How can a chatbot suggest job offers on


Contact with brands on Facebook has become as natural as writing with a friend. I have a problem. I'm just writing a message. What if we get to a recruitment portal? This is what the company faced. Candidates wanted to look for interesting job offers through Messenger. So maybe a chatbot will solve the problem? Let's check

Service design, research, workshops
Project completed, April 2018

How to design a chatbot to support job seekers?

The conversation model seemed most natural in this case. But how to make it attractive to users? The conversation is finally governed by its laws. Users can ask for the same job offer in very different ways. What if the bot answers incorrectly? We're ready for frustration. So there's a challenge — let's see how to use the job search engine, used on the website and in the application, in a new communication channel — chatbot.


Let's start with questions

There are no stupid questions. And only when we answer the basic ones, can we start to go into the details. Because what kind of virtual assistant can our virtual chatbot be? That's why we started to wonder:

  • When will you need it?
    In which situations will it be used (context)?
  • Why would you need it?
    What will be his role and what problems will he solve?
  • Who will you need?
    What kind of personality a bot is supposed to have?
  • Who are we designing for?
    We are deepening the personnas, i.e. people for whom this solution will be useful.
  • How will the bot work?
    We are developing communication logic and bot language.


We divided the work on this solution into three stages. Each of them lasted two weeks. And their main focus was the workshop. It was then that, together with team, we answered questions and designed further functions and features of the bot. Every day the bot learned how to solve new problems. He was growing up.

Okay, we have a base. Now we can take care of his personality. Well, what should it be? Serious? Curious? Or totally laid back? After all, looking for a job is an emotional, stressful moment — which was also confirmed by research. As with any service, the first impression is most important. Because if a bot sends a wrong offer or, what's worse, starts to flood you with complicated questions — it will only arouse frustration among the poor recruiter. And it may strongly discourage him from going on. As if he wasn't stressed enough? We didn't want to add to that.

Selection of the candidate

Just like choosing a partner, we wanted to feel that he was the one. So, candidate number one was Italo. Italian roots, clear and original language. Sounds perfect! And if we add that he can talk about the job search process as if he was creating the perfect recipe for the best dish... Madonna Santa!

Extreme solutions make it easier to evoke reactions from the test subjects and better verify the concept. That's why we decided to take a risk with Italo and test it with the users.

Agnieszka Wilke
Agnieszka Wilke

Okay, the bot was selected, but which test method to choose? After all, we didn't yet have a developed an advanced conversation logic. We have opted for the Wizard of Oz. How does it work? We ask the respondents to complete a task, and we play the role of the chatbot who assists them — we use the script prepared earlier. In this case, however, we were able to monitor the course of the conversation on an ongoing basis and immediately catch up on which path of the conversation is not natural or needs to be changed.

The Cogision team playing the role of a chatbot during the Wizard of Oz.

It's not a funny thing, looking for a job.

— Participant in the Wizard of Oz test

Let's choose again

Oh, Italo! We had to say goodbye to you although it broke our hearts. Respondents did not take it positively. They were irritated by his cheerful tone and language — he used too many metaphors. But for us, it was a valuable lesson. The respondents gave us a lot of guidance as to where to go further.

In the following iterations we decided to tame our bot a bit. We have opted for simple and understandable language. The tone of the bot's speech was supposed to be emphasized only by emoticons. It was time for a more balanced candidate named Radzimil. He referred largely to the advertising campaign, present in the media at that time. Let's see how it goes this time. Testing time!

The research has helped us better understand the expectations and emotions accompanying the job search. This allowed us to create a suitable bot personality.

Conversation logic

If a job interview conversation is to work, it must be logical. Because although a bot would speak the simplest language, the most important thing for a candidate is to get things done. As soon as possible. So the bot must always react accordingly.

Okay, we have directions and guidelines from the first research. What now? This time we used the Chatfuel. This allowed the respondents to enter the real experience of talking to a bot on Messenger. That's how the bot will work eventually. Great. We're proceeding with the study. Respondents' conclusions? They were clear. The conversation is to be quick and efficient. And finally, they're supposed to get the best matching offers. Simple.

And this was the clue. The Chatbot, with all its advantages, wasn't as fast as it should be. And the offers he presented proved to be inadequate. Hmm, not too good. What to do about it? After discussions we decided that the bot will talk to users in loops. What does that mean?

  • First loop, conversational, longer and classic. The bot asks questions, makes sure he understands the phrase correctly, and finally displays basic job offers.
  • The second loop, we moved part of the conversation to the classic form, which is filled in by the recruited person. He enters his own location or position and selects any filters he wants to apply. It's easier and faster than talking to a bot.

The workshops we could attend were the place where we had the greatest contribution to the design of the tool. The hours devoted to working out the ideal personality and scheme of Chatbot's activities resulted in Radzimil.

Patryk Müller
Patryk Müller
Specialist for Communication with Users in

Did Radzimil pass the exam?

Sure! Radzimił completed his training in Cogision in April 2018. He became a full-fledged virtual assistant for job seekers. He learned how to notify users of new job offers, redirect inquiries to customer service and talk about himself. It was finally implemented a year later and started working on the fanpage in 2019. Our pride!

Chatbot Radzimil :)


What we learned

Research is power! Our work on the scheme of operation and language lasted until the last days of the project. Each new function generated a few new paths of conversation. And these meant further tests, and again, and again.

Every day finding new blind alleyways or incomprehensible statements has become our habit. Very well!

The project has clearly shown us that MVP is not just about creating a working product. And above all, it is about creating a product that meets the most important needs of users. Obvious? It seems so, but this is what is often forgotten when creating new products or services. Let's test and improve!