Xsolla CEO says firing using an algorithm-based solution protect minorities | Hacker News Reader

Xsolla CEO says firing using an algorithm-based solution protect minorities

gameworldobserver.com

42 pointslogronoide41 comments

This article doesn’t capture the flavour of what happened here.

Here is the letter sent to staff.

You received this email because my big data team analyzed your activities in Jira, Confluence, Gmail, chats, documents, dashboards and tagged you as unengaged and unproductive employees. In other words, you were not always present at the workplace when you worked remotely.

Many of you might be shocked, but I truly believe that Xsolla is not for you. Nadia and her care team partnered with seven leading HR agencies, as we will help you find a good place, where you will earn more and work even less. Sasha will help you get a recommendation, including the one from myself. And Natalia will read you your rights.

Once again, thank you for your contribution. If you want to stay in contact with me, please write me a long letter about all your observations, injustice, and gratitude.

From the interview and this letter the CEO sounds like a really scary and entitled person. Stuff like this isn't just bad for business but when you see this kind of instability and recklessness in the leadership the company often fails down the line.

I gathered that he finds his business special and lucrative, especially by his native city standards. I think he presumes that even a "bad" employee at his company will find a nice spot in a more ordinary one.

He struck me as the kind of people who makes management errors, sees the company suffer for them, and blames anyone and everyone but himself for it, usually for extremely vague reasons like "not working hard enough"

Is there any study on what could be the false positive rate for models of this kind?

Or rather, how many productive employees are scared of being labeled unproductive and given the quick boot in the next round?

Who wrote that. The CEO? Sounds downright sociopathic.

I hadn't heard of the company before, but now I'll know to steer well clear of it, should the occasion ever arise -- thanks.

I have implemented payments using their platform. Even before hearing about this, I would not use them again.

what are the alternatives to use for their niche (merchant of record for game micro-transactions) ?

I don’t know of one. And maybe that was why Xsolla saw 40% growth until recently. But you can only go so long without fixing problems with your product before people stop using it.

I am also coming from the perspective of someone who has implemented Stripe in a SAAS application. I can see why Xsolla will appeal to game shops, but compared to what I can do with Stripe, I wouldn’t use Xsolla.

From a general strategic analysis approach though, based on my experience working with Xsolla:

  - if you already have a good customer service and know your users and fans, you do not want to depend on someone else’s customer support team. They won’t engage with your users the way you will, and now your customer service have to work through Xsolla customer service
  - if you need to customize the skin and experience of the payment flow, you cannot have someone else be the merchant of record
  - if your team already has competencies in writing web apps and operating infrastructure, you can do a better job than Xsolla at implementing virtual currencies, item shops
  - if the people paying for the games, virtual currencies, and items are not the gamers themselves, then the many payment integrations that Xsolla offers do not matter as much
  - if you are mainly operating in a few countries, then the international payment integrations Xsolla offers are not really much of an advantage

"Sasha will help you get a recommendation, including the one from myself."

Once this letter is public, not sure how valuable that recommendation will be.

I wonder what would the body of this recommendation be? "Worked poorly by our standards, but maybe will be more enthusiastic in your company. Otherwise great dude (AI-rated at 78% "great" and 87% "dude")"

I worked at a company where the CEO shared a chart of the top 10 most active people in company Slack, implying praise for the most active Slack participants. Everyone got the message that more lines in Slack == more visibility with CEO.

From then on, Slack became unusable because people were basically spamming Slack to keep their stats high.

It got to the point where people weren’t even typing full sentences, but instead sending a few words in every line or just posting multiple lines of emojis in response to everything.

1. Write bot to send slack messages. Maybe a mix of random HN articles, non-detailed responses, and DMs to another bot.

2. Go do your real work. Or slack off. Whatever.

Ministry of Work: Being productive is unproductive.

The last sentence sounds sarcastic and really really bad.

Busy != productive. This measures busyness and not productivity.

I wonder if the analysis can be trivially tricked by using a bot to post random crap using those services' APIs.

I'm always going, step away from the computer and think about what the fuck you are doing.

You omitted a dozen mistakes he made in the original text, and at least as many negligences like downcase and misuse of commas.

Firing people based on algorithms or arbitrary statistics doesn’t protect minorities.

It protects the people who know how to game the system. Those who understand how the metrics are calculated actually love these systems because they can use that insider information to make themselves look disproportionately good by going through the right motions. If they know they’re being ranked by the number of Jira tickets being closed, they might create 5 tickets for every piece of a task instead of 1 ticket with a checklist. Or if they’re being measured on lines of code committed, they’re going to be doing a lot of unnecessary refactoring and rewriting to keep those numbers up.

Indeed, apparently the developers of this big data employee ranking system are internally opposed to letting employees see their own scores. From the article:

> The company also planned to implement the so-called “digital mirror,” so every employee could learn what AI thinks of their work and engagement. However, the development team wasn’t enthusiastic about this idea, so its rollout is pending.

Of course the development team wouldn’t want other people to be able to see their AI ranking score or be given a chance to learn how to change it. That privilege is reserved for the developers of the AI ranking system, who no doubt score very well on the system.

It’s also bizarre that the CEO would publicly admit that his planned feature to let employees know their score was being held up because some employees didn’t want to release it. Why air internal drama like this? Why blame the employees? This CEO sounds like he’s trying to wash his hands of every CEO responsibility at every chance he gets.

“Tell me how you will measure me, and then I will tell you how I will behave. If you measure me in an illogical way, don't complain about illogical behavior.” – Eli Goldratt

Ite actually quite the contrary. Using AI as a tool to discriminate and then defer to a magic blackbox is one of the most profitable AI areas. Look at automated moderation for example.

I bet they have some visualization to show in court in case of a discrimination lawsuit. Also why bother with discrimination if your goal is profit increase?

It seems to me this guy should be fired. But good for him if he can disguise his inabilities to manage the transition to full remote. As a employee I would flee such a company.

Wow, usually unpopular decisions were outsourced to consultants who then would give back what the CEO wants to hear and do the deed. Now, that can be done for cheap via something mysterious called "the Algo-rythm" ;) . Bad times for consultant firms incoming.

Where does it say the algorithm protects minorities? Sounds like it does the opposite. Unlike humans, it is not "overly cautious" about firing minorities.

> Agapitov went on to say how companies nowadays protect minorities and can be overly cautios if they have to fire their representatives, which can make these employees almost immune to layoffs. “Our algorithm-based solution is as unbiased as possible.

It protects the company when you fire minorities, because you do it base on data. It's like proving that you've bought stock fairly (not according to insider information) by showing research.

>The list includes writing and reading articles in the internal Wiki, creating and closing task tickets, as well as dashboard activity and participation in internal meetings.

>He claimed that if a person’s “digital footprint” is not visible, they shouldn’t work at Xsolla.

Sucks to be the sysadmin I guess.

Why? Optimal position to gamble such a system. Everyone else needs browser addons to create fake activity with macros :D https://addons.mozilla.org/en-US/firefox/addon/selenium-ide/...

This is the "Lines of Code" metric of IBM reincarnated as social activity.

Servers working and being well setup doesn't look great in Jira statistics.

First step of resolving a ticket should be to create 10 tickets... now for each ticket create a new ticket to verify the other ticket has been done correctly... So work at Xsolla will become very slow but hopefully very predictable.

It’s going to be interesting to see if this gets used elsewhere. Something tells me we’ll probably see a lot more of this happening in the future.

From the article: 'The company’s decision regarding layoffs was inspired by Netflix, Harward, and other organizations that lay off 10% of the most unengaged and ineffective employees on a regular basis.'

That's the claim by Xsolla. But Netflix specifically say they don't do that:

"We have no bell curves or rankings or quotas such as “cut the bottom 10% every year.” That would be detrimental to fostering collaboration, and is a simplistic, rules-based approach we would never support" from https://jobs.netflix.com/culture

Say they don’t do that. Amazon also say they don’t do that. But it do.

We don’t cut the bottom 10% each year. We just ask managers to identify 10% as eligible for layoffs.

I've interviewed a couple times at Netflix, and they mentioned they do cut aggressively cut under performers every year, but not by a percentage.

I see a lot of comments about 'gaming the algorithms', but how is that different from a person?

Doing shady commits/tickets in a way that satisfies the algorithms, or play best-buddies with my supervisor would still result in the same outcome; I don't get fired.

A human could catch on. A algorithm has to be double-checked by a human in order for someone to catch on, and in generally people just trust machines to be right. Even if they initially double-check them, they do that less and less over time.

And if the person doesn't understand the algorithm, it's even worse.

Imagine this same kind of person the master of a plantation or 1800s factory or as a Dickensian villain. It fits.

Reply on news.ycombinator.com