There’s a lot of speculation surrounding how artificial intelligence (AI) will change our jobs, our lives, and everything in between.
But how can we cut through the hype and see the technology behind the bravado?
Many companies claim to have created AI recruiting software, but which ones are worth our time?
I can still remember when I was testing the first “AI” product I came across called Chosen AI.
This was 5+ years ago, so their tech might have changed, but I remember it being a really basic search fitted on top of LinkedIn with horrendous UI and probably the worst weighing of criteria I’ve ever seen.
The algorithm only assessed company size, so if I fed it candidates from companies like Nike, Adidas, and Vans, it would recommend people from BP, Coca-Cola, and Mott Macdonald. The built-in LinkedIn filters did 100x better at helping find the right people.
To help you avoid such an experience, I want to help you recognize when a tool has been made with actual recruitment in mind vs. a repurposed use of generic machine learning tools.
Alongside that, I’ll also review a few AI recruiting tools I’ve been experimenting with.
Talent acquisition teams spend a lot of time sourcing and engaging passive candidates.
It’s one of the areas that the earliest developers of AI-driven tools were looking to target. But, to see which tools are truly impactful on the quality of hire, we have to understand how the process is done manually and what AI can do to match it.
When you’re sourcing candidates you look at things like:
- Job titles—same or similar ones (e.g. sales manager or account executives or business development can potentially mean the same job and experience).
- Companies—similar to yours or ones you may have picked up have similar processes, products, and services.
- Keywords—skill sets, technical qualifications, education.
Something to bear in mind is that recruiters pick up information about candidates and companies that the AI sourcing tools may not be able to e.g. knowing that people from company X have to go through a specific grad program that makes them a good fit for a role in something else.
Keyword searches have gotten better at recognizing alternative phrasing and job titles (outside of the outrageous ones), but they are still not perfect.
Natural language processing algorithms have been improving more and more, but it’s still not 100% and it may miss out on candidates you would have made part of your talent pool.
What it would take for an AI tool to be good:
- It recognizes alternative job titles
- You can query why it recommended a certain candidate to be a fit for a role
- It has a good number of filters so that you can manually set the search parameters to help it learn what you need e.g. location, type of company, industry, and potentially even seniority.
- Fully sourced data so if you contact a candidate you know where you got their information from (especially in GDPR locations). If the database has info like email addresses and phone numbers you need to know where it’s coming from.
- If it includes automated outreach and outreach schedule, make sure that you can modify and control it (e.g. if the market you are reaching out to is not responsive to unsolicited WhatsApp messages, which let’s face it most aren’t, then make sure it doesn’t include that channel of comms).
Interesting AI-powered recruitment platforms for sourcing:
- Crew—more of a CRM where you can nurture a community of talent
- Fetcher.ai—Candidate sourcing tool with candidates from many platforms
- PersanaAI—Assistant for Linkedin prospecting helping write personalised outreach messages
- ProgAI—more eng-focused sourcing tool using Github as a source
- HumbirdA—another CRM tool with some capabilities of an ATS like reviewing inbound applications.
Chatbots are great for automating a lot of customer service tasks, so why not use them to help candidates navigate your application process or direct them to information about your company, jobs, benefits, or interview process?
They can help improve your employer brand by making more information accessible and not stuck behind the first interview wall.
It can also be a more interactive way of asking some pre-screening questions to screen out candidates who do not pass the mark on some very specific requirements e.g. a candidate who needs visa sponsorship but who you don’t have a license for right now.
A lot of these questions can seem aggressive when positioned in a questionnaire format on one page, but interactivity has been shown to improve user experience in UI and candidate experience is no exception.
There are standalone bots like eightfold and some ATS like SmartRecruiters and iCIMS are rolling out chatbots to help with anything from job matching, introducing hiring managers to candidates, and streamlining the application process via front-loading the questions.
What it would take for the AI to be good:
The company can demo it with your data i.e. your job descriptions and questions that it can be programmed to ask and respond to based on your company’s information.
This is key with chatbots as many vendors have worked hard on a demo with preset data and it crumbles when faced with real-life work.
Similar to how we decide on who to reach out to when we source candidates, we make a lot of decisions when we review resumes and decide who to screen out vs in.
Candidate screening often relies on nothing more than a resume and maybe a few questions that the candidate has answered during the application.
How can we automate that fairly time-consuming task?
There are a lot of applicant tracking systems rolling out features that give a rating or recommendation on the candidates that apply.
Some use the answers candidates give to questions you set during the application stage and others give a score “based on the CV”.
I’ve spoken to a few of these companies and not a single one has been able to walk me through their algorithm.
Via my experiments, I’ve noticed that this rating system seems to function more like a keyword recognition system based on the information that gets parsed from the CV file.
Note on parsing: Usually ATSs have built-in CV parsing which is supposed to lift out key information and auto-populate fields like name, email, etc.
These have been weak and have only gotten weaker with the rise of alternative platforms to write CVs (e.g. Canva or LaTeX).
For example, Enhancv—supposedly using an industry-leading ATS—parsed only 40% of my CV, which means most of the information remains “invisible” and it wouldn’t know 60% of the information it’s meant to judge me on.
Based on the above, I’m dubious about the state of automated CV screeners and their ability to recommend the best candidates without anyone verifying that recommendation.
I can think of a particular example where it would not work well: grad hiring. What keywords would it recognise?
While it could be helpful in high-volume scenarios, if you’re finding that your talent pipeline is lacking quality candidates you may want to check the auto-screener and see what candidates it may have missed out on.
What it would take for an AI tool to be good:
As mentioned above, the state of these tools is still very much a work in progress.
While there are many libraries in the world of machine learning doing things like text summary, sentiment analysis etc, those are not yet adopted fully by most modern ATS.
Also being able to query a specific decision is so important as human biases are likely to be baked into the algorithm.
Often AI has been touted as a weapon against unconscious bias, but the jury is still out on whether the use of AI will lead to more equitable outcomes.
Interesting products I’ve come across are:
- PDF.ai—So far the summaries seem fairly accurate and it helps generate questions
- https://accio.springworks.in/—Very much in the beginning stages but it’s adapted GPT to summarise CVs based on job descriptions
Overall when it comes to CV screening, I would still recommend a lot of human hand-holding.
You’ve identified a candidate you’d like to take to the first interview and now you need to create an interview process to unearth the skills you’re looking for.
Creating interview questions
AI can help you create questions based on the job specification and the candidate’s CV. Leveraging AI to create interview questions tailored to job specifications is a key feature of advanced recruiting software.
My recommendation is to create a base set of questions you will always ask each candidate to get data that you can directly compare.
For that I use either Bing or ChatGPT4, both being internet-connected as they can also look for new questions online.
As an experienced recruiter, I use these as baseline questions that I can either modify or use for inspiration as the recommendations can be quite bog standard.
After the baseline questions, you can start thinking about candidate-specific questions. You can automate some of the work to analyse the job description and the resume of the candidate and recommend questions for you.
What it would take for an AI tool to be good:
- Being able to articulate what it looks for in a job description and a CV
- How does it formulate questions based on the criteria e.g. when it finds something missing that the job advert requires and is not mentioned on the CV or any other way?
- Test it—see what it comes up with in the summary and see if that is what you would have seen in a CV and what you would have asked.
Another thing I’ve seen the moniker of AI tacked on has been “AI” or “Smart” schedulers.
Frankly, I am not sure that I’d call it AI when it finds a slot in the calendar of an interviewer and a candidate.
Nevertheless, scheduling is one of the most time-consuming and repetitive tasks and it is a super useful automation to have.
It’s one of the major reasons I always try to roll out an ATS with a good tool to schedule interviews as that saves so much admin time and can help with the candidate experience overall.
Text analysis, summary, and editing are major features of generative AI. There are quite a few tools out there that can do the general “analysis” of a text but they can be quite cumbersome to implement as part of an interview process because it means adding more and more tools outside of the ATS that you have to remember to use!
While there are a few out there that do more general “sentiment analysis”—even ChatGPT can do it fairly well— these have been designed with recruitment in mind more specifically.
I’ve played around a bit more with Screenloop and there’s quite an interesting feature where it highlights key points where candidates mention experience that match elements from the job description.
Both can integrate with your ATS making them easier to implement—you don’t even have to think about “switching it on” or pasting the interview transcript for analysis.
A Note On Chat GPT 3/4
One of the most widely used and widely trained pieces of “AI” or LLM (Large Language Models) is ChatGPT.
I am making a separate note on it as it can be a helpful tool in certain areas even though it’s not necessarily built for recruitment. It is especially useful for generating content or helping with editing.
I use it for:
- Job descriptions—either to get a first draft or as a way to edit the text to make it more detailed, sales-y, or concise.
- Outreach messages—While useful, I’ve found it likes to generate very flowery messages.
- Careers page content and social posts—both for ideas and editing ones I’ve written.
Tips I’ve picked up:
- Use “act as” before any prompts e.g. act as a tech recruiter, or HR manager with a friendly tone
- Use as many descriptors as needed e.g. “informative, brief message that aims to...”
- Don’t be scared to ask it to “generate more”, “summarise” the content, or “make it more detailed”
- I’ve found that the more you play with it the more it learns some of the things you need
- Give as much detail as you can on what you need. At that point you might be thinking “Might as well write it myself” but it can give you ideas or another angle or perspective.
For more on using ChatGPT as an HR professional: ChatGPT: A Working Guide for HR Professionals
AI is useful for tasks like creating job descriptions or LinkedIn outreach messages, but it’s not yet something we can unleash unsupervised.
Recruiters aren’t out of a job anytime soon and we cannot outsource our entire recruitment process to the current wave of machine learning models.
While writing this article I made sure to consult with a few ML/AI engineers to ensure I’m not stating anything erroneous and one of them said something I’d like to leave you with:
“Many new products and technologies suffer from evangelists that fall into the trap of having discovered a cool new hammer, so suddenly everything they see is a nail. It is the same now with AI. It's a cool hammer.” - Dr Mihail Morosan, Deployed AI Engineer.
Make sure you are there to guide the hammer and see if the problem truly is a nail!
Some further resources to help you recruit better: