AI in Recruitment: Automating with Impact at ThoughtWorks

Even though AI can demonstrate sort of quasi empathy, you know, it can be seen to be empathetic. It's not genuine empathy. And sometimes you need that human interaction to really get behind, you know, the, the real reasons that a candidate is looking, or not looking to leave their job or why they're looking at your job versus another one. You know, the best candidate tends to be interviewing simultaneously in multiple places. And that's where recruiters, I think, have to, have to still. They can leverage AI, but I think they still have to demonstrate the personal skills to make the difference.

Hello. You're very welcome to this special edition of the podcast. We're coming live from Barcelona, from Villarosa, actually outside Siches in Barcelona, where we bring together a bunch of fantastic TA leaders to discuss the hot topics of today. And joining me today, I'm delighted to have Marcus Thorpe from ThoughtWorks. We're going to be digging into AI and automation and discussing, you know, how we're automating TA and recruiting in what ways. Mark is going to share some stories around what ThoughtWorks are doing and what he believes are some of the kind of key opportunities for automation and maybe some of the risks in that as well. But Marcus, perhaps before we dig into it, I'd love you to introduce yourself to our audience, maybe tell them a little about your career, what brought you not just to here in situ today, but maybe to where you are in your career. And maybe tell us about your passports, which you seem to have many of, which is an international man of mystery.

Thank you. Yes, I'm Marcus Thorp. I'm currently global head of TA for ThoughtWorks. I've spent over 20 years, I think officially in TA. You stop counting at 20 years, but once you're over 20 years, so it's, it's definitely well over 20 years in, in, in talent acquisition, mostly in tech. I did one little stint in the Middle east working for a think tank that did economic and political research for the UAE government, which is another story altogether. All the rest has been in tech. So I worked at, started working agency, as most of us did on the dark side, as I like to call it, and then moved over in house, living in, in the Bay Area in San Francisco and took on a couple of contracts and then realized that in house was absolutely where I wanted to be. I've worked for some big international tech product firms like Google and Amazon and Skype and Twitter and yeah, most Recently I joined eight years ago. Now, ThoughtWorks were a big technology consultancy, so we build bespoke software for big Fortune 500 type companies. So basically big companies that have huge technical needs but won't have, you know, teams and teams of software engineers to help deliver them. Which basically means everything from starting a project through the completion of all the code and now the maintenance as well. We sort of have a managed services arm as well, but delving more and more into data science, data engineering, and we now even have a chief AI officer as our sort of the clients move more and more towards broadening their horizons from can you build us a website? To can you own the infrastructure to do complete supply chain management for us, that sort of thing.

What's the kind of scale of the organization? How many employees, how many people are you hiring in a given year? What's the size of your team, what's the scope in terms of geographies you work over?

Yeah, we're in five regions, five continents, which is North America, Latin America, Europe, IME, India, Middle east and APAC. And we're just under 12,000 people. We've had a couple of quieter years. We went from being a, a public company, well, from a private company to a public company. And now we're back to being a private company again, which is a story for another day. But we in a good year, and this will be a big year, it'll be north of 4,000 hires a year. Our record's about 5,000. This year we're on track for 4,000 and next year we're expecting it to be another 50% on top of that.

Wow. So kind of your team then. How big are they? Where are they based, how are they structured?

The team is about 230 or so now. We were at 300 at one point. At our biggest we were able to retain most of the team. Even when things went really quiet in 2023, they're all over. I have my. The structure of my team is I have a global head of sourcing and strategic hiring, which is exec hiring essentially there are heads of TA for each of the regions. I have someone who runs the program side of the thing of the fence and otherwise there's a recruitment systems team that supports us from an operations perspective. And then I also have the data analytics team reporting into me as well, which is super helpful because that level partnership makes all of those stakeholder meetings so much easier to facilitate. You get the right data, you can have the right conversations, the right narrative, you can steer the conversation the right way, etc.

From a business value perspective, literally what you are doing is you're producing the product for the business, like you are the sole resource, as in your team, you bring talent into the organization. The talent then gets essentially contracted out to customers. So the focus on TA must be pretty intense in terms of if you miss deadlines, you don't hit your numbers. It has a very direct effect, I imagine, on revenue and profitability and ability to the job. Talk to me about that and that, that structure. And how is that different perhaps to other recruiting teams?

Yeah, I mean, if revenue is flat, we can just move people around from project to project and rotate them in and out of projects as soon as revenue goes up. And historically, almost to the moment that we went to IPO, we were growing by 20% year on year on a very, very consistent basis. And so that 20% of growth every year is entirely reliant on us bringing in new talents. Obviously, there are backfills as well to facilitate, but that all of that new revenue growth is us finding the right people in the right place at the right time. That's changed over time. Historically it was all development, and then the sort of the, the QAs and BAS that sort of surrounded it. Now we're moving away a little bit from that, but so we have to sort of morph what we're looking for and how we're looking for it. But it's all about predictable supply. You know, know, we built a really good, credible partnership with the business. We're seen as a trusted business partner when it comes to talent. We actually report into the chief operating officer versus sort of an HR function. HR function is a parallel, which is super helpful in terms of sort of direct access to the, to the board, et cetera, when we need it. But we, we've had to build this sort of predictable model that we're able to say with a certain level of expectation that someone new coming into the organization takes n number of days to ramp. As a recruiter, they'll be delivering hires in their second month, they'll be fully ramped in their third month, and they'll be producing, you know, in excess of three hires every month, which isn't a huge number. But making sure we do that month in, month out across the team to make sure that everybody is in the right place at the right time is critical. And we're constantly playing a game of we will plan at the start of a year, as everybody does, but typically by the sort of second week in January, we've ripped that up and thrown out the window as demand has gone up and we're having to hire a Whole new set of a whole new team of people in a location that we weren't planning to do. So the vast majority of our hiring right now is in what we call delivery regions. So we have Latin America, Brazil, Ecuador, Chile, India, parts of Asia like Vietnam and Thailand. They're sort of our delivery regions. We have Spain and Romania in Europe. And then the other, the other countries are sort of the sales regions. That's where the, that's where the big Fortune 500 companies will be based. But thanks to Covid, we've been telling clients for a long time that we could facilitate their work by distributing teams all over the world at a fraction of the cost. And they sort of were insistent that we'd have people, you know, team of 40 flying in all over the US to go to Cincinnati or Austin, Texas. And now they're sort of saying, oh, thanks for telling us we can actually do this in a remote basis. So most of the teams are being built up in these new locations and that's where most of my growth is. But as you say, we are 90% reliant on TA to deliver any growth in revenue, which we expect to be somewhere between 10 and 20% every year.

You mentioned you're in the business eight years, so you've gone through some crazy changes in terms of the talent market. So in 2025, how would you describe the market for the talent that you're looking for? Obviously high tech, highly skilled engineering, software engineering talent. You know, is it a candidate's market? Is it an employer's market? How would you describe it or compare it to other years?

I think 2020 to 2021, 2022 were the, the absolute insane years in terms of, you know, it was completely candidate driven, it was a big resignation, etc. Now the, a lot of people are a little bit more entrenched, a little bit harder to get out of a job, but there's still plenty of talent looking if you ask them the right way and search hard enough. I think intelligent sourcing, developing sourcing in a new way to make sure the, the, the messaging is individually crafted rather than mass sent is critical. It's about, you know, we target 70% of our hires to come from either sourced or referral. We're trying to remove the reliance historically that we had on inbound candidates. We're still post roles, of course, but the quality, the throughput is nowhere near as it is, as good as it is for source candidates, for example. So all the reliance is on that. So it's constantly about evolving the sourcing team making them do a better job. Our response rates are really good. We have a good brand in most of the locations. Whenever we go somewhere new, like we're sort of expanding into Latam in new countries now we have to build that presence up. But it's about making sure the message is the right one and then that the candidate experience is exemplary. It has to be. You know, it's completely pointless being able to sell from a source of perspective, recruiter perspective, if that's not authentic and sort of seen throughout the interview process. So everybody has to be talking from the same hymn book. So now it's sort of. Yeah, they say it's, it's, it's not easy right now, but it's a lot easier than it was in 21, 22 people are hiring. The fact that we have clients coming to us for more work means that they're looking themselves and sometimes we're actually competing with the companies that we're, that are our clients for that same talent. And we have to work out who's going to, who's going to employ them and how we're going to have them work on, on a client site. There's a lot of work which is sort of staff augmentation, which we historically haven't dealt with very much, but in order to get some of the revenue back, we were looking at staff augmentation work to a certain extent, which allows us to get a foothold in a business, but the margins are a little bit lower and therefore a little bit more complicated to find the right talent at the right price. In certain parts of the world, like India, constantly salaries go up by 10% a year and we have to try and stay ahead of it. It's a constant battle to make sure that our ranges are competitive enough to be able to attract the talent, you know, because my biggest resource, most precious resource that we have is interviewer time. We're a professional services company, so we're constantly reminding our, our consultants that they work for ThoughtWorks, but, but typically they use client email systems, client client laptops, et cetera, while working on client sites. And they sometimes forget to check their ThoughtWorks systems, which means we're, you know, having to find novel ways to remind them about the interview they're about to facilitate or hopefully facilitate. So that's a constant battle. So I can't, we can't possibly waste any interviewer time. We have to maximize every interview. So that's why we focus so much on the most effective ways, most efficient ways of making hires like sourcing and referrals, it stops burning through those interviews at a rate we can't afford to do.

You mentioned interviewer time and maybe we can start this when we want to talk about AI and automation, Often people want to go straight into the conversation of what are the tools that you think are interesting and sexy stuff, which I think is the wrong way around. I'd like to maybe start with what are the, what are the use, what are the areas that you're trying to automate and why? Like first of all, what problems have to be solved before we dig into a solution for that that may or may not be out there? What are the kind of bigger problems that you're saying, okay, if I could solve a, or problems with automation or with different technology, these are the things I need to solve for what are those things that at a high level in your type of recruiting and volumes, you think of, okay, I need to make XYZ better. And why would you need to make them better or more automated or faster?

Yeah, I mean there's an awful lot, you know, I think over the course of the last year, I've always sort of, I've always sort of liked historically to geek out on new technology that happens in ta. And I thought I was pretty au fait with AI, but that was sort of about generative AI. And now the world is changing entirely into agentic AI, which allows basically to have all these little helpers running around doing things in an automated or semi automated fashion and then learning over time and making things better. And there's a power there which is fantastic. I don't think there's any part of TA that can't be supported in some way by artificial intelligence or machine learning of some description. So for example, you take sourcing, you know, outreach can be tailored based on the individual. You know, if you've spoken to them previously, if you, if you know, have information about the company they work for, you can basically be running competitive intelligence all the time to understand the movements in and out of your competitors and who's going where and gives you new places to go target. If you look at your workflows within the process themselves, you can set scheduling software to be able to interpret the calendars of all of your workforce and identify the skills that you have within your workforce and then match it as closely as possible to the people who are identifying times that they're able to talk to us. You talk about offers. You can tailor offers based on market conditions, automated ranges, benefits packages that are more compelling to individuals. If somebody's from a demographic which is single, no children. They probably want a very different benefits package than somebody who's married with three kids and has recently relocated from the US to Europe, for example. So it doesn't give a very precise answer. But generative AI, everybody should be leveraging it to write better messages, to teach machines how to write messages for you, which you then have to read and double check. AI has its flaws. AI will try and make you happy and it sometimes does that by lying to you. If it doesn't have the information, it will make things up. So it has to be checked. But it can be basically a very cost effective way of being an administrative assistant for you. But agentic AI I think has massive powers beyond that in order to you can basically give it instructions, you have to train it constantly, you have to make sure that the work it's doing isn't biased in some way. And you almost need AI to measure the bias of AI sometimes because the quantity of data it's processing. But it can take elements of your workflow and basically remove the human element. Give humans the ability to be doing the high touch white glove treatment for candidates. I personally don't believe that particularly in tech recruitment, AI will replace a recruiter, the need with candidates that we mentioned before to build that relationship. The best recruiters are still the ones who build the best relationships with candidates. They're not the ones who automate everything and send emails off and never speak to a candidate. They may get more candidates through the process, but they don't close them at the end of the process. So it's that marriage is a perfect marriage. There somewhere between automating the pieces which can and should be automated, that don't have direct touch points with the candidates. But then to maximize the conversations you have with candidates, either by being armed with more information about them, by using AI to extrapolate more data from the world from the web, or to take away the time consuming elements of recruitment which we all have. There's administrative tasks. The best recruits in the world are also the worst. At admin I found maybe myself is a case in point. You know, I hated doing the paperwork at the end of the process. And that can all be taken off your hands. You know, you can press a button and offers and NDAs and everything can be created pretty much instantly sent away for E signatures and never has such a human hand. If they renegotiate, it can be done very, very quickly. You can have conversations with chatbots to answer FAQs throughout your process. But As I say, it's a very delicate balance between over automating, which takes away the personal aspects. Because even though AI can demonstrate sort of quasi empathy, you know, it can be seen to be empathetic. It's not genuine empathy. And sometimes you need that human interaction to really get behind, you know, the, the real reasons that a candidate is looking, or not looking to leave their job or why they're looking at your job versus another one. You know, the best candidates tend to be interviewing simultaneously in multiple places. And that's where recruiters I think have to still they can leverage AI, but I think they still have to demonstrate the personal skills to make the difference.

What about in terms of your own experience in actually bringing in automation? I know you kind of have tended to look at a big project per year to try and get your head around as a team and look to make work. So what has been the most recent kind of innovation you've looked at? Why did you choose that? How has it worked out for you?

Well, we have, we have two that we're looking at this year. We had to, we were using Prelude as our scheduling tool. Prelude were bought by Calendly and then swiftly deprecated as a product. Thank you Calendly. And then we had to find an alternative. So we did an RFP and we looked at pretty much everything out there. We looked at tools like Paradox, which is much heavier, much more heavily reliant on AI, for example. But we chose Good Time mainly because of a personal recommendation from somebody who also works in the greenhouse ATS environment where greenhouse shop work day as our hrs. And we wanted tools that worked as seamlessly as possible with, with Greenhouse. And Good Time was where we chose. And it does two things for us which are super helpful. One is, you know, automation of scheduling. It does, if you tag your, your candidates and your employees correctly with the right skills. It's very, very good at matching and coming up with alternative solutions. And the communications back and forth are sort of excellent, very well, very well constructed. But the critical bit for us is about the, what we call interview readiness journey. So we do interview training. We just had it introduced into our new hire orientation because it's that important to us. Before we had online training with our learning management system where everybody was expected to complete it within three months of joining. But what would really happen is you join, you get staffed and then they'd forget about these things and they wouldn't happen. Now Goodtime allows us to manage that automatically. It's part, we put a, an Hour two hours aside, in everybody's new orientation now globally, to complete the training, they have the block of time, so please do it now. And then as soon as they've completed it, Goodtime knows it's been completed by talking to the LMS and then they start their shadowing journey. So we do two shadows and a reverse shadow and Goodtime managed that entirely for us. We can track at any given time how many people are being trained, how not. If we have to reschedule interviews, which is something we're actively working very hard to fix, it will reschedule a shadow as well if we ask it to. And that just means that we can, you know, as the volume of interviews increases exponentially, so we can get more and more of those people through and becoming interviewers themselves as quick as possible. The other thing we're doing is what we call virtual shadows because the pass through rate for us is somewhere circa 4 or 5 to 1 when it comes to face to face interviews to hire. It's sort of. That sort of which is pretty typical for big tech. And as a result we realize if you're doing a shadow, two shadows and a reverse shadow, and the reverse shadow being one where you're conducting the interview but being watched by the senior interviewer, the other two, you're being the sort of the silent watcher, the voyeur that with that pass through rate there's a good chance that you don't actually see a positive interview example. You know, you, you have, you know, three, one in one in five get through and you know you're having three conversations. There's a decent chance that you don't see a positive experience. So we created when we, we. The other tool that we're looking at is that we just brought on board is, is called brighthire. And when we were talking to brighthire and Metaview, who have some of the tools in the. In the same. And screen loops are the one that we looked at. One of the things they came up with was this virtual interviewing example. They actually used the example of you can record the interviews themselves and use them. We decided that probably wasn't optimal in case we lost either the thought worker or the candidate wasn't in favor of their. So what we are planning on doing is using Notebook, which is an AI tool and you can basically, for those who don't know, very, very clever tool where you can basically write a script. It might be. You told me about this in Phoenix, but. And it basically can conduct a conversation, a podcast between two completely fictitious individuals. To created AI created individuals. And so you can have this interview content basically written and you basically anonymized it and future proofed it because the person isn't going to leave because they don't exist as an individual.

I haven't ever heard of this use case of Notebook LLM and I really like it in terms of creating a mock interview scenario that people can learn from using AI characters running the thing which is there forever. And so it can be the best practice interview. You could probably do a. This is not a good interview example. This is a great interview example. You could do multiples and you could spin these up really quickly.

Exactly. And because you're not filming it, you don't have to refilm it. If you want to change a module of it or an element of it, or you change a question or you think another answer to a question is a better one, you just rewrite the script and it just recreates the. The entire thing.

Love that.

It's, it's, it's super clever. But the. We have it now for multiple roles. We, we run a process called Keep it Fresh. So every two years we look at our entire process for each of our core roles its entirety. Sometimes it's just a refresh of attributes and a bit of fine tuning in terms of some questions we ask or don't. But this year we're looking to revolutionize developer, which is 60% of our hiring process. And we're hoping to roll that out really quickly from a virtual interview perspective because that will affect thousands of interviews a month. Truthfully, in terms of its volume, scheduling.

Is probably a use case most of our listeners will be familiar with. But walk us through the use case for Brihire a bit more. In terms of what is the problem you're solving for, how does it work and how does that drive value? What's the business case behind that?

Well, when we were a public company, we were being audited annually, as you are for SOX compliance. And part of that is they choose a random number of offers that have been extended on an annual basis and then you have to prove that you follow the pre agreed procedure, the process to hire them. That there weren't, we didn't, we weren't spending public money on, on, on sort of fictitious offers, etc. And when we, when we were preparing for our first one, we realized that the quality of our interview feedback was pretty dreadful. Some, some cases it didn't exist. We fixed that there. So no problems there. But, but the. We basically ticked the box that There was always some feedback from every interview we had for every offer that went out. That didn't mean the quality of that feedback was very good. I've worked at place like Google where you basically have to transcribe the entire interview. You spend as long writing the feedback as you do conducting the 45 minutes of interview there because you're basically meant to transcribe the ent. This is what they said, this is what I asked, this is what I heard. Becker's response. And here's my qualitative evaluation of what I heard back and that submitted to a third party to review hiring committees, make decisions, etc. Our quality was nowhere near that. It was basically it's greenhouse. So we have thumbs up, thumbs down. So we had lots of thumbs up and thumbs down. And then if text was required, candidate was great or something like that. But at least we ticked the box that there was feedback. However, unsuccessful candidates, silver medalists, etc, we had next to nothing. Yeah, because they didn't have to go through. They weren't being beaten up by the recruiter to submit anything because we weren't going to be making an offer and it wasn't.

There's probably also fear in terms of I didn't hire this person. So let's, let's say nothing about them people. My experience is you just. You don't want to say anything negative so you just leave it as blank. There's a bit of a culture of that sometimes as well as the laziness are just moving on piece.

We've always tried to give feedback. We've always prided ourselves on giving feedback to unsuccessful candidates. But that's really difficult if it hasn't been captured in the right way. And often it would have to be a conversation, offline conversation to give the recruiters the ammunition to then go back to the candidate and say this is the reason why please apply again in the future. But without any data in the systems, they had to basically go through the entire process again. So Brighthire, which is our chosen technology, which is going through, we're live testing it with our TA team right now on all their recruiter phone screens which. Which every candidate receives. So at the start of their process it'll go live with the rest of the business probably in May. But what it basically does is it's Super Clever, uses AI to transcribe 99% accuracy. Allegedly we. We've spotted a couple of hiccups with accents, but that's to be expected. And we have accents all over the world. Most of our Interviews are conducted in Eng, not universally. It had to understand Portuguese and Mandarin. And what the tool does very cleverly is it transcribes the entire interview in the language it's delivered in, but then it translates it for summarization and at the press of a button, it interfaces directly with Greenhouse and even will identify which attributes the questions are being asked against. And so it auto populates, not just like a stream of consciousness in terms of, in terms of feedback, it will auto populate the individual sections for attributes that you're trying to assess in that interview. Then all the individual needs to do is apply the thumbs up, thumbs down, overall summary. One paragraph and we have everything. And that's for all successful candidates. Unsuccessful candidates. The other thing it does, which is really clever is as part of this keep it fresh process we're doing for developer is we always try and come up with example questions. We don't expect to be prescriptive. We don't expect them to follow an exact script that's sort of way too controlling. But for new interviewers who aren't perhaps as confident yet, maybe they've just done one reverse shadow and this our first. The other thing it offers is a sidebar, so it interfaces directly with Zoom. Zoom and Greenhouse are actually two of their clients from a TA perspective. So the interfaces are really, really seamless. But within Zoom, the instance that the interviewer sees, there's a sidebar with questions against the attributes that I meant to be asking them for that interview. So if they forget, if they have a, you know, if they're hungover and they've forgotten, you know, what they're meant to be asking, they all they have to do is sort of glance slightly, right? And there's a whole suite of questions that they can, they could, they could jump into, they can probe with, et cetera. They are then building rapport with a candidate. You know, it's always we 95% of our interviews, probably higher than that, are over Zoom these days, ever since COVID And it basically allows them to build a rapport. You know that most of the interviews are one on one, apart from when the shadow is involved. And sometimes the shadow turns their screen off to be less distracting, their camera off. But what it basically does is it just allows the individual to be sat there, you know, having a conversation, building a rapport, not taking notes like crazy. And it just produces a better candid experience. The recruiters that we rolled it out with in a couple of regions, there were some question marks about it. They felt, is this being Big Brother. What's this recording being used for? What's the transcription being used for? But now they find that again it makes their job so much easier. They're able to ask all the questions, maximize every second of their 30 minutes of a recruiter phone screen, build a rapport with a candidate, you know, start the relationship building which will allow those offers to be facilitated at the end of the process.

But how the candidates feel about it? What would I assume they have to opt in. So what's been the opt in rate so far?

They do, they do. It's globally. The data we had from from Greenhouse was globally. It's about 3%, 2 to 3% that will drop out, that will opt out. Everybody, everybody receives the same notification saying this will be conducted over, over bright higher Here the rationale why we want to make objective decisions, not subjective ones. And this allows us just to be able to do that while building relationships. And our response rates are better than that. Where you know we've rolled it out in IME and Latam and we haven't really had any pushback apparently Germany tend to be one of the most resistant to it as we haven't rolled it out in Europe yet. That's coming next but the response rates are really good and the candidates like it because as I say, you're having a conversation rather than just sort of being grilled by somebody who looks down every five minutes to write something down or to look up a question. It's all sort of quite a seamless experience. Those super well received.

So I take those two examples, I think they're brilliant examples of practical automation using AI in a hiring workflow in a professional environment. What has been or what was the business case justification? How is that built? Is it cost saving? Is it extra value added? Is it time back for let's say each of those. For the. I know you had to go back out and re or FP for your scheduling software but even beyond the original version of that and then the interview assistance, if you take those two, were they the same kind of type business cases or how did you go about building that to justify the spend the time, the effort, all that stuff?

Yeah, the, for the, for the scheduling it was sort of, it was, it was necessary because we'd already made the business case for scheduling it is about time saving. Not necessarily time saving on behalf of interviewers but definitely time saving on behalf of our team. Being able to match automatically being able to identify the right interview because you know, historically you might buy, you know, we have a system internally, which is sort of an intranet where you can see skill sets of individuals. But that was a very manual process that every, every coordinator had to go through for every interview they scheduled. And there was some gamification as well. Sometimes interviewers wouldn't want to conduct an interview in a coding language, which they're more incapable of doing, because they didn't want to be staffed with that language in the future. So take it off their profile and then we can't use them as an interviewer. This sort of overcomes that a little bit and just allows us to be fairer. Distribution is fairer as a result of it, because you can set very, very clear parameters within the scheduling software which say, if this person has done an interview this week, do not schedule another one and find somebody else and do it. And that ultimately the business found very compelling, because we weren't. We weren't rewarding the people who were martyrs and taking on 30 and 40 of interviews a month. We were trying to be as equal with distribution as possible. So that was the essential business case for that side, for the bright hire side. It's time saving of the business. We had been working so hard over the last three or four years to finally break the back of interview availability. But this basically said we can maximize the hour you spend speaking to candidates. You don't have to spend 15 minutes at the end of the process to complete your feedback. It's more or less an instant assessment. And if you have any suspicions of things like cheating, for example, you can go back and review the video footage. Now, the video footage, we've had to say internally, particularly with some locations where working councils have a lot of say in what tools can and can't be used, we've had to promise that we're not using the video nefariously. And, you know, if there is a video recording, you know, particularly if they haven't been recorded before, they might be using Zoom, but no recording of it. That was a little bit of a hurdle to overcome. So we, so we, We've said this is strictly for your usage. Recruiters don't have access. We do have admin to have access to it, but the recruiters themselves are not going to have access to it. So really, this is for your review. So if you suspect something, you see you saw something weird, if the screen went a bit funny or they sort of looked sideways and were slow with the response, rather than you thinking they had may have done something inscrutable, then you can now go back and check it, basically, and so that's been a super well received item too. So that was time saving, interview feedback, time saving, the ability to maximize every interview it does. And, and the best thing is about the training because it gives you a dashboard based on the role you're the interview type that you're interviewing for, the role that you're in that the candidate's in. It gives you aggregated data from your peers and will say, for candidates that have been successful, we would recommend that you allow the candidate to speak 10% more, 15% more, etc. You can even tie it into recruitment marketing content and say you answered these questions around the culture in this way. Here's access to some of our cultural alignment content, you might want to read up on it, etc. But again, it's not coming from us. It's coming from a tool which is interpreting what the content says and making suggestions on behalf of the company rather than on behalf of the recruiter.

When you say time savings, let's say we go back to the interview scheduling, does that mean the headcount reduction in your team because you don't need as many coordinators versus what you had years ago before you had scheduling. And then what does that mean, time saving? Because I imagine your CFO's office going, great, can I take that to the bank? Because sometimes we talk about time saving and then it's like, well, do our recruiters go home early now? Do they do two hours less a day? Are we actually translating this into we're getting more hiring for less people? Like, how is that translation on the.

Business case, the hope is we get more hiring for the same people. We're not seeing headcount reductions. We did lose people during the slowdown on the team, but we're now growing fast again. And we've always had a ratio of coordinators to recruiters that has worked optimally, which is sort of about 3 to 1. And we expect that to more or less continue. But we think that the ramp up is reduced because the amount of automation at the behest of all of the coordinators is so much better that they're ramped within a matter of weeks rather than months now, so they are able to get up to speed. So when we do add new team members that they're productive almost instantly. But the big one is around the interview. We found that half of our time to hire is within the face to face interview stage. So that's basically with the business. Yeah, us trying to find interviewers, us chasing feedback, us moving to the next stage. Etc. And so if we can reduce that by having more access to more interviewers faster, then we get to the table first with an offer. They can be places with longer notices like Germany and India. They can start serving that notice sooner and just beats us to the punch. So there are massive savings from that perspective. The mainly from the time to offer accept, but we probably will move to a ratio where we don't need quite as many coordinators to recruiters eventually. We haven't seen those savings yet, but we expect them to be more efficient and just be able to do more with the same amount of more value.

Adding more value, getting folks in fashion that could be built out to clients clients like four or five days faster translates to real revenue on the volume that you're hiring for.

Exactly. Every. Every day we save in time to offer accept is a day extra billing we have on client sites for sure.

Where do you go from here both in terms of what you're looking at for thoughtworks and then what do you think of the other interesting use cases? I love the LMM example you threw in there, but what else is out there that you're thinking, hey, I'd love to investigate XYZ or you're seeing in other applications in a professional hiring context that you think is interesting that we should be looking at?

Well, something we're looking at is sort of dynamic sourcing, if you will. We expect, as I said earlier, 40% of our hires to come sourced. We have a sourcing coe who are based in India that support globally, but that's finite in its size. You know, we're constantly getting to limits in terms of throughputs, etc. And we expect every recruiter to source as well. Some of our recruiters are far better at that than others. And so what we're trying to do now with the sourcing team is that they're creating dynamic dynamic Boolean searches, for example, based on success or failure of candidates and not just using and, and, or, or, but also using nots. By, by seeing unsuccessful candidates go through the process, they, they're adding a lot of nots to the Boolean strings. And then essentially we're running agentic AI to constantly look at success factors, no pun intended. We don't use that as an ats, but within the, within the process to, to work out how we can fine tune, streamline, add to search parameters so that their recruiters are all armed with the same information and the same capability as the, as the sourcing coe. So the sourcing coe are definitely the experts in the field and they're the ones who are helping develop this technology. But the goal is that everybody within the team will have access to the same AI generated content, which will then mean that their searches are better. Because if you run the same search in LinkedIn every time, you get the same search in the same order, etc. And what this does is it allows you to get some variety. And so rather than having to in a linear fashion, go through 450 people in the search and send outreach, you can, you can fine tune your Boolean search based on response rates to outreach. What the A B test outreach, which ones are more successful with which types of candidates, which companies, at which times, doing which roles are ones that are better fit for what we're looking for, etc. Or for specific clients that we're hiring for. So it's all about using, leveraging agentic AI to really take the sourcing to the next level. Again, it's not to replace people, but it's to arm everybody on the team with the same skill set to be able to be more successful in their sourcing, to improve their outreach and to get better response rates to put more candidates into the, into the flow. Because we do get the lion's share of people that complete the process. One they say from a candid experience perspective in the surveys that they've enjoyed the process. We have this thing called joy of interviewing, which is meant to be joyful for both candidate and interviewer alike. I don't know how true there is, but the candidates seem to like it and, and the vast majority of them that we give them an offer and we're not in the 100 percentile, we're not competing with Google and Amazon from a comp perspective, but the majority of them do say yes to our offers. So if we can get more successful candidates into the top of the funnel that we've sourced and that perhaps aren't being spoken to by 40 other companies simultaneously, then the better chances we have of hiring faster, better capabilities, new skill sets, et cetera.

As an example, I spoke on Monday with Dhiraj Gupta and Amazon, who runs the, I think the ads business recruiting and they're using, they hire huge amount of AI scientists. Average time to hire is about nine months. And he was saying they built a similar kind of AI product that, that is automating this, the Boolean searches and using learning around the exact keywords to use and not not use because of a lot of, lot of BS in the AI candidate world about what is A real candidate there. They've been able to fine tune this to identify talent much faster who are actually, who are the right skills and also probably going to be successful in an interview process. And that's massively shrinking that nine months down. So you know, you see those kind of top of funnel use cases around. How do you take a traditional thing like Boolean search writing, put AI on it and you get really powerful strings that are laser focused on the right candidates. But to your point, scale across every recruiter, you're not replacing the recruiters, you're giving them a better tool to go faster which has impact on product development cycles or on time. You know, out with, out with customers as well. Well, it's an interesting future. I love the, to start with, you know, you're talking about moving from, moving towards agentic from, from just generative as we are today. I think you're going to see the next year an explosion of new tools, new use cases. But Marcus, it sounds like you're, you're doing well in terms of, you know, really nailing good use cases and professional hiring. It sounds like you're, you have already a couple of candidates lined up for next year's projects. Would that be fair?

Yeah, I mean typically with the relationship we have with the business, we're able to look at one product a year to add to the ecosystem, add to our tech stack and then just improve things. So we're trying to work out what's next. We're happy with Greenhouse. We have another 18 months with greenhouse right now. There are some interesting looking ats. Ashby is the one that looks really interesting within the market right now. I'm sure my old colleague Jim Miller will be delighted for me to say that. But they do look really interesting. I don't know about the scale yet. You know, we're 12,000 person company. I'm not sure that they have many enterprise clients like that yet. But what they're doing, they start with the data first and that's half the, you know, when dealing with the technical audience that we have, the data is critical. Everything that we do from a business case, from a conversational perspective, from a. Trying to persuade better behaviors from the business, it has to be sort of embedded in data. Yeah. And so it has to be accurate, et cetera. And Ashby seemed to have approached it from that perspective. So that might be something we look at 18 months from now. But we've already started the wheels turning. There's a, there's a really good map, tech map which comes out every year. You know, with all of the products in the market and we're sort of plowing through that to see is there something else that we can be looking at. We haven't got a specific idea yet. I think some of it will come from the success or failure of our current rollouts and we have two simultaneously basically now. But there'll almost certainly be something new that we add to our tech stack in the coming year.

I feel like an annual update is too infrequent in the world of AI at the moment. You could be your quarterly updates for that kind of tech stack list. Mark, thanks so much for joining us today. Sharing your story, sharing your insights, plugging some ideas. Hopefully it's helped our audience with some ideas in terms of what they're looking at and in in terms of their use cases for automation for AI. Some of the tools you've used. If anyone wants to reach out to particularly ask about any of the tools or projects to get some advice for their team, what's the best way to reach you?

Drop me an email. My email address is and yeah, please, please drop me an email and I'm happy to answer any questions.

Hopefully the AI spammers won't then just jump on.

Yeah, I just thought about that. As I said, I'm going to get a lot of cabinet emails.

We might have beeped that out by the time we broadcast this to make sure. But maybe on LinkedIn or reach out to us and we can connect you real people only, not AI chatbots. Yes, we'll do the waving hand in front of the screen to make sure you're real. Thank you for joining as well this week. And to listen on the show, don't forget we're back at the weekly podcast. You can Watch us on LinkedIn, you can watch us on YouTube, you can subscribe, Spotify, Apple, wherever. Get your podcasts. Do do, click that subscribe button and you'll be back. We'll be back next week with another great guest and some more insights for you as well.

Thanks for listening.