RecBuzz stands as a must-attend event for professionals within the tech recruitment sphere. It delivers dialogues surrounding key themes such as performance-based pricing, workforce deficits, and the ramifications of AI integration, among others.
Yana Levchenko, Country Manager for France at Jooble, participated in the RecBuzz conference held in Barcelona on April 16-17 and shared Jooble’s experience with data, how the company uses it, and how sometimes it can be harmful if misinterpreted.
“I will tell you a short story about how Jooble–a proudly data-driven company–tricked itself with its own data. I should say that this isn’t a success story yet,” – Yana said. “At conferences like this, it’s much more important to raise questions and share thoughts with industry peers than to provide immediate answers.”
Yana suggested to the audience to imagine they’re looking for a job. Usually they see many job advertisements, all containing the job title. Some will also include information about salary, a brief job description and so on. The key point is that there are so many similar job advertisements on each site.
She asked, “Have you ever wondered how you decide which job to apply to when crawling the search results on every job site?”
In Jooble, we realized that each job posting could grab a job seeker’s attention for three to ten seconds. Is this enough to determine whether this job is suitable for you? We don’t think so. We also see that practically every job site now has a significant percentage of mobile traffic.
It means users may show the same behavior on job sites that they do on Instagram: just swiping posts with no particular attention while standing in line or riding home on a bus.
We’ve worked hard to increase this time span, but we realized that our hands are tied by how the human brain works. We realize that our job seekers make decisions based on fast thinking and implicit criteria. Moreover, these implicit criteria are usually not only for us but for them too.
In Jooble, we understood that to build a great product in job search, first we need to understand these implicit criteria. Second, and most importantly, we need to predict those criteria. Things wouldn’t be that bad if only job seekers were eager to help us help them find the best, most appropriate jobs.
Read also: Job Benefit Tags and Estimated Salary: The key to boosting conversions to 17%
So, we know that less than half of users who come to Jooble for the first time are looking for a particular profession, like hospital nursing, bank accounting or whatever. Another significant portion of our users are looking for jobs, defined by criteria such as salary expectation, whether the job is remote or in-office.
There’s another interesting part of users who are looking for “any job.”
For everyone working in a job niche, it’s not an open screen. The phrase “any job” doesn’t really mean “any”.
When we see the word “job,” for us it’s a search keyword without any semantic load. However, our users can hide so much information behind it. They may hide their salary expectations from a potential employer. They want to know the benefits the employer will offer, and each person has such expectations.
Here’s how it works on a daily basis:
Jooble collects millions of data points, which we use to analyze the search keywords that are entered:
Everything is important. So, the more data there is, the better. Such an analysis helps us build a system of recommendations based on similar behavior patterns among users who are looking for jobs without specifying any particular criteria.
We can see that at one point we have this data. We know how to collect it, analyze it and segment it, and we have two primary criteria to segment the audience based on the data: the country and search keyword.
Read also: Jooble’s Design Upgrade: Seamless Navigation for Job Searchers
As a result, we created segments to identify popular preferences and started to boost the jobs that would probably be interesting for job seekers looking for “any job.”
If we look at the UK market, we know that people looking for any job usually click on jobs titles like those shown on the tag cloud in the image below:
We also have a similar tag cloud for each profession and criteria-based search.
This approach really works. After applying this model, we got positive results in France, for example, in terms of interactions with vacancies. In Germany we got incredibly good results in terms of conversion rate.
Why wasn’t it a success story?
We were so happy with our positive results that we ignored something important: Even though every job recommendation was good on a country level, the performance could be different depending on other criteria, such as the region. For example, everybody knows that in every country there are industrial areas, business centers and touristic places. We hypothethesized that people were probably looking for different jobs in those regions.
We did some preliminary tests, which showed us that in London people looking for any job usually tend to click on Amazon Flex jobs or jobs for waiters because everybody knows how many restaurants there are London. In Manchester, people click on customer service or HR jobs, and in Bristol they click on airport and porter jobs.
This was a very interesting insight for us, and we blamed ourselves for not realizing it at the beginning. It’s an important lesson, and we expect to implement this new approach and improve the results such as these:
Read also: From data to decision: New Jooble Tools to Improve Your Campaigns
1. Big data and models built on it work.
You may have geniuses in your team with a perfect sense of business and a perfect sense of product who may invent a killer feature that will bring success to your business and your client. Nevertheless, it’s better to protect yourself with data and base your expectations and predictions on real information.
2. The more data you have, the more important it is to segment it.
In other words, the more data you have, the more likely it is that you’ll omit something when you aren’t analyzing the information from different perspectives. You need to look at the information from different angles in order to catch the tiny details that will help empower your business, help your clients and help jobseekers who always need our support in finding the best-fitting job.
3. Cool results can hide opportunities for further growth.
It isn’t the time to stop when you’ve just created a new reality for your business with positive results. On the other hand, it’s high time I thought about what else we can for our business. How can we enhance it, and what are the hidden opportunities that we do not see right now?
To build a great product in job search you need to analyze regional economies and enhance them with AI models. You could start with AI models, too. The order doesn’t matter.
The most important thing is to analyze the information. Jobble will scale all of this experience in its top markets–Germany, France and the United Kingdom–in the very near future, and even more markets later on.
Learn more about understanding user intent with Jooble from Yana Levchenko’s presentation.
