Depending on who you listen to, AI is either going to free us to do what we want. Or end civilisation. Whatever the long-term impact, AI is finding its way into so much of what we do.
But how is AI used in recruitment? And does it help or harm our diversity efforts?
There’s quite a lot to unpack here - so I’ll attempt to break it into manageable chunks over the next few newsletters. I’ll start here with what algorithms, machine learning and AI are. Then, I’ll look into the impacts of AI on diversity. And finally, how it’s used in recruitment tools.
Let me know if there’s a specific product or ATS feature you’d like me to cover. I can’t promise I’ll be able to find all the answers, but I’ll give it a go!
Also in this issue, we’ll look at copywriting - the art of writing to sell. We’ll cover tips and references to write copy to get more people reading your job ads and submitting an application.
P.S. If you have some tips on how I can improve this newsletter I’d love to hear them - just reply to this email.
Often when we hear that something uses an algorithm, we think it must be complicated. This is usually the intention. To make us believe that something is difficult and we need the algorithm to solve it. “You do this, and then the algorithm takes care of the rest.”
It’s true some algorithms are complex, but we use simple ones every day. An algorithm is just a set of instructions that describes how to get something done. Like a recipe. Or giving someone directions.
Software algorithms are no different. A programmer has a specific outcome in mind and codes the steps to get there.
A common use of algorithms in software is to sort data. Say you have a random set of files. What’s the quickest way to put them in alphabetical order? There are many ways you could do this. And as this video from Ted demonstrates, some are much quicker than others.
However, for some problems, the only way to get the answer is to try every single option - a brute force algorithm.
Imagine you were a courier. Each day you’d want to find the quickest route to make your deliveries. But how would you work it out?
First off you’d need to know how long it took to travel between each drop-off. Then you could find the total time for each possible route. It may not be the easiest task with a pen, paper and a map, but surely something a computer can handle.
Well yes and no.
It starts off easy. Let’s say you begin and end your shift at the depot. If you had 3 deliveries, there are only 6 possible routes. When that increases to 5 deliveries, there are 120 options. At 12 deliveries there are over 479 million routes.
And incredibly over 1,307 billion different ways to deliver 15 packages.
Even with a high powered computer, it takes weeks and years to calculate complex routes. It’s nowhere near quick enough to work out your journey each day.
The only option to get a result quicker is to sacrifice accuracy. If you get a good route in 5 minutes, you could finish all your deliveries before you even calculate the best option.
But we need to be aware of the trade-offs. Sometimes the loss of accuracy will be fine. Others times it may impact what we can do with the result.
This example is the travelling salesman problem - a good overview of which you can find here (at 10:52).
And this video shows some of the solutions, and how much they lose in accuracy.
In the section above, we looked at traditional algorithms. Meaning that someone decided the steps needed to solve a problem. They wrote them down. Checked they worked and probably asked others to test them as well. Then they updated them until they got the desired result.
This process is true for writing a recipe or writing software.
However, with machine learning, a computer writes the algorithm. It starts by guessing the answer to the problem. On finding out if it was correct, it learns and refines its code. Over time, it learns from this feedback, and its guesses improve.
To understand why this can be better it might help to think about sending a message on your phone. As you type a word, the app will try and guess or correct it.
For someone to write an algorithm for this feature they’d need to:
With machine learning, the computer essentially follows the same steps. But it can review very large sets of data and make changes very quickly. Therefore it can improve its results faster than a human could.
Despite this promise, there are a few reasons why it’s not the best approach for all situations.
Firstly, the machine needs a lot of training data to learn. Which someone needs to prepare so that the computer knows how accurate its guesses are in order to learn and improve.
For example, developing an autonomous car needs huge amounts of data. In fact it needs so much that Google has put us all to work on it. Each time you see a captcha image and are asked to pick the squares with buses, street signs or pedestrians, you not only prove you’re human, but you also prepare training data for Google.
Another limitation is that machine learning algorithms are like a black box. We know what goes in and what comes out, but we have no idea how the computer gets to the answers. While we don’t care how Netflix chose to recommend a show to us, we may need to explain why we selected one person over another.
It’s also important to note that people are still very much involved in machine learning.
Although the machine writes the algorithm, it does so in a controlled environment. A person, or team, decide how the machine should learn. They choose what data to use for training. They prepare the training data. They design the environment and set the boundaries for the program.
All of which can influence the algorithm and its results.
For a visual explanation of machine learning, I’d recommend this video. It’s only a couple of minutes and provides a good overview.
And if you’d like to learn more, this is a good article that covers the key concepts.
I’ve included machine learning and AI because I wasn’t too sure of the difference. Some times the two terms are used interchangeably. Other times as separate things, which is confusing.
Since we know about machine learning, what is AI? According to Andrew Moore, the computer science Dean at Carnegie Mellon University:
“Artificial intelligence is the science and engineering of making computers behave in ways that, until recently, we thought required human intelligence.”
AI is a very broad topic. It describes more an end state rather than how we get there. That’s where machine learning comes in - it is one of the ways to achieve artificial intelligence. This article provides an excellent explanation.
And if you’re interested in where AI is at the moment, the AlphaGo movie is well worth a watch. You should be able to find it on Netflix (in the UK at least).
In the last issue we looked at what to include in your job descriptions to appeal to the best candidates (available here). So I thought it would be helpful to also look at how to write a great ad.
A pretty comprehensive guide to copywriting. Broken into seven chapters, it provides a mix of skills, such as writing customer-focused and compelling copy, and formulas. It also has a lot of great practical tips and plenty of examples. Not all of it is relevant to recruitment, but a good resource.
A helpful introduction to writing effectively. A short read, it covers the key things to consider - writing for the reader, formatting and the call-to-action.
This article has 7 tips for writing copy that gets the reader to take action. Central to the approach is understanding what your customer wants. What they’re looking for. What appeals to them. Which is pretty much what you need to write job descriptions that appeal to candidates.
What better way to stand out than to focus on what you can offer that no-one else can. This article looks at how something you take for granted can be special. It’s written from a business perspective but there’ll be something about the job, your company or culture that will be a USP for candidates.
A fantastic copywriting guide. It’s clear, comprehensive, covers all the basics and provides many examples. Some of the sections are a little dated, and others aren’t relevant. But the sections on the fundamentals of copywriting more than make up for it.
This podcast is a little off topic, in that it is a conversation between a marketer and a product manager. But I’ve included it as it has a lot of tips on how to improve your storytelling and copywriting. You’ll just need to tweak some of them for recruitment and job ads.
And if you’re looking for more resources to keep learning, you might want to check out:
What you’ve been up to - from the community
Theo shares how he implemented a new recruitment strategy at NICE. With more and more specialist roles, he was finding it difficult to identify and attract the right talent. So he turned to evidence based recruitment. Learn more about his approach and the results.
Do you have something you’d like to share? Get in touch by replying to this email.
That’s it for the moment - see you again in 2 weeks. And if you know someone that would like this newsletter, please share this link.