How to attract talent as a startup? A text mining case
One challenge startups are facing nowadays is to hire talented people. Understanding what attracts candidates and their expectations would help to better tailor job offers and therefore increases the interest of candidates. We applied one of the machine learning techniques, namely text mining methods to figure this out.
Candidates applying for a job or an internship at a startup are appealed to the unique environment that only a startup can provide. Working in a dynamic and challenging team, meeting people and learning new skills are according to our research what makes startups appealing. Indeed, candidates are mainly looking for building human relationships and personal development. If you plan to hire any time soon within your startup and desire to raise the interest in the available position, these terms should be stated somewhere in the job offer. In this post, we will detail how we got to this observation, during the Acredius hiring process, using text mining techniques.¹
The startup environment:
We gathered the answer of each candidate to the following question:
“Why do you want to work for a startup?”
We used these alternative data to build a word cloud and a bar chart.
They display the most occurring words from the answers given by candidates. In both, the word “environment” has a prominent place. It reflects the importance the candidates granted it.
In the same topic, other words seem to be important for candidates such as:
team, people, learn, skills, responsibilities, dynamic, challenges
Now that we have the most important words for candidates, let’s analyse how they are connected to each other.
What do they mean by “environment”?
To grasp the relationship between these words we computed the term correlation and a dendrogram. The first is focusing on the word “environment” while the second gives a general picture of the association of words.
The correlation indicates which terms are mostly employed to describe the term environment. It turns out the human dimension is the most important part for the candidates. The two most correlated words are “people” and “team”. Candidates want to be part of a team and meet people. They are looking for an environment that stimulates or encourages interactions.
All other correlated terms can be summarised by personal development. Candidates desire to work in an environment where they can first express their ideas but also be challenged. It will push them to learn new skills which will improve their initial ideas and ultimately, they will make an impact. These last two points namely developing their skills and making an impact also show their desire to be an active member of the company.
These findings are at first confirmed by the dendrogram and at second expanded.
The words “environment”, “people” and “team” form a group by themselves. It supports that the main thought of a startup environment pictures the relations the candidate can build through the time spent within the company. Similarly, “ideas”, “skills”, “part” and “impact” are linked together and reinforce our opinion that personal development matters for candidates.
A major group is composed of “dynamic”, “challenge”, “different”, “innovative” and “great”. This whole group is then linked to the word: “motivate”. It is an easy statement to say that this is another facet appealing to candidates. They are looking for a working place that is in constant motion: today is completely different from yesterday. Candidates want to look beyond what is currently functioning and do things differently.
Interestingly enough, the word “startup” is not individually directly related to another word. It is associated with the whole body built by all other words. As it could not be specifically assigned to one and encompasses all these concepts.
Now that we have the associations of the word let’s make sure these are the most important arguments to candidates. To do so we looked even broader by applying to the full set of answers a topic modeling algorithm named Latent Dirichlet Allocation (LDA).
As expressed in the LDA description, it does not implicitly give the name of the topics but provides us with a list of words that fit respectively in each topic. From these 3 lists of words we can infer the topics are:
- Building relationships
- Personal development
- Dynamic and innovative nature of a startup
What about candidates’ personalities?
In addition to the answer to the question we also asked candidates to answer the 16Personalities test and report their respective personality code. For each job offer, we evaluated the number of candidates per personality type. One personality appears as a common result among all: the protagonist. Whatever your job is at a startup you have to believe in the idea/goal of the company no matter what the external opinions are. It is exactly what the protagonist people are known for.
Almost every position has a distinct category of personality that is not present in the other.
The Executive personality is only found within the data science applications. It is in line with the position since they are described as people that enjoy creating order and organising.
On the other hand, the Advocate is described as very creative. No wonder why they are only present in digital marketing applications.
In the same idea, we extracted the average number of words used by personality type to answer the question stated above. We observe that on average candidates belonging to the protagonist personality are by a consequent amount of words the most talkative. They are followed by the mediator. It is interesting to note that both belong, based on 16Personalities, to the diplomat class.
Concerning the least talkative, the executive and defender are on average using respectively eight and four times fewer words than the protagonists to answer the same question. Both belong to the so-called sentinel class. Did we just point out another characteristic of the sentinel class?
The sample used comprises candidates’ answer to the question
“Why do you want to work for a startup?” and the result of the personality test.
They applied for 3 different job openings at Acredius, an SME crowdlending platform based in Zürich, Switzerland. More information can be found below:
*A Special thanks goes to Cedric Higel, our data scientist intern who played an important role in producing and writing this article.