19 October 2017Katya LebedevaSupport Lead

In the last article, I gave a detailed explanation as to why hiring more agents doesn't help in the battle to answer users' questions as quick as possible. Nonetheless, there is still some turnover in the Support team. An agent might move to a different department within VK, someone might gather their own team and sail off into the open ocean or someone might decide to dedicate their life to meditation and set off in search of Shambhala. So, that means that from time to time our team needs to find fresh blood. And with that being the case, we need to find it somewhere. But the question remains as to where?

The story of how VK Support agents are found deserves its own story, and the time has finally come to share it.

In the last article, I gave a detailed explanation as to why hiring more agents doesn't help in t.., image #1

Support history 101

A long time ago, in a galaxy far, far away, there were already some targeted ads on VK, consisting of a couple of promos under the left menu. It was there that chosen (in the truest sense of the word) users saw mysterious Buddha images which asked, "Got zen? Then get over here, we need you". This was in the long-gone year of 2011. We searched for the right people to join our team using many different methods, and this was just one of these experiments.

The first members of our Support team came from a primordial soup of volunteers that organized themselves around VK and began to help other users (by the way, a few dozen of these first agents are still with us on the team). Long after came the search for new agents employed the use of targeted ads, word of mouth and the /jobs section.

Alongside posting vacancies to the /jobs section, we created the Support Team community. It helped deliver important news and just share our idea of what Support should be like (and at first, just to let users know that it exists). Screenshots of our answers went viral and spread across VK communities and beyond. And as our glory grew, so did the number of those who wanted to join our ranks.

Glory be to glory, but to say that those screenshots really played into our hands well in regards to hiring new agents would be wrong. They provided a distorted view of the work that an agent actually does. Candidates had the impression that all that agents do is have pleasant chats about eternity, joke around and have fun. However, these chats only account for about 3% of all questions with the rest of them consisting of serious work that requires diligence, perseverance and a keen desire to constantly learn. It was only after being hired that the new agents truly faced this reality.

And so, we've never suffered from a lack of candidates. Actually, we've had too many. And at one point, it became a real struggle. How does one pick out the right people from thousands of candidates.

In the last article, I gave a detailed explanation as to why hiring more agents doesn't help in t.., image #2

How do we choose candidates?

It's our job to remain impartial and sympathise with all users equally. When making a decision, the agent has to set all emotions aside and work according to the rules of the Site and Terms of Service, use common sense and understand the consequences of every choice they make. Learning to take a philosophical view of things can prove useful, as it will make it a lot easier to live happily, even when someone on the internet is wrong. Back then, we needed people who understood this, and we still need them now.

In order to find the best, we subject the candidates to all sorts of tests. The first step is to sift out candidates for formal reasons: applicants under the age of 18, foreign citizens without necessary documents (as we hire by the book), as well as all those who could not come up with at least a couple of coherent sentences about themselves. Those who pass the first step are sent a few test tasks which help us evaluate their aptitude for finding answers to non-obvious questions, formulating their thoughts well, bringing people over from the dark side and creating.

Traditionally, we include an unexpected question in the test. For example, back in the day when I was answering such questions, I got this: "Hello, Support! So the thing is that I invited a girl to my house, but I don't know what movie we should watch. Could you give me some suggestions?" I had to describe in detail the principle of searching through thematic communities and tell what to look up: the "romantic" tag (movies which can cause tears of joy and set the mood for a lovely date night) or "horror" tag (movies which might make her cuddle up to you in fear).

We evaluate answers to such questions with utmost attention. It's through them that we are able to understand how a candidate would perform in such non-standard situations and manage their way through uncharted territory.

As in all other internal spheres of an agent's life, there prevails an openness that comes when selecting new team members. Each agent is free to take part and express their opinion about the answers of any candidate. However, we are only talking about the answers. The candidates are assigned numbers and their names and pages hidden. This way, not even a candidate's good looks would be able to sway the evaluation.

If after the first two steps a candidate is not chosen, extra questions are asked. In the end, only the best of the best remain. We bring them to the Singer House, name them younglings and start teaching them.

In the last article, I gave a detailed explanation as to why hiring more agents doesn't help in t.., image #3

What's the outcome?

For a while, the system worked quite well. We were able to find new talented individuals and provided them a place within the Team. Everything would have been great if there just hadn't been so many applicants. In October 2013, we took a look at the collection of applications for the job and were overwhelmed. Given the minimal amount of staff turnover and a fixed number of jobs, the applications for the vacancy that had piled up, with the most modest estimates, would have kept us set for about the next 100 years. It was then that we decided to remove the vacancy from the section.

Potential agents were tested on a first-come, first-served basis. Every few months, we sent out assignments to the next hundred candidates and from all of them, selected no more than five to become our students. Needless to say, the stack of thousands of job applications melted away slowly, so that meant that many had been waiting years to get their assignments. And then they forgot that they were waiting. They were surprised when, many months later, they were actually given a chance. We, of course, planned to test how patient our future agents could be, but this has already passed the boundaries of what could be deemed reasonable. Something had to be done.

In the last article, I gave a detailed explanation as to why hiring more agents doesn't help in t.., image #4

What did we do?

In the beginning, we quietly hoped that the situation would just sort itself out. By our calculations, with the amount of time that had passed, most of the candidates should refuse to go through some strange testing procedures. Who waits years for a chance to get a job? We underestimated ourselves. Someone reacted to our sudden offer with an expectable refusal, but most still wanted to try their hand at the position.

And so, 3 years later, we still had 1800 applications for the vacancy which had been yellowing with age, but perhaps hadn't lost their relevance. 1800 people, many of whom had long forgotten that they had submitted an application, now had to go through the long and exhausting (for both sides) testing process. To simply throw out these applications and start a new search was unthinkable. Surely there had to be people suited for the job, and we had to at least try to find them.

We realized it was time to modernize the process and figured that VK's giant platform probably had all of the tools we would need. And before long, we found the tool that would do the trick. We pulled out look alike, the ace that had been up our sleeves all along, to use its algorithms which would themselves choose individuals which would best suit our team.

Look alike (or a similar VK audience) are users which have similar behavioral characteristics to those who are already your clients (and in our case, Support agents). The algorithm finds a similar audience, taking into account gender, age, interests, education and other aspects. The types of communities they are in, the types of posts they like and other factors were also taken into account.

Kill your darlings! We gave the machine the right to decide. We used the profiles of our current agents as the basis for "look alike" and got around 200 "similar" candidates. Out of 1800 applications, 200 was a number that could actually be used for testing. And so, test tasks were sent to these two hundred lucky few. And they did not disappoint. From these lucky few, we found three times as many younglings than we usually would from a similar "raw" sample.


And there you have it! The tale of our search for VK Support agents is wrapped up nicely with a happy ending. A few more people found their place in our Team and spend their days helping users. We have cleansed our karma and finally sorted out the obsolete legacy of 2013. We're not planning to return the vacancy to the /jobs section quite yet (I don't think we need to step on the same rake again), but if you read the articles about Support and catch yourself thinking that we are perfectly suited for each other and it's the job of your dreams (at least for the next few years), then you certainly will find your way. Nothing escapes the attentive eyes of a future agent.