#145 – Arpita Bagchi | How LuxExperience uses AI to make TA more strategic

Thomas Kohler
Today’s guest: Arpita, Head of Talent Attraction at LuxExperience. I’m really happy that we do this episode today. But maybe first we start with a quick introduction about yourself.

Arpita Bagchi
Thank you, Thomas. Thank you so much for inviting me, first of all. I’m Arpita. I’m currently the Head of Talent Attraction at LuxExperience. I’m responsible for all the recruitments in the company with regards to all our brands. So we have three major brands: Mytheresa, Net-A-Porter and Mr Porter. And I’m basically recruiting for all the three brands, all corporate roles. We kind of hired about 600 to 700 roles in a year. And that’s across locations, both all across in EMEA, APAC, North America. That’s where our remit is. Prior to joining LuxExperience, Mytheresa for that matter, I was working with Micron. It’s a semiconductor company. And before that, I was working with zooplus. Both roles were in the talent acquisition space.

But previously, I was also living in London, where I was working with Unilever, which was my role in the leadership development side. So I was responsible for developing leaders for Unilever, which, as you may know, is one of their biggest pet projects in the company. So I devoted quite some time doing that. And before that, I was living in India, and I did different kinds of roles across various large companies in India, from HR business partnering to talent acquisition. So length and breadth of all, I mean, touched almost all aspects of HR while I did that. I also studied HR. So I graduated, or rather did my master’s, in human resource management and labor relations. So that’s a little bit about my journey and my career.

Thomas Kohler
What I find interesting is that you were also in a talent development role and then now lead a talent attraction. How does your experience from a talent development role affect the way you look at talent attraction? Because I see a lot of, I would say, talent attractions or recruiting leaders that were always in recruiting and maybe even come from an agency side, which is maybe even more focused on top-of-funnel filling. How did this affect your thinking about talent attraction?

Arpita Bagchi
So when I’m hiring for people, we are just not looking at filling numbers. It’s always also about the experience that we want to give the candidates. How can we give them an experience that makes it memorable for them and also helps them? And we also try and give them a picture of what they will see in the company once they join us. So retention part of me has not left. So I want to make sure that they have a window to what they will experience in the organization. I think an experience in talent development is very important for everyone, just to understand what ticks for people and what doesn’t tick for people, because that is a very important linkage also in talent attraction. You need to find areas and areas that will impact people. So when you have those conversations with people, try and understand what is it that ticks for them, what will make this job very memorable, very exciting for them, and then focus on those aspects.

Thomas Kohler
Do you also hire based on skill matrices, or do you have also an overview internally where you have a gap of skills in the organization, or maybe a density of skills, and how you shift the workforce from a development perspective, but also decide when you hire internal versus external, for instance?

Arpita Bagchi
So when it comes to skill-based hiring, that’s a topic that is developing in the company. What we do is, and that is something that we do within our remit, is when we have these intake calls with the hiring managers, we try and have a very detailed discussion with them, just to understand what is it that they are looking at, what are the skill sets that are missing in the team, and what is it that they want to hire for in this new person. And we have now started, and this is moving to the AI side of things… What we do is we have… we built a gem within, because we are a Gemini workspace company. We’ve built a gem which basically, based on this discussion, helps the TA manager to identify the skills that need to be looked for in the candidate. So we basically develop our talent rubrics, and we shared it with the hiring manager. They have been very appreciative of what they’ve been seeing, because it’s something very nice. Recently, we’ve introduced in the talent attraction workflow, and they can basically check mark and say: Is this what we need to hire? Is this a green flag? Is this a red flag? And they find it very useful. And I think that is helping us in doing that skill-based hiring. It’s a project that is in development within the organization, but from our side, when we are doing the hirings in the organization, we are looking at it already.

Thomas Kohler
You are the first person I talked to that used the LinkedIn Hiring Assistant. It’s their AI product, what they, I think, try to launch and sell and roll out, that is using it and has good experience with sourcing with it, right? How?

Arpita Bagchi
It’s very important. The inputs that you give are making sense. So what we’ve learned, having used AI for now quite some time, is: input is equal to output. AI has no intelligence of its own. So unless you give it very, very clear instructions and very precise instructions, it doesn’t understand. And that is something we have observed with different tools as well. If the instructions are unclear, blurry, it just doesn’t get it. So we put in a lot of effort to make sure that our instructions are very, very clear.

And it’s a problem with LinkedIn Hiring Assistant at the moment, or rather it was, where the only language they were supporting was English. So it was naturally English. People who were native speakers found it easier to communicate with the tool. The moment you had people who were in their head translating from one language to the other, it didn’t work so well, because the conversation was not flowing effortlessly. And that is where we noticed, when we had, let’s say, some of our London team members working on LinkedIn Hiring Assistant versus, let’s say, some of the people here in the German team, there was a big difference in the output they got and big difference in the satisfaction that they had from the tool. And I think that was a little bottleneck with LinkedIn Hiring Assistant. But now, having launched in German language as well, I think it’s working better for people who speak German. It’s been doing a pretty good job. You also have to do a little bit of tweaking here and there to make sure it understands your input pretty well. And you have to give it, constantly give feedback to the tool to make sure it understands.

Thomas Kohler
How do you do that specifically?

Arpita Bagchi
So every candidate that you open, and let’s say… and you have different sections, and in each section there is a thumbs up and a thumbs down. So you have to go to each and every detail and say: Yes, this is right, or this is not right. And if you give it the input, it understands and learns. And you have to constantly make sure that it learns. Sometimes it has its flaws, and it doesn’t pick up. But if you keep on doing it, it’s after all an AI tool. It learns, and then you get very good output out of it. We’ve found the tool quite useful.

Thomas Kohler
So you are already at a stage where you said you trained it and you put in a certain amount of time, and now you’re already in the phase where you got at least back what you invested in terms of time savings, and even more, that it’s already running efficiently and saves you hours of sourcing or screening candidates. Is that already the case?

Arpita Bagchi
No, because every project you will need to train it. It doesn’t work that you’ve trained it once and it understands. Every project it needs training, but now we know how to do it quickly. So we don’t spend so much time training it, because previously we didn’t know where to click, which one is actually the input for the assistant to learn. There were multiple these thumbs up, thumbs down things. But now we know where to click, which is the right button to click.

Thomas Kohler
Give me a role, for example, where you trained it.

Arpita Bagchi
For example, let’s just pick up a random role. Let’s say that I’m hiring a recruiter for the company. And when I’m hiring a recruiter, I would give it inputs like this: So I want the person to have three years of experience, very specifically experience in technical recruiting. And I will maybe, for my company requirements, I’ll tell them, tell the tool that I need somebody who has Workday experience. Probably I’m looking for somebody who has… And then I will put in nice-to-have, like, you know, should have been using some AI, should have been working in the same industry as we are, those things. And then I run the search.

Now, it’s very generic what I’m talking about right now. So when I say three years of experience, or do I mean three to five? Three plus: Ten is also three plus, and five is also three plus. So I have to then start being… becoming very specific, tell it that: Listen, I want somebody between three to five years of experience. When I’m saying technical recruiting, I want experience recruiting in digital marketing roles, like… and then give it some examples. So for it to understand that if the person is… has it in the CV or not, or in the LinkedIn profile.

Thomas Kohler
That’s, that’s not so… yeah, sometimes I think then you’re faster by just using the filters or the Boolean aspects, right? And then the question is also: When does it make sense to train it instead of just doing a quick search? Because sometimes it maybe doesn’t really give you an ROI on the invested time. Do you have a model for that, when to train it and when not?

Arpita Bagchi
So always we start with the AI tool, because it’s natural language. You can type, you can type as much as you want, and it tries to understand and make sense out of it. If we are struggling, and we are saying, okay, we are unable to do… train it, then the Boolean search and/or the filtering works for us. But we’ve noticed, the more you kind of give it feedback, because it’s also throwing you more options when it’s doing it, it makes your life a little easier. Because most people, when we are… rather, when we have these intake calls, these are just statements for us to make… convert those statements into very specific search topics. Of course, we have those gems as well available with us to kind of turn these into Boolean search strings. But it’s always easier to put those talent screen rubrics right in the description and ask it to look for these exact skills, because we have the… as I said, the green flag, the red flag on the skills. So we know these are the exact skills that we are looking for. Sometimes the drop-down will not give you so many options in when you are filtering, which makes it a little dicey, and which is why we prefer the natural language to begin with. And then maybe, if we are struggling, then we start with the filtering. But as of now, not so much.

Thomas Kohler
You mentioned what is important for, I think, the listeners: the talent rubrics. So you already have a system in place that is defining roles very systematically. What else do you have in terms of systems that enable recruiting? Nothing to do with AI, but in order to then leverage AI.

Arpita Bagchi
Yes. So, for example, as you know, Workday has its own AI tool, but most companies do not use it because it’s so expensive. We have Gemini, we have the Google Workspace with us, but they don’t both talk to each other, because we do not have an MCP in place as of yet. But you can always download the Workday CSV file. And what we do is kind of use that to identify also bottlenecks in the process, because we’ve had situations where hiring managers have not screened profiles for a long period of time. This is a pretty common thing. We would use not AI directly, but download the report, use it to find a solution and identify the exact bottlenecks.

Thomas Kohler
Based on their recruiting data, like the recruiting process for certain roles or certain roles, to say: Okay, there are certain time steps, and on average, in a certain requisition or opening, the average time in stage is 14 days. That’s long, for instance.

Arpita Bagchi
Exactly. Because a lot of times when we are doing our intake calls… what we also have something in place right now is a start date predictor. This is nothing but an Excel sheet. So there’s no AI in it. It’s just a pure Excel sheet.

Thomas Kohler
The prediction in itself is already fancy.

Arpita Bagchi
Yes, it’s very cool, actually. It’s not something very old. It’s very recently introduced. I’m trying to push it across the team members. Basically, it’s a pure Excel sheet. You have the budgeted start date for the role and the intake call date for the role. And then you have different stages per hiring process. The stages are more or less common. And you just need to enter what is the SLA per stage. And that’s in pure Excel sheet entry. And then you have a couple of columns which says during which period you will not have the hiring process running. For example, the hiring manager is going on vacation. So right in the intake call, you can just fill in those dates.

The model will calculate: Okay, the hiring manager wants the person to be staffed by 1st of October. But given that the position has notice period, so we have that cell as well. So you put in four weeks, or in Germany it will be 12 weeks. It doesn’t mix. It doesn’t look like it. Maybe the realistic start date is 1st of January. And we can actually immediately show that to a lot of managers. And it’s a dashboard. So then, this is an Excel sheet, but then we can do it very fancy. We have a Looker Studio, which I learned over the weekend. And I was like, okay, you can just transport it to a dashboard, and they can see the visualization. They don’t need to see the Excel sheet. You can do a lot of these things.

Thomas Kohler
So what you actually then do is something what I really believe in. And what we also do is: You get, let’s say, a start date of a role from the budget. And that is finance put it in there as: Okay, from this start date, you could hire. Then some hiring managers don’t have this full context or see it in that way, that they could hire. They said: Hey, we have the role budget that the person needs to start on that date. And then there is the recruiting perspective to say: Okay, now it’s September when we record. And let’s say you get roles approved starting October, and then they want you to hire now, right? To kick it off, to attract the right candidates into the funnel, to have the candidates through the funnel, and then to have a signed contract until they start. This could be months, even half a year sometimes, right? Sometimes even longer. And a lot of managers, they don’t see it in that way. And then what often happens in a dynamic, if they have pressure, they put the pressure further to talent acquisition, and you never really had a chance to deliver it, but then you’re the one being blamed.

Arpita Bagchi
Exactly.

Thomas Kohler
And this is what you do to have a hiring prediction model. Very simple. It’s perfect for that, because then you can set expectations clearly. And then they also can have a different conversation with the business and push back: Look, we have the talent acquisition data. This is, yes, budgeted, and of course we have pressure to deliver some projects, but this is also most probably realistic. Let’s consider that as a plan. And then they are also enabled through the talent acquisition team, which puts you more into a strategic role than into this transactional role.

Arpita Bagchi
Exactly. Just receiving a date doesn’t really help you. You have to be able to tell people: Listen, the date is there, but that is not the reality. And fill in some days of vacation, that is definitely out of reality. So that is the conversation I want the team to have internally as well.

Thomas Kohler
Overall, what are other things that you do that make your talent acquisition department more a strategic partner than a transactional delivery role?

Arpita Bagchi
See, some parts of the transactional delivery role can never be eliminated. It’s important as well. And I know there are a lot of AI tools that are available in the market, which are also selling that we will take out the scheduling part of it. But in reality, we are a 4,000-people company. It’s not going to happen. For the tool to work seamlessly with our hiring manager’s calendar, the calendars need to be updated, which is never the case. So you have a situation: The tool cannot read the calendar. So some of those operational aspects will remain.

Where I want the team to focus is the strategic part of it, the areas which they themselves can control and have an influence over. Like, for instance, spend more time on having these intake calls, get more clarity, and make sure when you have these intake calls, you send what has been discussed. It’s not just that you need a confirmation on the salary for the role or the start date for the role. You need a confirmation before you post the job about the specifics of the role, because I don’t want the hiring manager to turn around two weeks later and say: Well, this was not what we discussed. We need to start the search again. Get a confirmation. Make sure you provide them with abundant data, like what is the market reality of the role. Like, those little gems that we have actually given output of worth seven pages. So seven pages of data is available. You decide what you want to share with the hiring manager. Make them aware that this is the reality, this is going to happen. Do that during a call. Rather than having a 30-minute short call, have a longer one-hour call, because it will make things easier for you.

What we are also trying to see is how can we make our TA partners more strategic in terms of just the interview process that they do. So, for example, the CVs have come. They have screened CVs, and we received plenty of CVs, considering we are in a fashion industry. There are lots of people applying into the company. How can we screen those CVs faster? Is there a way to do it better? We are looking into tools which can kind of help us in doing that. So that’s a very important aspect that we are looking at, because we want to empower our team. We do not want them to spend countless hours looking at CVs which are below what is the skill matrix that we have set up for the role. So we want them to look at good-quality CVs and make a decision whether the person is good enough for the next round or not. We want them to focus on that.

We want them to have quality conversations with the candidates. Make them understand about the role, what is it to expect, the skills, the timeline, everything should be part of the conversations. Those are aspects that we can control. When it goes to the hiring managers, we also want to be a strategic partner, where we tell them that: Listen, this was our conversation with the candidate, and this is what our insights are, and what we think you need to ask in the next round of interview to these candidates. And given the skill matrix that we have developed, maybe these are the areas that you need to dig deeper on. Those are the things I want the team to do more and more as we see AI coming in to do the basic stuff for most of us.

Thomas Kohler
How do you envision, let’s say in the next 6 to 12 months, how AI can also affect your work in the talent attraction space? Because speed is moving rapidly, and you’re also actively building agents, gems, how you call them, and so on. How will it look for you?

Arpita Bagchi
I think we will need those tools to do the admin part of the job. So the heavy lifting will still need to be done by the recruiters. As everybody says that the human has to remain, which is also as part of the EU AI Act. They will not get replaced. I know, I speak to a lot of the team members, and there is a little bit of reluctance in using tools, and they are kind of like: But what am I going to do? No, you are not replaceable.

Let’s take, for instance: AI interviewing tools have become very common now in the market. But that AI interviewing tool is not going to do any assessment. You need a human to listen to those video recordings or audio recordings to make a judgment. We’ve been testing some tools, and we can actually hear people typing things. But the summary that gets generated by these tools is excellent. But only if you hear the recording, you know that somebody was probably using another AI while they were doing this interview. So it’s quite funny. So if you take out the human from this process, who’s going to check it? There is nobody to check it. I will solely rely on that AI summary to decide whether the candidate is a good fit or not, which is probably not the right measure, because the person was actually using AI during the interview process.

So people will never be replaceable. What we need to do is convince our recruiters that they need to take a step up. It’s a brilliant opportunity to upgrade and upskill yourselves. We’ve always complained that we are saddled with administrative stuff. This is our opportunity to kind of say: Okay, maybe it’s not bad. Let’s use AI for all the admin, low-key jobs that we were doing. And let us be more productive. Let us be more strategic. Let us talk to our hiring managers, become partners. Right now, every recruiter has 10 to 12 positions. Who has the time to have these strategic conversations? It is not possible. But these tools will enable you to do that, because you will have a lot of free time. Because based on the trials we’ve done, a lot of people who are testing this have told us that: I do have a lot of spare time. What do I do with it? I’m like: Yeah, this is the time we give you back to have those strategic conversations with the hiring managers, with people or with the candidates, and do some more sourcing, look at more candidates in the market. This is the time you have with you now, which you previously complained about.

Thomas Kohler
Thank you, Arpita, that was it. It was really nice talking to you. Thank you so much.

Arpita Bagchi
Thank you so much. Thank you. Thanks a lot. It was lovely having this conversation with you.