In the broader scheme of corporate evolution, Artificial Intelligence (AI), is no longer an underdog. AI is devouring data at a ravenous speed, like a Pacman and the kind of intelligence  and insights we are receiving is helping us make quicker, better and informed choices.

Below are some of the view points as shared by the panelists; Mr. Satish Rajaratnam – Senior VP HR, Mphasis and Mr. Pulak Ghosh, Professor – Decision Sciences, IIM Bangalore in conversation with the Moderator Ms. Nitya Vijaykumar, Sr. Knowledge Advisor, SHRM India during the Roundtable Discussion:

What are the most compelling use cases for artificial intelligence in recruitment?

For the past few years, organizations are using AI as a combination of NLP (Natural Language Processing) in speeding up the recruitment process through automation of screening resumes to find an ideal fit for a particular job. In the USA, MIT graduates have created a program where some organizations use this combination for new hires where the algorithm is taught to analyze the first 2 months of their email communication to identify their flight risk. Based on what email the new hire is sending out of their mailbox, to whom is it sent, what time is the email being sent, what words are being used, how short or long the email is, etc. the program identifies if the new hire is able to fit in the culture and if the employee is happy or not.

Some organizations are taking these forensics a step forward by analyzing some of the aspects such as the tone used by their managers, escalations received or harsh words used that would reduce the productivity or the efficiency of that individual during the day. This helps identify the attrition risks and also the dissatisfaction levels within the organization.  This is done with the precursor of the caveat that the organization lets the employees know that it  might pick up some tonalities and words from their emails.

The other use case from a recruitment perspective is about parsing – a cliche word in any applicant tracking system. However, most organizations overlook the interviewer’s or hiring manager’s behavior. Mphasis analyzed the audio recordings of a set of interviews over a period to time, using NLP in terms of heuristics to identify the panelists who spoke more about themselves.

or an  It was observed that the panelists spoke more about the skills they are seeking out than about the organization’s EVP, did not give enough airtime for the candidates to voice out their views, bombarded the candidate with a lot of questions etc.

Further, the organization  focused on the hiring manager evaluation to understand the higher or lower rejection rate for the 100’s of candidates applying for a job. Using such a technique, Mphasis could address the pain point of the recruiter in the supply chain ecosystem.

As the focus  is shifting from the candidate to the hiring manager; one important question that organizations are seeking is –  “Can AI successfully eliminate bias and are there any inherent limitations?”

From a practitioner point of view, the answer oscillates between an yes and a no. Over the years, AI has been able to screen resumes to match a JD. Today, there exists tools that can identify if a candidate has taken help of other tech tools to write the resume or it is written by self.   It is seen that, whenever a new technology pops up, there is another technology to beat that.

Mphasis wrote some heuristics where they started interviewing using a conversational AI bot to remove the bias. The bot was able to learn and mature to the point where it would set up a meeting with the candidate for an interview and learned to ask probing questions based on the response from the candidate, revalidating every single item that is needed in the JD. The process started off the conversation with questions like –

• “We thought we will go through your profile together. Could you quickly take me through? It’s a very interesting profile. “
•  
• “You’ve been in this company for X years. What were your key responsibilities?”
•  

After a 45 mins interview, the bot will report to the recruiter if the candidate is ideal for the job role. However, the bot will not be able to take decisions and will only be able to through insights as well as suggestions. For example, if the JD requires human capital strategy, but looks like the candidate has experience only in HR operations, you might want to probe more again, which the bot will not be able to do. The NLP program is also able to filter out candidates by identifying lip sync issue with the audio to check out ifis the candidate looking at someone else when answering, and  identify if the candidate is being prompted with the answers (by analyzing the gaps between the questions asked and the candidate’s responses). Another example is where the AI asks a candidate to type a code with intelligent complexities; the AI then analyses the code if it’s written by a human or an AI (Chat GPT).

This technology is helping recruiters  and  sourcing teams  to become more efficient and achieve the metrics both from a time and quality perspective.

What do you think are the inherent limitations of AI in recruitment?

Currently, AI is not intelligent!

Imagine it being an Almirah full of files stacked. You are looking for a particular page; the “AI” searches through the stack in a second and displays the result. And that is where the bias comes in. To elaborate on the analogy; the entire system works on the dictionary. The software learns by the information you provide and if your information set is biased then eventually the program will become biased. For example, if an organization recruits more females, the algorithm will prompt the recruiter to recruit more females. If a bank provides more loan to individuals with a higher salary, the algorithm will prompt loans to individuals with a higher salary. It is important for us as users to leverage AI as aids and not to replace us. It is important to figure out when and where we need to interject and correct the path. When talking about bias in AI, it is problematic from a regulatory perspective. Who would be responsible if the algorithm makes a mistake? A suitable answer to this question is going to remain a challenge for a  long time until the algorithm becomes more intelligent.

Where can AI really be put to good use?

Faster decision making and minimizing error is the base. Technology is best used to start by reading a resume – figuring out the skill set required for a role, training the manager – figuring out where the manager may be going wrong etc. There are a plethora of applications coming in and technology can help when the recruiter is not sure of the skillset required for the role and further achieve speed and accuracy.

What are the typical challenges an organization encounters when embarking on this journey of AI in recruitment?

Often HR teams project the productivity, efficiency, quality of hires that can be achieved, while pitching AI to the CXO team.  However, they tend to miss out on the underlining area of revenue leakage. With the help of AI, time to fill will shrink. Mphasis built a case for requiring AI in recruitment in the organization. The team reported that their conversion ratio from applied to screening is about 23%; meaning the sourcing team will keep searching for the ideal candidate. Using a slightly elevated parsing tool, they can onboard a candidate as much faster as by 34 days which can save revenue and save on revenue leakage by 34 days per candidate for X number of people. They also took it a step further by introducing design thinking by applying the Cartesian principle to the 5 areas of recruitment: sourcing, recruitment, offer rollout, post offer follow up and candidate experience. The recruitment team leads were asked what will happen to the 5 areas if they were to move to AI?  It was identified that the perks of this would be  the reduction in complexity for the hiring team; reduced  time to interview, faster intake of quality hires, as well as reduced outsourcing interview cost.

The teams were then told they may not get investment for all 5 areas but were asked to identify the areas they think the investment on tech will minimize a lot of their effort. They looked at  minimal effort and maximum impact as the key criteria in identifying the area for investment.  They went step by step and that worked wonders for the recruiting team because at every stage they showed productivity  and cost improvement..

Matching the skill set and candidate to the JD is a challenge as today recruiters receive over 50,000 resumes for a vacancy and most of the resumes are misleading and do not qualify for that particular JD.

The ROI on AI in recruitment is another concern because of which organizations shy away from implementing AI. Today, recruitment teams want to standardize the hiring process across all the verticals and clusters. For example, if a vacancy has only 15 candidates in the market, an investment in AI won’t be economical. On the other end, if there were tons of resumes available for a vacancy, AI is a suitable investment. The dilemma is, should the team invest in only this segment and not the other? This dilemma is solved by standardizing the process to implement AI across.

What are some of the skills that will be actually required to start making AI more impactful ?

If you look at the algorithm, there are two aspects to it. One is using the algorithm, the other is interpretation of the output of the algorithm, and that interpretation is becoming way more complicated than running it. And this is where in the western world, the Big 5, Amazon, Facebook etc. for past 7 to 8 years,  are hiring 5 non engineers for every engineer required. Basically, an organization needs to have an interdisciplinary team to make AI effective  and  not just engineers.

For AI or for the algorithm to work,an organization needs to save their recruitment data including the interview data. This will ensure that AI is being leveraged for the right use in the organizational context of demand and skillsets required.

What aspects of the current graduates should a recruiter be aware of now?

Candidates should be aware of some amount of coding no matter what stream they belong to. If one knows coding, their thought process becomes more streamlined. Top technology firms of the world are generally 5 to 7 years ahead in what they want when compared to the rest of the world. They are clear about what  and when they want it. It could be a good starting point to look at how their recruitment structure is changing. For example, Facebook ensures their leaders write codes frequently as a mandatory requirement for their role even if they are not actually writing codes for the website. Irrespective of which industry one hails from, everyone should know what coding is. Three aspects that are key to this:  one – they should be savvy in their domain, two – they should be strong in their functional capabilities, three – they should possess leadership behavior traits.

Is there a threat to recruitment professionals from AI taking over?

Only if one is not aware of AI and how it works, there is a threat. It’s better to be qualified (AI for dummies)  to be a little tech savvy with the technical terms like: what is a front end, what is the middleware? What is the back end? What is the database? What is cloud? etc. to know what AI can do and cannot, as well as where the recruitment professionals can come in to bring in their intellect.

AI will create plenty of jobs. However, what are the specific skills required in future to manage AI?

One fundamental concern with AI is that, it is rapidly changing in nature and in 2 years every technology becomes obsolete. This concern requires the human to constantly keep upskilling to survive in this era. However, one needs to be very careful and aware about  area where upskilling is required. One needs to look beyond India towards the west to figure out what technology is being developed there and learn accordingly. The Skilling Horizon needs to be worldwide.

The second part to this question is human engineering; prompting to the engineering side of AI. The emphasis is practice. For example, from a maturity perspective; a 5 year plan to making AI a mandate process in recruitment gives enough time for a professional to upskill and learn the technological skills required to operate an AI tool.

To summarize, AI in recruitment, What  it can and can’t do?

AI can enhance efficiency depending on how and where one applies it, what insights the human wants for taking decisions. AI can save time, effort and cost.

However, there are certain risks that recruitment teams need to be aware of before leveraging AI.

1. 1. In a few years, AI bots interviewing candidates are going to be a lot smarter than the people who they will work for; and this will be a challenge as the younger generation embrace technology.
2. 
   
3. 2. The organization may develop a very good AI system but the biggest question to ask is who will take responsibility for the system. IIM Bangalore recently worked with Duke Medical school to develop a system for predicting what other kind of diseases would the patients with a heart disease have. However, when the system was implemented; there was a revolt. Before the system was implemented, as soon as a patient was admitted, the nurse would be the first point of contact to diagnose. If there was a concern, the nurse or the resident doctor would inform a senior doctor to make a decision. The new AI tool when implemented was directly notifying the senior doctor about a patient with a 92% chance of a disease. The senior doctors realized that the AI was over-empowering them while skipping the matrix, which did not go well with them. In other words, the more AI intrudes in your day-to-day life including your systems, you won’t know how to deal with it.

In summary- there’s a lot that artificial intelligence can do, but not without human intelligence.