Case study: AI CV creation tool for Saudi marketplace
AI is not only a revolutionary tool, which can help bring innovation into products, but it can also be a key to improving users’ experience and eliminating user pains. In this case we will uncover how AI was implemented in Bab Ajeer, a product built by Boldare for Ajeer, and how it helped reach product KPIs in a quick and effective way.
Table of contents
The client
Bab Ajeer is a product within the Ajeer portfolio, a set of products tasked with allowing companies and individuals to sign B2C and B2B contracts for the temporary transfer of labor within Saudi Arabia. Whilst some areas of Ajeer focus on temporary leasing of employees between companies, others allow job seekers to sign contracts directly with new employers for temporary contracts. These are very frequently used for formalizing temporary work during eg. Hajj season in Saudi Arabia, a time of the year which sees thousands of individuals temporarily changing their place of employment, often associated with jobs related to transport or hospitality.
Bab Ajeer is the marketplace of the Ajeer portfolio, allowing job seekers to create their CVs, publish them on the market for potential employers to view, as well as apply for open vacancies using those CVs. These serve as the core profiles of the users and contain all the most important information about their past work experience, education, and skills, among other details. Based on these applications, vacancy owners can decide to accept them by sending the candidate an online contract proposal.
User pains
The marketplace, similarly as in many other platforms of this type, replaces the traditional profile with a CV format, which the users fill out in order to effectively use the platform and apply for vacancies. These are very important as the CV itself serves as an application and the quality of the CVs greatly contributes to the success of such a marketplace. Similarly to social media networks, if users neglect their profiles and do not fill in key details, the value of the product can drop significantly as its core value lies in the users’ input.
For users the need to fill in their profile, adding many details regarding their education, experience, etc. is a troubling task which not only requires extensive work, but also a lot of time spent on filling in fields. Anyone who has ever taken part in applying to job offers and creating job posting accounts can relate to this user issue of being required to fill in the same details on numerous websites. This is all despite most individuals having a file version of their CV on hand.
In fact, in Bab Ajeer many users were seen in Hotjar recordings spending over 30+ minutes meticulously filling in their in-app CVs by hand. They were also frequently seen pasting in details from, what is most likely, other files or accounts. What’s more, there were multiple recorded cases of users sending their CVs to customer support with a request for the team to fill in their in-app CVs for them.
Our solution
Upon conducting our regular analyses of competitive platforms and new technological advances, we began researching the topic of utilizing AI in Bab Ajeer. This came with a surge of ideas and possibilities. What was, however, key, was the identification of low-hanging fruits, ie. areas where we can most easily and quickly bring the highest value to the product and its users. The case of the filling of CVs was one of the highest priority user pains, which ended up being one of the easiest to implement as the first AI feature in the product.
The planned solution was an introduction of a CV importing tool, which, by use of AI, could interpret a user’s CV file and fill in their profile based on that reading. This would not only save users tens of minutes spent on filling their in-app CVs manually, but it would also decrease user frustration and annoyance caused by the repeatedness of this action, as well as bring value to the product by means of higher quality CVs filled with more data, and subsequently, more applications sent to vacancies and more contracts.
A/B testing
While the preparation and design of the tool was pretty straightforward and based on extensive user research, analysis of good practices related to the UX of AI tools, as well as solutions created by other platforms, an important question arose - how will the target audience interpret AI features - with fascination and desire, doubt and suspicion, or maybe neutrally? This is an important question to be asked, as the topic of AI is surrounded with mixed opinions and the public’s attitude towards the issue is not only unclear and not yet fully studied, but also continuously changing and dependent on factors such as age, nationality, and occupation.
In order to make a data-driven decision on whether to promote the feature as AI-based or simply introduce it as just another feature without AI-related copy in the banner, we planned an A/B test to see which version of the banner on the CV page would perform better and result in a higher conversion rate. Such an introduction of an A/B test is not a very high effort on the frontend, yet it can yield high quality results which allow us to make decisions based on real metrics.
What’s more, the data from such a test can help identify the approach for future AI-related features. Some platforms add such new features without mentioning AI, whilst others add expressive icons, animations, and sometimes even launch entire campaigns for them and utilize personification to give the feature a name and face.
The results of the A/B test indicated a 18% higher conversion rate in the non-AI version of the banner, which debunked the original hypothesis and allowed the client to make the decision to keep the non-AI version, but also select a similar approach for future AI features.
Next steps
Events in multiple areas of the feature are continuously helping the team to monitor the effectiveness of the feature and changing conversion rates with time. In addition to this, logs are tracked to monitor the cause of errors during importation and the ratio of accepted to rejected imports. This is thanks to the addition of a preview page displayed to the user, guaranteeing a fallback for dissatisfying results of the import, allowing the user to go back to the previous version of their CV without losing the initial version.
Such a preview page was quicker and less expensive to implement in opposition to eg. an undo button, whilst allowing users to make the decision and not feel worried by an incorrect import. This also allows us to track the amount of accepted and rejected imports after each subsequent release of the feature with improvements.
The next step with all AI-based tools is the observation of the feature’s accuracy and efficiency. Accuracy can relate to the satisfaction of the user, which we track using a survey where users can rate their import and leave feedback regarding their experience, while efficiency is tied to the time required for processing the prompt and the amount of errors experienced by users on production. By means of backend improvements this time and error-proneness can be improved.
Conclusions
The introduction of AI to any new or existing product is a challenge and is associated with many risks, including those related to data safety and privacy, its effect on speed, users’ experience, as well as complexity of working with new technologies. It is however crucial to understand that AI can not only be a fun and exciting gimmick, but it can also be the key to solving some of the biggest user pains with the least amount of effort and least expensive execution. We are consistently learning from our mistakes, similarly as is AI, improving based on our observations and results.
Check out the visual case study for this project on Behance.
Share this article:
