous maintenance and feedback

Tools will invariably become more sophisticated over time, but either way and particularly at this stage, AI tools should be considered a part of the solution, not the silver bullet. Continuously monitor the output and improve the input. There are even AI tools to monitor AI tools.

One area particularly within the Talent Acquisition application of AI, that is imperative as an industry we remain vigilant to is the risk of AI introducing biases. These biases can be, but are certainly not limited to, sampling bias, measurement bias, confirmation bias, selection bias, recency bias etc. The first step in mitigating this is improving data quality and diversity at the input stage, ensuring that the data set your AI is wielding is representative of the talent pool you intend to serve.

Whilst counter intuitive, in the current iteration of available AI tools it is important we examine the output data for any associations or correlations that could be unfair and or discriminatory. 

With over a decade’s experience in the space, the potential upside of AI implementation into Recruitment and Talent Acquisition is, in the main, exciting. It has the potential to improve access to a more diverse talent pool, enhance productivity and efficiency, freeing up hours in the day or days in the year to focus on other endeavours, professional or personal. It has the potential to improve candidate experience, engagement and foster a feeling of belonging much earlier in the process. It has the potential to drive inclusion and accessibility, and facilitate community. It has the potential to allow smaller businesses to cut through the noise created by bigger businesses, purely by virtue of their size. It is exciting, it’s effing exciting! 

However, this is not, we believe, a revolution in the world of Talent Acquisition and Recruitment with AI at its core, but rather an evolution enabled, supported and augmented by AI.

Keep me updated

Recent posts

PR, Communications & Digital Salary Guide: Agency 2024

Our salary guides are put together with a view to creating something as user friendly and helpful as possible that doesn’t rely on opinion but rather built from multiple data sources to remove any form of bias or subjectivity.