Bonuses, part 2: How

Your business has had a good year and there is some surplus that the owners decided to distribute as a bonus among the team. So far so good. But how do you actually do it?

Finding a bonus schema that is as fair as possible and not prohibitively expensive to implement is not an easy task.

We at OSI have made our own such schema and are constantly looking for ways to make it better. Since I joined OSI about 4 years ago we do an annual exercise that I lovingly call “sweetening our lives” – each of us distributes a certain amount of fictional candy between everyone else. Adding on top some management voodoo, and through a bit of Excel jiujitsu, the annual bonuses are determined.


We started with the idea of getting a fair assessment based on each employee’s opinion for his/her colleagues. So, we started with innermost team assessments. We invented a single dimensionless metric called Lukche[1]. Everybody got assigned 100 Lukche to distribute inside their innermost team as an appreciation of their coworkers’ work.

The results were encouraging so on the next iteration, which was the next year, we created two metrics – Lukche and Violetka[2]. Lukche was to be given inside the immediate team, Violetka to be given outside of the immediate team. Here we also introduced a third type of candy that only managers got to assign usually to their reports for things that they see from their vantage point (the voodoo part).

In 2019, on the next iteration, we tweaked the metrics even further. Now the candy given are M&Ms and Gummy Bears. M&Ms to be distributed for any help that the employee received directly and Gummy Bears – for anything else that improved our work environment in general. Everyone could give Gummy Bears to people that he/she thinks have contributed most to the success of the company as a whole. No more hard boundaries who can get each candy. Why we changed the rules? Because at OSI many people work in multiple teams, and many people’s job is to work with everyone. And again managers received their third type of candy to express their assessments of who should get a higher bonus.

After all, candies are collected, I put them in one single spreadsheet, do some normalizations, add the employee’s annual compensation, then the business does some team multipliers[3], then finance department determines the total amount to be paid out as bonus based on the business’s annual performance. Then through the Excel jiujitsu (example) and each employee ends up with a specific amount to be paid out as a bonus.

How well does it work in practice?

Well, it does work. And since we are relying on an algorithm[4], we are for sure avoiding some of the more prominent forms of biases.

Some teams love it, some teams not so much. Some teams do have clear stars in them, other teams decide and distribute all their candy equally. Some teams are more self-absorbed than others. Some employees are deliberate, others are somewhat more random, yet others distribute candy purely on sympathy. Some employees are eager to submit their evaluations, some need a little nudging.

The benefits

Today, with 3 years of experience behind us, I can say that the efforts are worth it. The benefits of the rating system show us that we are on the right track. We already have more involvement, greater motivation for teamwork and we would argue better business results (although those are very hard to attribute directly).

Yes, for sure it is messy, it is uncomfortable for some people, and it yields surprising results at times[5]. Some employees get only a fixed minimum bonus but it also rewards the best employees proportionately. It cancels out favoritism when enough people participate. And above all, it ignites many conversations between employees and their managers, and many, many more among managers. The system pushes people to discuss with managers various initiatives to improve their own performance, such as presenting, mentoring, participating in different kinds of teamwork. In this way, growth is achieved organically.

Upgrading the system

To plot the contributions curve as accurately as possible this kind of assessment should happen on a daily basis. This, of course, is not immediately achievable for many reasons and we decided to start with annual assessments although I am still devising ways to make the daily assessments feasible[6].

To begin with, we plan to start applying this system every quarter when we do the quarterly performance review as well. In this way, we will achieve even greater decrease in “recency bias” and will improve our results.

Overall though I would say it works better than any system that relies on a cabal assembling in a smoky back room determining our bonuses and we are always looking for ways to evolve that system to make it better for the next year.

[1] Lukche is a type of hard candy sold in Bulgaria for the last close to 90 years. Think of them as lollipops.

[2] Another type of Bulgarian hard candy.

[3] Even though we row in the same boat, not all teams contribute equally to the bottom line. And even though differentiating in such a way may seem unfair at first, not differentiating at all is also unfair to the top contributing teams.

[4] Smarter people seem to think that algorithms are far better than humans in such matters. See the work of Orley Ashenfelter, Paul Meehl, David Kahneman, and others for examples.

[5] I guess this is part of the reasons why we do it this way.

[6] Considering that organizations like Bridgewater Associates have done even more this should be achievable in practice.

This post was written by George Kremenliev