Developing an AI-powered job search to boost conversions

Now that we knew our weaknesses, we began brainstorming ways to address the problems mentioned above while also looking into the quite exciting possibilities an AI-powered job search could offer. I love this phase, as we usually completely neglect the budget, the timeframe, and our limited resources and come up with some wild ideas, which is always fun.

Budget, timing, ressources…cutting it down and making it something a HR team wants to invest in

After all the brainstorming and coming up with great ideas, this is always the hardest part. But let’s be honest: budgets in employer branding are usually small, resources are limited, and we are not developing a product we want to sell to make big bucks or that has millions of users. From a business perspective, it is still just a career website. It’s important, but not that business-critical. For me, it is important to really create value with what we do, and most often this means we cannot develop the product we would love to. Instead, we need to focus on creating the product that actually delivers the results we need. So, while we decided not to use ChatGPT, both from a budgetary and technical perspective, we still wanted to leverage AI technology and incorporate open content in our search, making it possible for users to use the search bar as the central element to navigate our careers website, tackling both our high bounce rate, and improve our conversion rate for job searches.

Putting the search module front and centre

Okay, I admit it: putting the job search module front and center isn’t a groundbreaking idea for a career website. Sure, we previously made it easy for users to access the search with a prominent button on the main page, but the search bar itself wasn’t the first thing visitors saw when previously entering the New Work careers website. Based on user feedback, we wanted to make the search the central element of our site, encouraging every visitor to interact with it—no matter what they’re looking for. So, we decided to eliminate all unnecessary elements from the first view port. We removed images and took a very minimalistic approach on the design, making sure that the search bar truly stands out and grabs our users’ attention.

We also included a section below the search that shows what other users have been looking for. I admit, there is a bit of an overload on recommendations, but we really wanted to make a point that users can search for everything.  We programmed the recommendations in a way that we can influence what appears there and suggest terms that might spark our users’ imagination. An interesting finding after launching the search: the suggestions are actually the search terms with the most hits, and they work great for promoting certain jobs or job fields!

Having the search everywhere at your fingertips

Opening the doors to conversational searching

We wanted the search experience to feel as close as possible to a fully conversational interaction. Inspired by technologies like ChatGPT, we aimed for our AI search to truly understand conversational queries and provide users with the results they’re looking for. By prioritizing this conversational approach, we ensure that interacting with our search feels natural and intuitive, making it easier for you to find exactly what you need.

Conversational search still needs learning: Both for the user and us

We wanted the search experience to feel as close as possible to a fully conversational interaction. Inspired by technologies like ChatGPT, our aim was for our AI search to truly understand conversational queries and provide users with the results they’re looking for. We wanted interacting with our search to be natural and intuitive, allowing users to input as many job-related wishes or keywords as they wanted to discover the perfect job.

Interestingly, it seems users aren’t quite ready for a conversational search experience when looking for a job anyways. Despite our efforts to encourage users to engage in a conversational manner, only 3% of them utilize the search in this way. Most job searches still rely on two or three keywords or simple job titles. This was evident from our initial tests, and even after the search has been live for 3-4 months, we haven’t noticed a change in user behavior. Job searching is a skill that has been developed over many years, and it’s unrealistic to expect users to adapt their job search habits overnight and suddenly write lengthy and detailed wished on how their next job should look like. But I am sure that is where the future of job searching goes.

Optimising conversion rates by integrating content into the job search

One of our key ideas for the new search was to integrate content, allowing us to offer not just job listings but also relevant resources from our website. In the beginning we learned through our data and user feedback that many visitors quickly bounce from our career site because they can’t find what they need. From the beginning we wanted to create a search that not only offers job openings, but also answers career relevant questions. By integrating content into the search, we give our users a search bar where they can simply ask whatever question brought them to our site and receive appropriate answers. Things like who is my recruiting contact, how does the application process look like or even questions like what are your benefits, what is your culture like or why should I actually work for you will now be answered by our search.  

Adding content to our search results also created a great opportunity to boost our conversion rates. User feedback revealed that much of our content goes unnoticed — around 80% of our users come for job ads rather than directly for our content. But pretty much all the content we have on the careers website has the goal to actually convince people to join us, by showing them what they can expect and how it is to actually work for us. But most people do not take the time to scroll through our website to find all relevant employee interviews, videos etc. Now that we included content into the search we can actually promote it within the job search. Meaning if someone types in “backend developer” they will not only see suitable job ads but also an interview with a backend developer that can give them more insights on the job, increasing the chance for an application.

The good, the bad, and the AI!

Honestly, I am really happy with the results we came up with for this project. The numbers after launching the search are great and speak for themselves. We nearly doubled our application conversion rate, which is a huge win for the Recruiting and Employer Branding team at New Work. More than 80% of our website users now actively engage with the search feature, and we successfully reduced the bounce rate to the mid-60s. Yeah, this could probably be even lower, but a website called New Work really does attract a lot of “bad” traffic. Many people come via Google expecting the New Work Group website to offer jobs from around the world, which of course we don’t. So we are also pleased with reducing the bounce rate to the mid-60s. The decision to integrate content into the search was also a real win—our content has seen a great increase in traffic simply by being accessible through the search. Even more exciting, we’ve observed that visitors who explore the employee interviews featured between job ads are applying at higher rates, positively affecting our application conversion rate.

But, of course, there were also many moments when I hated this project. A real bummer was, and still is, that we clearly had limitations regarding our AI model. In the beginning, it was always the dream to create something like ChatGPT that has a good answer to everything when searching and feels more like a conversation. Unfortunately, we did not manage to achieve that due to budget, timing, and resource constraints. Additionally, conversational searching has not really taken off yet. But given that our AI is not advanced yet, it turned out to be a good thing for us. That brings me to my last point: setting up a search, teaching it how to behave in certain situations, and even monitoring it takes a lot of time. It’s a lot of fun and definitely worth it, but it required more resources than originally expected.

Overall, developing an AI-powered job search has improved the user experience and lifted pretty much all our key performance metrics. Compared to what we had before, it is nevertheless so much better and was definitely worth it. I can only advise you that if you have a career website that attracts a good amount of traffic and you have a substantial number of job openings to fill, do invest in your job search. It makes a difference!