AI in Recruitment: Transforming Public Sector Hiring with Andy Headworth

Yeah. So I generally think it's going to be, how do we get our workforce to accept, learn and adopt a tool that's going to potentially, for some, change their jobs, change their roles, you. And I'd say in a positive way, but some people look at it as a negative, and I think that's going to be the biggest challenge for many organizations in recruiting, or whatever it may be.

Hi and welcome to the show. I'm Johnny Campbell, CEO and co founder of Social Talent. And this week on the podcast, we're speaking with Andy Hedworth from H or C about how the public sector is, to be quite frank, kicking the ass off the private sector when it comes to AI and recruiting usage. Andy, you're very, very, very welcome to the show. Andy, we've known each other a long, long time. You've been on the show a couple of years ago, but for our audience, would you mind introducing yourself? Explain a little about your career, your background and maybe about who HMRC are if you're not in the uk?

Yeah, absolutely. Hi. So currently I'm the Director of Talent for hmrc, Director of talent acquisition, director of Recruiting, all rolled into one. My colleague will get annoyed when I say director of Talent, but I'll leave that to her. My background is fully recruitment and Jonny and I go back a long, long way, so I've been in recruitment, starting as a recruiter way through, and then that. That came into the first recruitment revolution with the Internet and then that morphed into social recruiting when the social came along. And we were both very early adopters in the social space. So very much driven by innovation and how new technology and new ways of working and new thought processes can improve how we do, how we do recruitment and how we recruit for others and how we help others recruit. I was consulting for about 10 years, working with organizations around the world, and then took the OR got the call, I should say, to join the public sector and join the UK civil service. Not something I'd ever considered. And anybody who knew me would all said the same thing. Why have you done that? Well, I'm still here eight years later, and I have to say, I absolutely love it. On the other side of the fence now, from my point of view, AI the journey and where that's come about, we all know when you know, when OpenAI landed and ChatGPT landed a couple of years ago now, and going back to the social days, my interest has always been whenever I see a change in technology that's a scale of social or AI, my brain Immediately flips and I know Johnny's does as well to how can we use this for recruitment? How can we use this? What can we do with it? How can we source people, how can we engage people? And so very quickly started playing with it after Christmas that year 23 I think it was, and started to bring it in. So straight away it's how we can look at it from a point of view of using it in all aspects of recruitment, which I'm sure we'll touch on in a minute. In terms of HMRC and who we are for our international audiences, we are the UK's tax authority. It's as simple as that. We collect, manage and own the tax for the uk. So we have a customer base which is every working person in the UK and some abroad. And our role in terms of what we do in my function is to recruit people across the whole cross section. So we have about 65,000 employees across the UK and all the major cities and we recruit from call center people right through to senior leadership and it goes across 25, 26 professions across the whole organization. So people think we're taxed, but we do everything underneath. So we recruit a very wide range of skills into the organization.

Andy, thank you for that intro. I think you've framed it nicely, given us a good understanding of the type of organization you're in, type of background. You know, AI stories probably aren't expected from the public sector civil service. I know when I talk to looking to the tech companies, to the leading edge companies and many of them still haven't done anything. I was at a conference last week and most people in the room, 50 different leaders of TA, I'd say five of them have done anything with AI, the other 45 are thinking about it, talking about it, asking for permission, etc. Etc. So I think it's a nice surprise certainly for me, for others to see that a public sector body like yourselves are trailblazing, if you don't mind me saying so, in this area and we'll leave that to the audience to decide at the end. But if I can get to the meat, you've kind of outlined several different opportunities, some of which you've already executed on, some of which you are currently about to execute on, and some you're planning for the future. I wonder, could you walk me through the first two which you have already launched, already got up and running, what they were set up to achieve, what your experience was with the team, getting them up and running and what you're seeing from them so far. And then we might jump into one of the bigger projects, your third one, which is not just is ready to launch, but not just quite there yet.

Yeah, the first thing we did was so for all of you, and I'm the same, by the way, in every conference that I've been to, have the same numbers in terms of who are doing it, who aren't. So the first thing we did was to start to get my team involved, to get them interested in it. So what I would say to everyone is you have to get your team interested in using AI. If they're not, then you've got a hiding to nothing there to start with. So we, and bear in mind this is summer of last year, so it really was quite early. So we put a team together. We have an open, we have an hour a week, every week where we just basically came together and lunch and learn, bring it to the table, bring the new tools, let's talk about it, let's explore, have some fun, create pictures, create videos, create radio jingles. But it got them prompting, it got them to understand and it got them to understand how you could use AI and how that could affect the different aspects of recruitment. So first of all got them playing with it, that was really important. And we got the buy in from the team. The second thing is we then chunked up our recruitment and we said, right, what if we break it up into simple blocks? How could we then use AI for each block? So the sourcing block was the easy one, quite frankly. And we, you know, we use that for sourcing in very many different ways. And you know, you've done a fantastic post on that a few weeks ago, Jonny, and how we use it for sourcing. And we were doing the same thing then. It's like the next big problem we had was the hiring managers. How can we help hiring manager get an accurate job description, a current job description, one that's got all the relevant hashtags and it's got all the behaviors, the competencies, everything else that you need as a hiring manager. How can we help them? The next stage is the assessment and selection process. I'll come back to that bit later on. And another piece that was really big for our team at the time last year was Talent Insights is trying to understand if we're based in every major city, we want to know everything that goes on within an hour of that city. And that's a conference around there. Everything from schools, colleges, universities, who's recruiting, who's employing, who's hiring, who's firing the social mobility ONS data, census data, all that great stuff. So I need to know everything about my audience within an hour of that location. So we focused on that tool and we focused on the hiring manager tool because they're the ones that were causing us the most pain at that time. So the hiring manager one is called skillscribe and it's a simple concept. We want to make our hiring manager's job so easy they don't even have to think about it. And one of the challenges we all know in recruitment, and it's been going on for years, is your hiring manager has the job description in his bottom left drawer, his or their, his or her drawer. And they always get the same one out every time to advertise. It's probably out of date, it's probably, probably illegal in terms of HR methodologies and it just doesn't deliver. But they'll just keep using it anyway. So we wanted to stop that. We wanted to keep consistency. We wanted to make sure that we integrate the evp, the brand. We want to make sure that we give them the tools to then take that job and take them to the social place they work, whether it's Twitter, Facebook, whether it's, you know, obviously LinkedIn and allow them then to convert that and take it to a social space. So in very simple terms, what the tool does is and they can do self serve or the recruiter can work with them. They put a job title in and it then pops up and says, I have a grade. So you then put the grade and it's obviously, you know, written towards our grade structure. And behind the scenes, the large language model is built off all our old online job descriptions that we've used. So that gives tone of voice. We've got the evp, we've got a brand, we've put our career site in there, we've got all the civil service guidelines, all the, we use success profiles. So all the behaviors and trying there's anything we've missed and we put all of this into the large language model. So then when it comes back to the beginning, so you put your job title in grade and then it comes out with a job advert, a large job advertising job description, social adverts. So it produces an advert for that you can post on your LinkedIn profile, an advert you can put on your Twitter profile. It then comes down and says, right, these are behaviors we suggest based on success profiles, based on the job description you're looking for. We think you should be looking to go after these six to eight behaviors. It gives you the reasons why and advises you which ones you select. It then does the hashtags for that role and then it starts to produce the interview questions based on that role, based on the behaviors, which interview questions you should be asking. It then does a secondary set of interview questions if you want as follow up ones to qualify the first one. And then produces hashtags at the bottom. And then it produces your outreach messaging for you to engage with people on Twitter or LinkedIn or Facebook, wherever you want to engage with them. It does this in about 90 seconds. One golden thread right the way through, then takes you into a Google Doc, effectively a Google Doc and allows you then to edit any aspect of that in the Google Doc using AI of course. So you can say, oh, can you just improve the job description, can you improve the adverts or can you change this? And you use the AI exactly the same way and it changes every aspect. And then once you've gotten you're happy with it, you just download it as a PDF and you have your consistent golden thread product for a hiring manager there with current thinking merged with our own tone of voice merged with our own evp. It's all publicly available data, so there's no secrecy involved. And suddenly the hiring manager's got everything there and it's proving very, very popular. It takes about two and a half, three minutes to do that. Once you've got the questions answered as well, give it five if they're a bit slow and suddenly you've got the document there. And the concept for us is what we want to try and do is just focus on focusing them on thinking about the interview and allowing you to get the best quality people coming through the adverts. Now we have currently got tell you the exact number at the moment, how many people we've got currently using it. We've got a notice the other day, we have 168 of our hiring managers currently using it at the moment in the business. So we are still, we're still appraising the full set of data anecdotally so far. It's much quicker, it's much more efficient, it's modern is one word we've had described. And the other thing as well, it allows that we've got better quality people applying for the jobs than they've seen before because they had decent adverts that we're placing. And that's, it's been a really interesting product. So that's the first one. Johnny anybody?

Sounds like it probably took six months, 200 grand, team of 10 people.

One, one, one person, two weeks to build the pilot. We're iterating it continuously, but I would say probably one person with a person qualifying it in about four to five weeks. And in terms of cost, we use a product called Mind Studio, which is a freely available, brilliant piece of kit. We pay $20 a month for it each. There's three of us that pay for two of us that pay for that. That's sounds so simple, doesn't it? It's just a series of. It's a clever series. Well, it's not simple. It's a clever series of prompts built on a bespoke large language model to you.

There's nothing confidential there. You've uploaded your guidelines which are publicly available. It's producing a job spec. It's not any personally identifiable private information. It's something that I think anyone in the world could set something up like that. And it's phenomenal whether it's used by the recruiters, as you said, or it's self served by the hiring managers. Even better, the value is enormous.

Yeah, absolutely. And this is why, I guess Wreckfest was a good example. When we were at Wreckfest this summer and talking to all the recruiters, the suppliers, everybody that was there. I don't know you, Jonny, when you just come back from the States, they're all talking about things as if, yeah, we can do this, we can do that. I said, well, you can do it now. You could have done it six months ago. We did this last September. So don't stand there and say to me, we can't do it. We've got to put through R and D. We've got, as you say, get an R and D budget, a couple of hundred grand, go and bring the software vendors in, go and pick the big five in. You don't need to do that. This is available now. It's just having the thought process and a bit of logic and mentality of what have we got to put in that bucket underneath? What things do we need? Can we add to that bucket of information that we can then put AI above it and then prompt away.

Tell us about step three. A little bit more complex potentially, though much more impactful in terms of efficiency across the team.

Which one do you mean? Do you mean the step two or three? Which one you want about. Oh, three. The Insights one. So the insights one was born out of a frustration from mine to know exactly what's going on around a recruiting location. It just comes back to being A recruiter in my head, I want to know everything that's going on around the location. So same principle, same product. We use Mime Studio and this took one person two weeks to build. And then we took, we looked at every single available piece of data around that location. So we started with ONS data, we had census data, all current. We used LinkedIn insights data. So we would take the data and scrape it for that location and add it into the large language model. And just for awareness, LinkedIn were available, were aware of this and we're quite happy with this doing that. So there's no issue with that at all. We then added in things like local information, salary surveys, council information. We added in everything we could possibly do that situated around a location. So we have suddenly this data set for, let's just say Manchester. Then you have effectively an interface where again same interface and all you have to do is put in the location. So for us we've got, you know, we've got 20 locations. So you pick drop down box of one of 20, put Manchester in and it will produce a one pager to start with of the current state of affairs of recruitment and talent acquisition in and around the Manchester area. So bring into whether anyone's hiring, firing, the salaries, the benchmark. And it also always gives a recommendation at the bottom of the. If you're currently recruiting in Manchester, we recommend that you adopt this approach and this style based on your EVP and your brand. So it gives us a little steer of what's going on at the time of year. And then if you can imagine a chatgpt on a location, it's basically that we have the ability to drill down and ask that location anything at all. So you might want to look at, you know, for example, what's the, you know, I'm looking to recruit data analysts. What's the, what are the numbers in that location, what the salaries, who's hiring, who's firing, social mobility, looking at the universities that are doing courses where you might be able to plug in if you're looking for data analysts. So it really takes it down and then produces this information where you can drill down on anything you want to. Now we built it with the intention of recruitment and unintended consequences which makes this really interesting product is that our locations and our places for growth, which is what it's called in government, which is where the government are trying to move the focus of London out into the regions. It's been going on for a few years now and we demoed this product across the whole Civil Service across Cabinet Office, this team have picked this up and gone, do you know what? This is amazing for us. It allows us to interrogate locations, to know where to go to, not to go to, how to position ourselves in a marketplace. So they've started now using this product, our own locations team, which deals with so much more than recruitment. It looks at everything from the buildings, locations, everything around a particular location. They've now embraced it and have been using it now for well over six months as a tool to help them do their planning, their logistics and everything else. So suddenly, a recruitment tool that we built for recruitment has got many uses. So, again, fantastic piece of simplicity in theory. Obviously, it's a bit more complicated than that. And both products, we demoed these before Christmas last year with the whole. With Cabinet Office, and we've done it to all the directors across the main departments. And these will both be taken up across the wider Civil Service and utilized by different departments over the course of the next three to six months, once we've done our, you know, our formal. Our formal testing and reporting, which we have to do. So something from a little acorn is going to grow into a tree and spread across a wider organization, which, for the two people that have built them primarily, you know, what a fantastic story for them, you know, what an amazing story.

It's amazing. I think, again, super practical examples of how anyone could use these. And you mentioned at the start, obviously you mapped this originally by looking at the stages of recruiting. So at this point, we've got the insights, we got a plan, we've got job specs, we're advertising. What about when you get to the point when you have candidates applying, Andy, how did you think about that and what are you working on?

So this was a challenge for us because we thought we could crack this one out really simply. And we thought, why not? We use the same tool, same process. So we scoped this out and spent, tried to solve it. And for us, we call it sifting. In the Civil Service, it's the selection process. So this is after people have applied and it's the process. The ATS we have doesn't do a great job of this. So it's, how do we take 1,000 applications, 500 applications, what it is to get it down to an interview or number of people, and that's done manually across the Civil Service and it takes a considerable amount of time. So this is like, for us, this is with the nirvana, this is the one to really start to make some monetary, you know, inroads into monetary savings that you can only just dream about. And so we thought I would crack this out. It's easy enough. How hard could this be? So we did and the team had a good go at building this and we thought we cracked it. But we just could not get the AI to sift each individual CV uniquely and then stack rank them at the end. That was just a problem we had. So we went sourcing. We went searching for a company which was a bit of a problem, trying to find organizations that were able to provide that service. We found one eventually. I came across them at the CIPD conference this year in the summer and we're currently working with them and we've tested it. It works. And just to give you an idea of efficiency, we had a. We tested it on a. On a roll that's been closed so it's all finished. And we know, we know we knew exactly the ranking. We had 200 plus candidates for this particular role. We know it took I think it was eight panels of three people, a full day, a full day to go through and select the people from that. Probably a bit longer in honesty so you can work the maths out on that one. And then we put this through this particular tool. Nine minutes. That was it, nine minutes. Now the criteria here is we're not looking to, we're not looking to recruit and select using AI. That's the most important part here. We are looking to deselect 80, 85% of the applicants that are not suitable for the role and leave the 1015 depends on the numbers 10, 15% left for the hiring manager and his team to make the decision on who they want to select and go ahead and interview. So be really clear for everyone. AI is not making the decision here. This is just a deep selection tool. But the amount of time as you can, you know, you can imagine how much time and money that that is going to save by taking those people out of the process and just allowing them to focus on a much smaller number of people. And that's where we're currently at. At the moment we're just working that.

With a supplier is the plan to also automatically disposition as in the messaging and personalized messaging for the dispositioning of those 85% of candidates that don't make.

It through 100% they can have. This is the cool thing about this particular product because it assesses a CV against the job spec individually it will provide full briefing on why that person has not got the job and recommendations on what they can do to get a job in the future, they can either have a couple of lines, they can have a paragraph, they can even have a whole page. And that can be inputted back into the ATS and fed out back to the candidates, which is something that, as we all know, candidates don't get a lot of feedback even for rejection. So the ability to do that will also be significantly improved the candidate experience. And also the other problem is fast as well. We don't have to wait three or four or five weeks for them to find out. They get to know about that as quickly as we can turn it around. So as opposed to some of the ats that just do, you know, keyword rejection and automated rejection, this is proper insights into why someone didn't score well for that role. Based on the cv, obviously that's the data the AI is looking at at the time. So, yeah, super excited about this because this is, this is something that, I mean, this is where the industry is going to go and, you know, we'll all be there in three or four years time. This part of the process will be, you know, will be automated, I'm sure. But for the moment, while we're on the cutting edge of this, this is just. And no one else is doing it. You know, no one else is looking at this aspect. I'm sure they are, but no one else is coming out and, and talking about what they're doing and how they're doing it, which, which I find quite strange in our industry, Johnny, because they're usually quite vocal, aren't they, with things like this?

They're getting there. I heard two great stories from 711 in the United States and Compass Group, which is Sodexo's biggest competitor globally, and how they're doing something similar with Paradox and their AI software. So I think we're beginning to see the use cases and we'll hear more of them, hopefully more in this podcast over the next few weeks as well. But Andy, what's next for you? You know, that's. You've got a huge way into the journey of, of the hire here at the moment or the recruiting. What do you think is next up after this goes live?

I think we're always going to be iterating and some of the. So another big space is onboarding, but that's the place we're going to go next and how we now there are some fantastic onboarding tools. You know, we know that. So how can we make that experience a good experience? Sometimes, certainly in the public sector, they get, we get accused of not giving good experience. I think for us, the big step forward for us and for any organization is going to be adoption is the people, you know, people talking about copilot and bringing in copilot testing, for example. And we are, you know, we're testing copilot like a lot of many organizations, but it's, it's the individuals working for us, whether it be in cpo, in the business, or whoever's going to be exposed to the ability to improve their working day by using AI and getting them to adopt it in an organization. And I think that's going to be the biggest challenge for many organizations. It's not the fact that we can roll AI out. I mean, my recruiters use it, but not all of them use it because some of them still have an aversion to wanting to use AI in the workplace. Yeah. So I generally think it's going to be how do we get our workforce to accept, learn and adopt a tool that's going to potentially for some, change their jobs, change their roles. You. And I'd say in a positive way, but some people look at it as a negative. And I think that's going to be the biggest challenge for many organizations in recruiting or whatever it may be.

I think if the TA team don't embrace it, the organization's going to force it regardless, and we don't want that. And Andy, we were here, as you said in social media, took years weirdly for people to, to trust that there was something here and that they should be using it. And we the same issues over people banning usage, saying it's going to be the end of things, it won't work. And look where we are today.

I know. And what's next? I mean, it's a hi today. What's next? I don't know. I'm looking at a lego kit of C3PO on my table here. I just wonder one day, you never know.

Andy, thank you so much for sharing those insights and sharing your story. If anyone wants to reach out to you, any leaders who want to reach out and get a bit more detail because they're about to embark on this or you've inspire them to embark on something different. How, how can they best get in touch with you, Andy? And is it okay to get in touch with you?

Of course it is. And the best thing is just to jump on LinkedIn. Andy Hedworth. Yeah, you can find me easily enough on there and just reach out and I'll, you know, take it from there. So no problem at all, Andy.

It's been a pleasure having you on the show again. I really appreciate you being as honest and candid about what you're doing and I love what your team is achieving. And hopefully we'll catch up with you beginning of next year, maybe, to see how that went, what's next and what else you're doing for. For 50 quid in two weeks of effort.

Yeah. Cheers, Tony, thanks very much and thank.

You for listening, listener. We'll be back again next week with another great show.