Let me guess: You’re someone who is supposed to have all the answers (and usually does), but right now when your head of compliance, legal, or comms asks you to define your strategy for AI in recruiting, you’re swinging from the heels and coming up empty.
Good news: You’re not alone. AI has proven to be quite the wicked curveball.
This an entirely new and uncharted era of talent acquisition — one where the possibilities seem endless, but unfortunately so does the confusion and uncertainty. For an industry that felt like it was stuck in the same place for a couple of decades, the whipsnap speed in which AI technology is evolving has made it nearly impossible to keep up. Many of the talent leaders we work with who have been trusted to know all the things at all times are now also being asked to be AI experts at a time when even AI experts don’t always have all the answers. Everyone has been reduced to a series of search queries and scrunched up faces.
But here’s the thing: You don’t actually need to know everything about AI in recruiting and talent acquisition. Just some of the things. The most important things. The things that matter to you and your teams and your problems … and ultimately, your solutions.
These are those things.
This guide is meant to help you focus on the “right stuff” when it comes to AI, with practical advice, void of all the buzzwords and acronyms and senseless “what if” hype.
Feel free to click around, scan, peruse, and parse at your leisure. The goal is to turn you back into an unstoppable force and transform AI into a very movable object and another tool in your toolbelt — one that you can bend and shape to your liking. If AI is a curveball, consider this your crash course on how to knock it out of the park.
A brief history of AI in recruiting and talent acquisition.
It wasn’t too long ago that the term “artificial intelligence’” conjured up images of walking, talking, honest-to-god robots in most people’s minds. As far as cultural footprints go, the Tin Man left a pretty big one way back in the 1930s, and that legacy of unrealistic robotic expectations was carried through the decades on the inorganic shoulders of icons like HAL 9000, C-3PO, and Johnny-5.
Of course, we all know now that AI — for all of its recent advancements — is still far more rudimentary than that. Our current cultural benchmark for AI is less sci-fi and more Siri; it’s largely seen and approached as a useful tool to help support humans, not outright mimic or replace them.
That change in perception happened fairly gradually (and lately, not so gradually) so it’s gone mostly unnoticed, but it’s actually quite a profound shift from where things started.
The Turing test (also known as the imitation game, as made famous by Benedict Cumberbatch) — probably the most ubiquitous AI-related testing format — was conceived in 1950 in order to measure a “machine’s ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human.” The goal back then was to determine if a machine could actually think. In the 75 years since then, we’ve learned that the answer is unequivocally no.
And so the game has changed.
As AI advanced, our expectations for what it can (and should) be used for seemingly has gone in the opposite direction. Even though we’re closer than ever to actualizing it, the dream is no longer centered around robot butlers or humanlike replicants; we just want some functional stuff that’s going to make our lives a bit easier. Very simply, AI in its current form has shifted to how it can help us get work done and be more productive, instead of how it can replace us all and create immeasurable labor market chaos.
Which brings us specifically to AI’s use within recruiting and talent acquisition, and how we got from there to here. Specifically, there are some very clear “jobs to be done” where AI can be a force for good in our work. Consider the things that humans do today — and then consider whether they should be doing those things at all:
- Screening resumes for basic requirements
- Reviewing calendars and manually scheduling (and rescheduling … and then rescheduling again) interviews
- Answering repetitive questions, like “can I bring my dog to work” or “what should I be prepared to answer in my interview” or “what should I wear on my first day”
- Writing job descriptions and copy for email or text campaigns
- Collecting feedback from managers and candidates after different stages of the hiring process
But before we get into whether AI can help get that work done, let’s take a step back to examine AI’s historical role in recruiting and hiring.
A timeline of AI’s rapid rise and evolution in recruiting and talent acquisition.
It’s admittedly pretty hard to pinpoint exactly what the first examples of “artificial intelligence” were in talent acquisition, because there’s varying definitions of what constitutes “AI” and even more names that people have used to describe it over the years.
You might remember that one time everyone redid all their marketing to include so-called “big data” — then later just decided to call it “AI” instead. But essentially, what we generally refer to as AI came to be used in talent acquisition in order to solve two main problems.
The first is “matching” — essentially, taking a group of resumes and a group of jobs and seeing how the words match each other. There have been many attempts to solve this, from parsing vendors like Textkernel, DaXtra, and Sovren to Bright Data (sold to LinkedIn), My Perfect Gig, and many others in the jobs space. Many servers burned brightly using the technology of the 2000’s and 2010’s to infer the meaning of words and get high quality matches. For example, it was pretty easy to figure out that C# was a computer term (or a music note, I guess), but it took some understanding of other words in a resume to know that java and javascript aren’t related, and that neither of them have anything to do with coffee.
These approaches were okay, but never really moved the needle in the industry. The application of machine learning to understand what terms really, really mean in an organization by seeing what recruiters actually do was the key advancement of companies like HiredScore and Ideal.com that led to their acquisitions. Alongside heavy compliance and explainability guardrails, the matching industry took leaps forward. But there were also some false steps, most notably by Amazon in their famous disclosure of an internal project to grade resumes that showed significant bias against women and minority populations. It was a cautionary tale that the guardrails weren’t just nice to have — they were essential.
There were also companies along the way that looked to automate other parts of the process. Several companies (Entelo, humanpredictions, and others) made attempts to model when employees may be looking for a new job. HireVue made waves by using facial recognition software in video interview processes for a small number of clients, which sparked lawsuits and legislation in Illinois against the idea.
But the other major application of AI really came in 2016 with the early recruiting chatbot companies, seeing the application of that emerging technology to recruiting. Mya, AllyO, Wade & Wendy — and Paradox, of course — were among the first to explore conversational AI, harnessing its power to automate routine conversations around screening, interview scheduling and the many other not-that-interesting conversations that happen in recruiting processes.
(Quick aside: It’s worth mentioning that at this time there was the enigma of Watson, IBM’s supercomputer that could do chat, beat you at Jeopardy!, and diagnose healthcare challenges. While Watson’s promise never really materialized into anything meaningful for the talent acquisition space, its substantial marketing and pop culture influence opened many eyes to future possibilities).
Now, in most of the late 2010’s and early 2020’s, chat was good, but not great, lacking some of the humanizing elements that it needed and often giving frustrating “does not compute” error messaging. Because of those limitations, chatbots earned a mixed reputation, especially as a customer service tool (although if we’re being honest, it was mostly just negative). As deeply flawed as it was, early chat used in talent acquisition was still able to make a huge impact by simply automating narrowly defined tasks such as candidate screening, interview scheduling, and question answering.
Even still, the idea that a “chatbot” would ever be able to truly transform the hiring process and candidate experience felt farfetched. But as conversational AI became more sophisticated in the early 2020s the idea of truly “invisible” and “messaging first” hiring software started to seem not as crazy. Then, in late 2022, everything changed with three letters: GPT.
ChatGPT — the first mass adoption of a chat-based assistant based on a large language model (LLM) — was truly a Mentos-in-a-Coke-bottle moment; an instantaneous explosion of awareness and attention that finally put the AI in “mainstream”, showing the world what generative AI could really do. It wasn’t (and isn’t) perfect, but for the first time, a future fueled by AI-based natural language felt real. Tangible. Doable.
It brought tomorrow to today’s doorstep. And now we’re all getting ready to walk through it.
An even briefer explanation of how AI is built.
It’s important to understand that talent acquisition is usually a later adopter of other technology trends. Typically, we see widespread adoption in consumer industries first, and recruiting follows once it’s been validated or proven to drive value. TA posted to newspapers until readership declined and then followed that audience to the web. In technological change, it’s the tail not the dog. This makes it fairly straightforward to figure out where technology is generally headed in talent acquisition, though the specific application of those technologies is always a wild card.
First, the AI nearly every application uses is based on a “foundational model” that understands patterns. Foundational models are made by OpenAI, Google, Meta, Microsoft, and a few other focused companies that apply their efforts toward this task. These models require huge computer power (fashionably now referred to as “Compute”) to understand patterns in massive amounts of data. They create “large language models” (LLMs) like GPT-4 or Llama-2 that are trained on the entire history of human knowledge that lives in written text, or image models like Midjourney or DALL-E that can generate images by training on every image imaginable to arrange pixels with stunning accuracy.
As a talent acquisition practitioner, I know that all probably sounds like random gibberish.
The good news is that you don’t need to know much about these models other than where they are hosted (where data you send it will go) and how they will use your data (will they train on your data or not). For most enterprises, staying within a trusted infrastructure and minimizing training on any proprietary information is important. There are pros and cons to foundational models on things like speed, cost, trainability, accuracy for certain applications, but for buyers of talent acquisition software, the analysis is best done on the application, not the foundational model.
How do generative AI models work?
In short, really big computers are making extremely good statistical projections. So good that it can predict what words should go in order to respond in a human-like way. What’s different about these LLMs is their accuracy and their ability to use human language in ways that haven’t before been possible. It’s all due to the increase in underlying “Compute” power that makes this possible at a staggering scale.
Fine tuning of models.
The next “layer” in the AI cake is fine tuning. Fine tuning is where developers will use a small set of training data (remember, almost anything seems small next to a LLM trained on all of human knowledge) that is specific to a domain. In order to work really well for a specific task, many companies add additional data inputs to train an AI model to a certain task. For example, you might give AI tens of thousands of X-ray images and note which ones have broken bones and which ones don’t. You might give it thousands of outbound recruiting messages and ask AI to write in a certain style. Through feedback loops and lots of trial and error, the AI model can be trained over time to recognize new patterns.
It’s also important to note that there are many different models for applicant developers to use for different use cases. For example, in Paradox’s AI we use a trained AI model as a routing tool to understand what the context of an interaction is based on the history of the interaction, the context, and the words. From there, it routes the work to other models based on what you want to do — if we’re trying to capture a name, a name recognition model is used. If you’re trying to ask a question, a question-answering model is used. These different models work together as a team to get work done, each with a specialty based on its training.
One important type of model is called retrieval augmented generation, or RAG. This is the type of model Paradox uses for Q&A to be sure that correct answers are given when a question is asked. This technique involves using a knowledge base to get answers to a question rather than generating them using probability from a trained model. It’s less “creative” but in a good way for many tasks, like answering questions about a job.
The four core functions of AI in recruiting and talent acquisition.
Artificial intelligence is such a dense topic that it can often feel impenetrable. Even the most basic questions unearth labyrinthine results.
How do I use it?
Yeah, good luck untangling that twisted web. Better clear your schedule and call the babysitter.
However, when it comes to using AI in recruiting, the answers are actually fairly straightforward in these early stages. Really, they are. You can essentially distill the ever-growing number of AI tools and platforms into four core functionalities. And within those functionalities, there are a clear set of tasks — and solutions — that AI can provide your hiring teams.
Now, depending on the tech vendor, AI can actually serve more than one of these functions, and in some cases it will end up serving none at all. The critical thing here is developing a very clear understanding of what your root problem is, so you can work backwards to find a technology that serves the exact function(s) you need to best support your business.
In the early days of AI, here are the patterns of how AI is being used and deployed today:
Copilot.
It speaks to the ubiquity of AI tools that there is literally an AI chatbot from Microsoft with the same name, but that’s not what we’re talking about here. When we say copilot, we mean it as a concept. The idea here is that an AI assistant is ever-present in a recruiter’s flow of work, serving as sort of a support system to make them better. The copilot is there to answer random candidate questions and to own specific types of work. If you’re late to an interview and haven’t had much time to prepare, you can ask your copilot for a list of the best questions to ask. The great thing about using AI this way is that it clearly puts the user in the driver seat and allows them to dictate how much or little they use AI to enhance them. On the flip side, it can sometimes be a challenge to fully understand their limitations and the ideal ways to utilize them.
Writing assistant.
You probably know this better as “ChatGPT”. While certainly not the only game in town (Grok, Gemini, etc.), it’s easily the most ubiquitous generative chatbot going right now. It’s basically a prompt machine — a prompt goes in (write me a limerick about purple squirrels) and ChatGPT uses a large language model to generate a (mostly) accurate response. You can even “prompt engineer” the AI to meet your desired length, style, and tone. If you’ve messed around with these tools at all, you know that a lot of the outputs end up being more fun than function, but there’s already some pretty obvious use cases manifesting within talent acquisition. One of the most perplexing and counterintuitive parts about companies using AI is that candidates are also using AI in their job search.
This creates incredible problems for AI matching tools that have sought to match the words on a resume with the words on a job description. If the “perfect” resume for a certain role is created using AI, finding a signal is going to be a challenge. TA traditionalists might view that as “cheating”, but it’s simply current; technology has superseded the need for everyone to write everything manually. In fact, it’s just a plain old waste of time. You would have thought people were insane if they kept walking everywhere after 1908, right? Recruiters can leverage ChatGPT (and GPT-adjacent tools) to help craft things like job descriptions and standardized candidate comms so they can invest time and brain power into more important work.
Conversation assistant.
Conversational AI is baked into the technology of some of the things we’ve already mentioned, but it also has standalone functionality. The idea here is that you can use AI to make the hiring experience more human by giving candidates a way to communicate and navigate their hiring experience in a purely conversational way, through text or chat. AI making things human might seem counterintuitive, or even outright dystopian — but it’s actually more of a throwback concept than anything else. People are hardwired to get things done through conversations, we’ve just never been able to do it at scale until now. A recruiter providing a high touch experience to one candidate for one role? Easy.
Doing that for hundreds of candidates across dozens of roles? No way.
Things like CRMs and ATSs were created to help scale growing hiring demand, and while it sort of worked, it also created a lot of new barriers and friction points. AI eliminates those barriers and allows for real-time, automated conversations with candidates using natural language, 24/7. Applications become conversations, and candidates gain a reliable, ever-present touchpoint in familiar format, just in case they have any questions (what to wear to the interview, directions, company policies, etc.) In a sense, it makes hiring software “invisible”, so clunkiness like logins and forms get ghosted, but your candidates don’t.
Matching.
This is perhaps the most controversial function given some high-profile misfires in the past, but when used correctly it’s a valuable tool that serves as an extension of the recruiter or hiring manager. You can take the term “matching” quite literally here — AI matches things to other things based on pattern recognition. You can take a role and define certain keywords (skills, experience, etc.) that indicate a “match”, and the AI can identify those in a resume to provide a “recommendation” on good fits. Now, recommendation is in quotes, because the AI isn’t really recommending anything; it doesn’t pass judgment in any way, it’s merely making binary, fact-based decisions based on parameters set by humans.
Does a candidate have at least five years experience? Do they have a programming background? It’s either yes or no — and matching AI can identify these things with speed and accuracy far surpassing a recruiter.
Essentially, AI’s overarching value right now in recruiting is helping employer’s hire with more speed, accuracy, and intention than with traditional methods. There’s this misconception that AI is somehow adding in completely new complexities, but it’s really not — it’s simply augmenting (and improving) the same old recruiting and hiring challenges organizations have been faced with since everything was pen and paper.
Here’s a final, very critical thing to emphasize as it pertains to all of these functions: Humans are in charge. They’re the captain, now and for the foreseeable future. AI is augmenting people and transforming the work they do, not outright replacing them. It is replacing some tasks — but it’s the tasks that your hiring teams shouldn’t be doing, anyways.
The difference between chatbot, conversational AI, and generative AI.
As more and more AI terms enter the mainstream lexicon, the more everything seems to meld into an amorphous blob of buzzwords. We use certain phrases interchangeably, even though they mean completely different things (to thaw or unthaw, that is the question).
The three listed above are the biggest culprits. So for the sake of posterity (and clarity), here’s what they actually mean. a
Chatbot.
Chatbots are computer programs designed to simulate human conversations with its user. Early versions were rudimentary, with significant improvement in the mid 2010’s with the use of Natural Language Processing technology. Chatbots were often frustrating to use and a lot more “bot” than “chat”, only capable of rudimentary responses and a limited understanding of user prompts. Many people have allergic reactions to bad chatbots used for customer service that weren’t good enough, so the term “chatbot” has become associated with frustration until breakthroughs in technology made them much smarter.
Conversational AI.
Here’s a little secret: ChatGPT is a chatbot. Grok is a chatbot. Paradox’s Olivia is a chatbot. Now, we don’t use that word because of the visceral, negative connotations associated with it (we prefer to call Olivia an assistant), but it’s a chatbot by definition. ChatGPT has the same chassis as the maddeningly unhelpful chat widget sitting on that retail site you’re trying to return something on — but what’s under our hood is much, much different: Conversational AI that is capable of more advanced, human-like responses. That being said, there’s still limitations; chatbots powered by conversational AI need to be “trained” on their responses, and are typically not capable of responding to complex intents (multiple questions in the same prompt, for example) and don’t possess advanced contextual awareness. Standard conversational AI is predictive and great for when you need a response to be super specific.
Generative AI.
Generative AI is the newest evolution of conversational AI, infusing it with the ability to generate more complex text, images, and video based on prompts. The magic of generative AI is the large language models they’re based on, giving them a wide breadth of information to draw from in their responses. Generative AI possesses contextual awareness, which means it doesn’t require super specific prompts in order to generate an accurate response. It can understand multiple intents, follow up questions, and context in a way that feels more akin to how humans interact with each other in real life.
Ultimately, generative AI layers in a level of nuance that helps power “chatbots” to provide the best candidate experiences we’ve ever been able to build. The slight downside, of course, is that generative AI — by design — isn’t completely predictive. It’s possible to put guardrails on it based on the language model it’s pulling information from, but you can’t predict exactly what words it will use in what order.
What parts of the recruiting and hiring process benefit from using AI?
That’s pretty simple: All the ones you don’t want (or need) your people to be doing.
When it comes to technology, it’s very easy to get pulled into a never ending game of “keeping up with the Jobs’s” — a rat race of one-upmanship that ultimately results in a junk shelf filled with tools and tech that you didn’t really need and never actually used. Not to mention all the wasted hours on implementation, training, etc. Yikes. Instead, we advocate for essentially the inverse methodology when it comes to AI: Pinpoint the high-value tasks only your people can do and have them focus on that. Then, find the right AI tools that can automate everything else.
In other words, a “human-first” approach to AI that strikes the ideal balance between AI automation and human judgment.
You want your recruiters and hiring managers to focus on people-centric tasks and critical decision making (interviewing, overseeing store operations, hiring). Yes, that opens up a lot of extra admin work — hours, actually — but that’s what AI is best at automating right now.
Fast, accurate, scaleable. Fact-based, binary decision making. Always on, conversational communication. In other words:
Pre-apply (career sites, learning).
Most career sites have become “brochureware” — pages and pages of static content filled with generic stock photos of smiling faces and hollow brand values. They’re hard to parse through and harder to trust because the content relevant to any given candidate is buried under layers of genericity (yes that’s a real word, we checked).
AI transforms career sites into something much simpler but also much more dynamic; something that can actually evolve in real-time based on what each candidate wants to know. Imagine if instead of digging through pages and pages of career site content to find an answer to the one thing you’re interested in, candidates could simply engage with a friendly, helpful chat assistant on the site that’s powered by conversational AI.
This would allow the candidate to immediately find what they want, in a frictionless, conversational way. They could ask it anything: “what’s your PTO structure”, “do you have any benefits for veterans”, “can I bring my dog to the office”, and the site will literally serve up related content right on the page. Essentially, the career site becomes a unique and specific experience for every candidate, cutting out clutter and delivering only the information they care about. We’ve seen this reduce dropoff, increase conversion, and — as a bonus — save hiring teams tons of time they would have otherwise spent handling pre-apply or pre-interview questions.
Screening, matching, and applying.
This is really where AI starts to hum — but also where risk can be introduced based on how the AI models are used.
On the positive, less risky side, AI can help with screening, matching, and applying by turning formerly (no pun intended) exhaustive, form-filled applications into short, sweet conversations. Candidates can enter this stage in a number of ways — maybe they’re already on the career site doing some research (see above), or they’ve seen a “WE’RE HIRING” poster in a window, or they’ve scrolled past an ad on Instagram that has a text to apply shortcode — and once they’ve started to chat with the AI assistant on the other end, the whole processes happens right there in just a few minutes. No extra links, passwords, signups, or intervention needed. AI can serve up the most relevant jobs based on keywords (through the conversation itself or uploading a resume) and ask pre-determined screening questions to determine qualifications.
The whole process can take just a few minutes, and can happen at any time of the day. For some high-volume hourly employers who need to hire thousands of a certain role, like package handlers at a distribution center for instance, this is literally the entire hiring process; candidates apply on their phones, get qualified based on minimum qualifications, maybe even get a job offer in minutes, and start working within a day or two. Store managers never have to shift focus away from actually managing their stores, and candidates never do anything other than text. That’s truly invisible software.
Now, for the more risky side. AI today can also help you “rank” and “score” candidates based on their profile, background, and skills. While this use case seems like a panacea for recruiters overwhelmed with candidates and hiring work to manage — allowing them to quickly sort 500 applicants by the ones who are the best “fit” — it also introduces incredible risk. If the models being used to rank and score candidates are doing so on subjective criteria and those models aren’t transparent (what’s sometimes called “black box AI”), then you as the employer assume the burden of proving that the AI’s assessment of a candidate is bias-free.
There are countless examples of this gone wrong, but the most famous was technology that Amazon built in-house discriminately disqualifying or deprioritizing women from certain positions because, historically, very few women had been hired in those roles. The AI was modeled on historical data and made decisions logically based on that data. But it also did so without nuance and context. And in recruiting and HR, that can be dangerous.
Interview scheduling.
You no longer need to light a lantern when it’s dark or make a fire when it’s cold, so why are your hiring teams still scheduling their own interviews? This is the BIG ONE — the literal no-brainer. We’ve simply advanced past the need to manually schedule candidate interviews; with AI, it’s as simple as flicking on a lightswitch. Actually, quite literally. So if there’s just one thing you use AI for in talent acquisition and recruiting, it should be this, because automating simple tasks at scale, with accuracy, is what AI does far better than humans. Conversational AI can sync up with the calendars of all stakeholders in the hiring process (from hiring managers to recruiters all the way up to the C-suite) and serve up only available times to qualified candidates. Once a candidate selects a time, the interview block is automatically placed on corresponding calendars. If the candidate needs to reschedule, all it takes is a simple “what other times are available” text to the AI assistant and it happens automatically. Interview reminders are conversational and automatic, too. Gone are the days of playing calendar Tetris. The ROI on using AI to automate this one thing is a complete chart smasher — we’ve seen 50,000 hours and up to $2 million saved annually.
The electric future of motor vehicles has arrived.
Candidate communication.
For starters, robots don’t sleep. So while it’s physically impossible for recruiting teams to be working 24/7, 365 (unless you have an unlimited TA budget, which you don’t), AI is always on, ready to chat. But in addition to the conversations your people can’t handle, AI can also automate the conversations your people don't need to handle. Basic company information, simple logistics, random FAQs — this stuff doesn’t exactly require an elegant human touch, it just needs to be fast and accurate.
Offer, onboarding, and post hire.
There’s a misconception that the hiring process ends at the offer. It doesn’t. In fact, there’s a significant amount of new hires who never actually make it to Day 1, especially in hourly roles. Lack of communication (or the complete absence of it, in many cases) post-offer leads to confusion and disenchantment, which leads to no-shows. In other cases, the clunky process of receiving and completing certain forms is enough to drive your new hired employee into the arms of a competitor down the street.
Conversational AI assistants give candidates exactly what they need here: information and confirmation, fast. Assistants send onboarding documentation in the same text channel the application took place in, and candidates can quickly fill everything out right on their phones. Plus, the AI assistant is always there to help answer any questions the candidate may have, and can send reminders that nudge the stragglers.
AI assistants act as a safeguard to late-stage candidate drop-off, and reconfirms to candidates that they made the right decision to apply at your organization. At a part of the process where candidates just want to move on from all the legal documents, they’re given a fast, simple way to do so — leading to a 70% increase in post-offer retention.
How does AI impact recruiters and hiring managers?
Let’s start with the bad before we get to the good. Or rather, the lack of bad. Here’s what AI is not impacting (that you might have feared it was):
- AI is not completely replacing recruiting teams.
- It’s not making recruiting processes more complicated.
- And it’s definitely not creating more work.
- Oh, and it’s not making things feel “less human”, either.
AI doesn’t replace recruiters and hiring managers, it makes them better; AI doesn’t create more work for them, it removes it; AI doesn’t facilitate robotic, stilted experiences, it actually puts an ever greater emphasis on person-to-person interactions.
In other words: Augmentation of historically dreadful processes and roles.
And in other, other words: Enabling recruiters to actually recruit, and store managers to actually manage their stores.Imagine if you had an assistant to help you get recruiting and hiring work done. What would you have that assistant do for you?
In a sense, this advancement in AI simply allows us to get back to basics. Recruiters were never meant to toil away in front of computer screens, exhaustively navigating clunky software for hours on end. Their true value is cultivating a sense of trust with applicants and identifying the right candidates to hire so their organization is constantly fueled with the talent it needs. Only a human can do that. Well, only a human with adequate time and resources.
AI gives them that time.
So what’s the biggest impact AI has in talent acquisition? It’s automating the tasks that recruiters and hiring managers shouldn’t do (scheduling interviews) or can’t do (screening dozens of applicants at once), and giving them the time back to focus on the tasks that only they can do (preparing for interviews, building trust with candidates, making final hiring decisions). AI is augmenting what it means to be a recruiter and enabling them to focus on more valuable, fulfilling work.
In a sense, this advancement in AI simply allows us to get back to basics. Recruiters were never meant to toil away in front of computer screens, exhaustively navigating clunky software for hours on end.
OK, now about that word. Augmentation. It could be a little scary, but don’t let the TA industry’s glacial adoption of technological advancements or aversion to “replacing” people get you hung up — the concept of “augmenting" work isn't a novel concept. It’s been around for years and is applied effectively in a number of different fields. For example, AI is used in military defense systems to target dozens of missiles traveling 5,000 miles per hour and shoot them down instantaneously. No human could do that. But when the tables are turned and the military needs to decide to fire at an opposing force, only a human can do it. AI does not possess the moral, ethical reasoning, nor the accountability to make those types of judgment calls.
The decisions being made in TA everyday are less dire, of course, but still critical in their own right. And the same logic applies. Olivia handles the interview scheduling for many of our clients, scheduling thousands of complex interviews per day while speaking over 100 different languages. No human could do that. And yet, when the emotion gets deep on a life-changing move, convincing a candidate to move across the country or take a job with a promising startup, AI can’t gain that person’s trust to convince them to join. Only a human can do that.
Without AI, the job of a modern (let alone a future) recruiter is simply not scalable. And without humans, TA as a function is simply not controllable or manageable in a way that it needs to be to ensure business success.
How does AI impact candidates?
Well, it makes everything faster.
And easier.
And better.
It’s really that simple. Research shows that the number one frustration for candidates is lack of communication — and that’s been an indelible law of hiring since TA was primordial soup. It’s like a research report showing that water is wet. Yeah, tell us something we don’t know.
Of course, the problem has never been lack of understanding. There was just never anything organizations could do about it. But now they can.
AI eliminates the “comms problem” completely, giving candidates a 24/7, 365 touchpoint from the moment they want to apply all the way through Day 1 and beyond, essentially paving a two-way street right over the dreaded “blackhole” organizations have tried to work around for years. What Siri and Alexa have become for people looking to simplify everyday tasks like ordering groceries or setting the right ambiance at dinner parties, AI recruiting assistants like Paradox’s Olivia have become for candidates that are trying to get a job, fast.
The reason AI tools like Siri work is because they’re seamless. And the reason candidates have always hated traditional hiring processes (tons of forms and obscure information needed, third party application systems, tons of waiting) is because they’re not. But the magic of AI is that when it works, nobody thinks of it as AI anymore. You don’t have to go out of your way to use Siri — it’s just there whenever you need it, however you need to use it. There is no UI, it’s just there.
Candidates now have that for applying to jobs; AI makes it seamless. Easy. Invisible. And the best part about it is that AI can automate as much or as little of the process depending on what’s expected for different roles. Imagine a world where:
- A customer in a drive-thru line can scan a QR code at the window, apply for a cashier position, and get scheduled for an interview in two minutes before they even get their food.
- A truck driver sees a better position advertised on a billboard at a rest stop and is able to easily apply on their phone (truck drivers don’t even have desktops) at any hour of the day.
- A nurse comes across a social media ad for an open role after a long night shift and has the ability to learn more about the company values in real time, at 3 a.m., to see if making a switch is right for them.
- A package handler applicant goes from apply to hired to Day 1 within one literal day without ever needing to interact with a human.
- At the other end of the spectrum, a candidate applying for a more senior corporate position receives a high-touch experience that effortlessly blends AI automation at certain parts of the process with invaluable human connection.
- And all of this is done through texting on a phone or chatting on a computer.
The biggest impact AI has on candidates applying for jobs is that it no longer feels like they even are; UI is invisible, processes are streamlined, and frustration and waiting is almost completely eliminated.
The conversation is the application. And so much more.
What are the risks of using AI in recruiting and talent acquisition?
All of this is pretty new. And very early. If this were a race, we’d barely be leaving the starting blocks.
Is there some risk? Of course. New things always carry inherent risk. And being “first” is always scary. But is the payoff worth it? Well, it depends on the circumstances, but it usually is. Just ask Neil Armstrong.
Using AI in talent acquisition isn’t quite the same as landing on the moon, of course — but in many ways we are venturing into an entirely new frontier of recruiting and hiring. Things are moving at warp speed, and that’s scary. There have been some fairly high profile missteps, and that’s scary, too. But by and large, the positives overwhelmingly outweigh the negatives. AI works in TA, and the organizations who use it are seeing massive business success. In fact, the most risky approach right now is the status quo.
OK, but what about the risks? For starters, we’ve already seen some level of regulation at both a state and national level. New York City passed legislation in 2023 that made a clear delineation between hiring software that uses AI to make critical decisions and software that is just automating simple tasks. While there was a fair bit of consternation over this, we ultimately viewed it as a good thing. The law is fairly narrow in scope and only applies to “tools that almost completely replace human decision making processes” and subjugates them to third-party “bias audits” to ensure the AI technology they are using is “free of racist or sexist bias.” Everyone else (including Paradox) could carry on as usual. That’s to say: Continue leveraging AI to automate admin tasks like screening and interview scheduling, and leaving the judgment calls to humans.
By and large, the positives overwhelmingly outweigh the negatives. AI works in TA, and the organizations who use it are seeing massive business success. In fact, the most risky approach right now is the status quo.
That’s probably as good a segue as any to focus more on how we at Paradox personally mitigate risk in our products. It’s not particularly helpful to take a thousand foot view of such a nuanced topic, especially when perspectives vary wildly from company to company.
This is ours:
Improving inclusivity.
This is without a doubt one of the questions we get asked the most. How could this negatively impact us? Even though it sounds like a statement pulled directly from the Snake Oil Salesman’s Manual, it’s still quite true: it doesn’t. We’ve built a product that has no adverse impact. When you think about traditional hiring processes, they’re rife with steps and stages that could go awry. Static forms that are only in one language, third party apps accessible only on certain devices, recruiters and hiring managers only available at specific parts of the day, human subjectivity, etc.
Conversational AI eliminates all of that and creates a streamlined process that transcends all demographics. Candidates can apply on any device, in any language, at any time of day — and the AI assistant on the other end will be there to provide the same, consistent experience. AI reduces bias to practically zero through judgment-free decision making. And through that total neutrality, we’ve seen positive results when it comes to DEI; AI doesn’t care about your name, gender, race, or when and where you applied, it only cares about your qualifications.
More candidates of different backgrounds experiencing the same inclusive, unbiased hiring process can lead to more diverse candidate pools.
Ensuring data privacy.
There are lots of products today that are relatively thin wrappers on OpenAI or another generally available model. However, that also means that companies are sending candidate data places without their permission, and they can’t delete it or control what it’s used for; it could very well be used for training of other AI models and all sorts of other things that are likely to get your organization in hot water.
Our stance is to keep client data right in the same cloud-based infrastructure, with models that are rigorously tested to ensure the same enterprise privacy and security all our clients have also applies to our AI models.
When to not use AI in products.
There are some areas of the hiring process where AI may not be the right tool for the job, and believe assessments are one of them. There’s dozens of startups out there that are trying to assess candidates with AI in increasingly creative ways: text analysis, vocal biomarkers, game-based puzzles, facial recognition, and basically anything else you can think of. The tech is neat, sure.
But in basically all cases these end up creating more problems than they solve. Instead, we use AI to solve the “boring stuff” (admin work), which frees up time and allows people to engage with people to better inform critical decisions, rather than trying to automate the decisions themselves. So instead of fancy bells and whistles, our assessments are delivered in a simple but unique way – without any sort of AI involved — but grounded in the same science that has predicted hiring outcomes for more than 50 years.
What is the business case for AI in talent acquisition?
When it’s applied to the right things, it just works.
Of course it’s not that simple. But it also kind of is. The ROI of improving manager and recruiter productivity, candidate conversion, and candidate experience is loud. Try reading this next part without the cha-ching sound effect playing in your ears:
- $2 million save in recruiting costs in under a year
- 35,000 hours saved just by automating interview scheduling with AI
- 600% increase in candidate interviews
If you hear it, your CHRO — and your CFO — will hear it, too. There have been unbelievable business transformations through automating the simplest tasks with AI. It’s a business case that practically makes itself, but here’s one anyway:
Measuring success with AI.
Every organization has specific hiring challenges they’re trying to solve, which means “success” looks differently for everyone. But we focus most on what tends to be the root of all problems in TA: cost, time, and conversion rate. Not necessarily in that order, and not necessarily separately. Those three things actually tend to flow into each other in varying directions, and if you solve them you’ll usually find tertiary issues (like applicant flow or high dropoff rates) get fixed as well.
For example, we’ve seen large enterprise companies like General Motors save their recruiting team thousands of hours by automating interview scheduling with AI — which in turn led to reducing overall recruiting costs by millions of dollars. In the hourly world, we’ve seen organizations crying out for more candidates end up fixing their applicant flow “problem” simply by converting more of the candidates they already had — which also led to cost savings through a reduction in paid job spend.
Affording the right AI.
First of all, it’s important to think of new technology as an investment, and not just “more stuff.” Like any good investment, you’re only going to get out what you put in. Also — the reason you’re even shopping for AI technology is because whatever you’re doing isn’t working. At least not well enough. Don’t let sticker shock of the right AI tool scare you into settling on cheap band aid solutions. AI automation fixes the root problems of TA. Yes, there’s undoubtedly an upfront investment you have to make, but we’ve seen companies net millions in ROI within a year.
Why your CFO / CHRO / CTO will care.
Well, money. Let’s just be honest about that. This is good for business, pure and simple. But beyond that, infusing TA with AI leads to real, exponential organizational change. It saves money, sure — but it also helps make it, too. Better, more efficient hiring processes lead to fueling your organization with better talent, which leads to better work, and ultimately better products.
Companies that have already found success using AI in recruiting.
McDonald’s.
McDonald’s has over 36,500 restaurant locations worldwide. That’s a lot of Big Macs — and a lot of restaurants that need to be staffed. To take the hiring burden away from franchise managers McDonald’s and Paradox teamed up to launch McHire, a one-of-its-kind hiring platform that used conversational AI to automate the hiring process. Now, McDonald’s conversational AI assistant Olivia automates nearly the entire application process. With Olivia, McDonald’s cut their time from application to interview scheduled from three days to just three minutes, and reduced their total time to hire from 21 days to under three. Yes, that is in fact putting the fast in fast food.
Compass Group.
Compass Group is one of the largest employers in the world; they need to hire 120,000 new people every single year. A global contract foodservice leader, they were previously grappling with the challenges of high-volume hiring, a complex tech stack, and the need for more efficient candidate engagement. After they integrated conversational AI atop to work alongside SAP SuccessFactors, Compass was able to achieve their hiring goal of 120,000 employees … with just a recruiting team of 20. Twenty. That’s a recruiter-to-hire ratio of 1:5000.
Johnson Controls.
With more than 100,000 employees working across 150 countries, Johnson Controls’ success is largely dependent on their ability to scale and localize processes to individual regions. They use a conversational AI assistant named Emma that can speak over 18 languages to schedule interviews with candidates across the globe, taking away the need for recruiters to manually — and painstakingly — translate languages and time zones. With Emma, Johnson Controls has seen a 14% increase in hires, while nearly halving candidate response time. And candidates love it; Emma has a 98% candidate satisfaction rate.
What’s next for AI in recruiting and talent acquisition?
When things are going well, people like to use the term “up and to the right.” Well, currently the growth of AI is beyond that — it’s just up. You’re probably hanging on for dear life, wondering when things will slow down.
Look, things have changed. A lot. And with the rate we’re going, everything will change pretty soon. If you look at AI augmentation in other areas, the standard has been that after about a decade AI surpasses humans and is able to operate autonomously. Think about Chess. Gary Kasparov fought Deep Blue tooth and nail back in 1996, but by 2006 AI was unbeatable. We were still using physical, comically large maps to take family road trips as recently as the mid 2000s, but most of Gen Z thinks an “atlas” is that big bald guy with the earth on his back. Things change, usually gradually and then all at once. The goal for you — for all of us — is to not be at square one when that switch flips. There is no path forward to the status quo. Things are going to progress with or without you, so the only question is do you want to go along for the ride or get left behind?
The future of recruiting and TA work will likely look like this:
- People who serve as “systems operators”, managing the AI tools and overseeing their performance.
- People who do “people stuff” — interviewing, connecting, building trust, recruiting
Here’s the good news: There will always, always, be a place for humans in talent acquisition. This whole “people thing” simply doesn’t work without people on both sides of the fence. But as AI continues to improve and is able to do certain recruiting and hiring tasks better than people ever could, it’s imperative that humans continue to get better at the human stuff — because that essentially will be the only recruiting work that’s left.
Things change, usually gradually and then all at once. The goal for you — for all of us — is to not be at square one when that switch flips. There is no path forward to the status quo.
While I’m sure that sounds scary, it’s also amazing. Because I’m guessing you didn’t get into this industry because you enjoyed clicking buttons or sending emails. No, you’re a people person. We all are. You know that the magic moments happen when people get to spend time with people.
And if I had to distill 8,000 words into just a sentence, it would be this: AI in recruiting and TA will simply help us all do more of what we love.
But hopefully you still read everything else. Cheers.
