How to organize a Performance Review in IT companies: Badoo experience

Hi, Habr! My name is Alexei Rybak, I am the head of development at Badoo. In February, in our Moscow office of Badoo, there was a Techleads Mitap, where I talked about our Performance Review process. This article is based on my talk .

Performance review is a highly controversial topic. In Russia, it is still widely believed that no KPI, “measurements,” or performance review — in software companies not only do not work, but are completely harmful. This categorical nature has always surprised me, but the final decision to highlight this topic ripened in detail after reading the article right at RBC itself .

The article describes the negative experience of the company ABBYY and it is a fairly common conclusion that it is better not to measure programmers, but you just need to give them an interesting job and leave them alone. And, they say, everything will be fine. And because ABBYY is far from the only company whose experience is cited as an argument, there is a much better known story about Microsoft .

The indisputable value of these negative examples for objectivity is worth adding with examples from other companies: Google , Facebook and many others - in which the Performance Review just works successfully. This only illustrates that it is possible to implement a review process both successfully and unsuccessfully. Therefore, the main purpose of this article is to tell about one of the approaches that leads to a working review. In Badoo, this approach has been working for almost seven years, and taking this opportunity, I want to express my great gratitude to Gena Sokolov (ex-Google, ex-Badoo), who “infected” us with this approach.

In this post I:

• I will tell about why review is necessary from the point of view of the engineering manager;
• I will briefly describe how this process happens in Badoo;
• I will formulate questions that are worth answering in order to understand whether your company needs this tool or not;
• and happy to answer your questions in the comments.

Company Review and Values

Performance review, if very briefly, is one of the system-wide methods for improving organizational effectiveness. Mechanically, as simple as possible, a review is a retrospective process that allows you to identify "weak" and "strong" places in the company and tighten the "weak". Each employee talks about his results, gets a “rating” and a comment on which projects or personal qualities you need to pay attention to (sometimes in the form of general recommendations, but better in the form of specific verifiable goals).

Badoo values

However, before proceeding with the review, it is necessary to tell a few words about Badoo and our values. Each company has some properties - external or internal - that cannot or is very difficult to change. Examples of such “boundary conditions” can be the direction of activity, how exactly a company earns money, what values ​​prevail in its corporate culture. In different companies, “to be effective” means different things; therefore, discussing the review in isolation from values ​​has no practical meaning.

Badoo develops dating applications and social networks to meet new people. Our main products are in the dating dating niche, but we are also developing projects in the field of social discovery - locating people nearby by interests. In a sense, Badoo is a platform on which many different applications are made. Inside projects can be very complex, but from the outside everything looks quite simple. A very large number of teams in the world at the same time doing about the same projects. Thus, in the market in which we operate, you need to be very fast: very quickly cut features, test them if they did not go - close them, if they went - twist something and repeat the cycle of experiments again. This determines the whole culture of our company. And all the pain of engineers: it is quite difficult for them to find a balance between this speed and comfortable, unhurried development, when everywhere, as they say, straw is laid.

When in the process of testing a project at an audience, requirements begin to change, it seems at the idea level that nothing has changed at all: some syntactic sentences are replaced by others, but at the design level changes may require serious reworking of interfaces, and at the programming level, it may even end up weeks of work. That is why engineers are painfully aware of changes in requirements, and special follow-up of these changes is required so that teams have high motivation. Since one of the components of an engineer’s motivation is the author’s, it is more comfortable for an engineer to work on “long-running” projects, to be proud of what he has been doing for years, it is hard for him to part with some cool things that needed to be done in a relatively short time, but they did not shoot.

Thus, the balance between speed and engineering values ​​determines the culture of the company. And we try to regularly explain our values ​​to all employees; to a man from the outside, they may seem barbaric. It cannot be said that they are ideally suited to all engineers: it strongly depends on the character.

At this stage, our value system consists of three components.

Value 1: Delivery

The most important thing for us is delivery , delivery of the working product to users. To some extent, we do not even care how perfect it is to use in all audiences. We try to make the most of all features-flags to run in some part of the cluster or in some kind of audience, measure something in the process. If we do not understand how the feature should work for some part of the audience, it will be easier for us to start it only on the “understandable” part of us: what if it turns out to be useless.

A lot of time is spent on conducting A / B tests in order to test various ideas practically without the participation of programmers. First of all, we write software in an experimental way: this is the advantage of the developers of “inalienable” software (strictly speaking, the slow cycle of publishing in mobile application stores limits our speed, but server development and web / mobile web can be released very quickly, usually we will be released strictly twice a day, except for the days before the weekend).

Value 2: Quality

In second place is quality - quality. One of our approaches is that quality and QA as a whole are part of an organization that owns processes along with engineers and programmers, and even more so than they. Thus, we get a balanced distribution of roles, because programmers (especially those who work quickly) on some things that limit them, simply clog, and without balance in the form of QA - completely independent of development - we risk to get a low quality. And this is not at all what we are striving for.

Value 3: Retention

Finally, as I said above, it is very important for us that people come to the company and stay in the company longer. This value is retention , retention. Our threshold of entry is quite high, and finding people is not easy.

When a company is small, everyone knows each other, they work as a team, they celebrate birthdays and other holidays, and even exchange gifts, everything happens naturally simply because the size of the company fits into the usual scale - families, groups of friends, class or student group. But when she (the company) begins to grow, and the number of employees increases, the “usual” methods instantly stop working, instead of hiring a couple people a year, you need to hire dozens. And the key parameter of success (first of all in development) is how fast you can grow, hire and retain people. Therefore, the issue of hiring and retaining qualified and efficient employees is very important for us.

Who and what for review?

Give it back now to Performance Review. No matter how good people in the company work, the more it becomes, the more complex the interaction becomes. If we represent any company as a model, we will get such a branchy structure, when many conditional arrows go from one part to another. The larger the company, the more these arrows and the more losses on these arrows - on interaction, communication.

If you have a small company, you need only one person - a leader, an enthusiast - who is able to run around and “ignite” everyone. When a company is large, it is very important to have a layer of techlids, preferably synchronized with each other, each of which solves problems in his or her area. If something in this scheme does not work - fuzzy goals, weak inactive leads, crookedly divided powers, provoking sluggish cooperation, or even hostility and sabotage - productivity begins to subside, and in especially neglected cases, the company simply begins to “faint”.

So review is a process that allows you to systematically deal with such “swelling”. The review is not a panacea, but I look at it as one of the ways to ensure regular performance checks throughout the company. And do it at the cost of a small “tax” (bonus system).

Badoo is more than ten years old, and it is important for us that people grow with us, staying with us for a long time. Some guys work here for nine to ten years. We strive to always keep abreast and be aware of the expectations of employees, to know what is happening in their lives. Of course, their expectations do not always coincide with the position of managers, but at least we guarantee regular interaction and provide employees with the opportunity to grow within the company. Projecting onto society as a whole, one can say that it is happy if its best representatives have the opportunity to lead the society and lead. And the lack of social elevators for the most active representatives of society can lead to the fact that these people will be apathetic at best, and at worst - prone to revolutionary moods.

But back to the “review.” This is a process, a part of the framework, which shows you where it is stupid. The key issue here is the presence of people who evaluate other people and not just give feedback, but receive detailed information about their projects.

Among other things, the review updates the expectations that I mentioned above. Every time after it there is a conversation between the leaders and members of their teams. When a company grows quickly and former IT professionals become managers, it is important that people understand the meaning of these regular conversations, communication, and regular expressions of gratitude for the work done, this process should not be allowed to flow. First of all, it is of great importance just for retaining employees in the company. If somewhere there is a framework that pushes everyone to this, then the number of such interactions and the number of “thanks” said is steadily increasing. And this number is very important for the company.

Unfortunately, in some organizations, the review is done in isolation from the bonus system: most often it turns out “for show”. This is bad and inefficient - I am convinced that it must necessarily lead to some kind of encouragement, people must participate in this process in order to get something in the end.

So review:

• regularly shows that where “stalls”;
• talks about successes (or failures);
• actualizes long-term and short-term expectations, teaches "good";
• forces me to say “thank you” regularly and to notice shortcomings
• all pleasure for a little “tax”.

The review is bad and good

A few words about a bad review. The worst thing is when it is too bureaucratic and complicated. “It is always bureaucratic,” you will say, and you will be right. But here, as in everything, there must be a measure.

Another property of a bad review is irrelevance. For example, when the premium is paid for what has been done long ago, the focus has already been lost, there is no longer any opportunity to collect information or receive feedback. In this case, people will certainly be considered an unfair review. Note that a large percentage of such people will always be - completely satisfy everyone, so it is very important to come to a consensus.

Finally, a bad review does not contain explicit values, and there is no point in participating in it. There may be a lot of these values, but still the main (what would you think?) Is money.

What should be a good review? First of all - as simple as possible; it should be absolutely applicable to any situation. In addition, it should be relevant and desirable (if we are talking about the bonus component or collecting feedback) as close as possible to what is being analyzed. And since our projects exist for some specific time (even if for a very long time, conditionally, a month, two weeks from which we did something, two more - we watched how it lived in sales, and finally decided that we needed to completely change or throwing out), shifting the review in time is very wrong in this case.

So a good review

• simply;
• applicable to any situation;
• quickly / regularly and regularly;
• understandably;
• nice ($).

What is the review

I look at the review as a process around some kind of framework. What does it consist of? 3 parts:

• peers - colleagues;
• the review process itself;
• ladders - career levels; and also brackets - salary plugs attached to them.

A review is a complex process, and someone must answer for it. Ideally, HR specialist, someone with an administrative function. A correct review takes place in several stages, and since, as we found out, it is always a kind of bureaucracy, it will not move by itself - someone must constantly push people, look at what’s left, and transfer the system from one state to another .

The process itself consists of issuing grades - grades; these may be grades during the review period (good / not so), or global grades regarding progress in professional growth. There are different approaches to the formation of scales of grades, but in general, they are all built around certain expectations.

There are very simple formulas, there are awesome options with clever KPI. I adhere to the principle of simplicity, so I think that it is best to build a scale either around the degrees “good” or around the degrees “conformity to expectations”. For example, an employee fully meets expectations, or with some suggestions for improvement, or does not meet them, or vice versa exceeds expectations, or exceeds expectations so much that it deserves promotion (promotion) and so on. Promotion is a long-term part of the framework that allows you to move a person up the career ladder.

The career ladder in large companies is often quite long, with a bunch of numbers: Vice President No. 1, Vice President No. 2, and so on. We at Badoo practice a simpler system and, like many software companies, give engineers the opportunity to move both the manager’s career path and the engineer’s career path. And one of the most important components of the document that regulates it is work with the expectations of the employee. If he wants to grow as an expert, here is a description of the levels along this ladder. Thus, a person has an understanding of what level of career he is at. This is a rather boring story, but in companies with a large number of employees, this is indispensable.

If your company also has career levels, you must regularly review salary forks. This is done by analyzing the data collected by either your team or external consultants. Often, the analysis does not look very informative, but in one way or another it gives guidelines for changing salaries, for example, when promoting employees or conducting negotiations, issuing counter-affiliates.

The review process itself

What is the review procedure? There is a practice in which the subject of the assessment is given earlier promises or a plan. We never suggest that people tell in advance what they are going to do: plans can change for reasons beyond the control of the person, and he will be upset, and there should not be an unfair review. In the form that each employee fills out, you just need to make a list of the most significant results, and then it is proposed to choose peers.

As a rule, during the reporting period, a person works on three / five / ten projects, and he points them out. We ask people to indicate only what is on the production, or, in extreme cases, those parts of the projects that have been completed and are already working on the developer site. Thus, we once again say: guys, the main thing is a fully completed task, we do not evaluate the degree of fatigue - only the final result.

We have OKRs (objectives and keys results) for employees who work on long-term objectives. Sometimes we do a comparison of what we have planned and what we have done, but in general these indicators are decoupled from the review. Again, this is very important, because it can lead to severe irritation. Since Badoo is a very dynamic company, we cannot use this approach. Therefore, a review is made in fact: this and that is done.

Peers

To get an assessment, you first need to collect feedback from colleagues. The most important part of the framework is peers. Employees choose peers, colleagues with whom they worked on projects, testers with whom they tested tasks, members of the product team with whom they interacted — anyone other than their manager. After that, the manager looks at the peers and says something like: “Listen, why don't I see a single QA engineer here? You had an important task, there was a good (or not very good) feedback. I would like to receive feedback QA-engineer to complete the picture. " There is a certain approval process, the manager can add or remove an employee.

Moreover, “remove” is a fairly common story: if, in the boundary case, everyone adds to the peers of everyone, you get such a bureaucracy that the motivation of employees will tend to zero. It is desirable that the peers were three to five; in the extreme case, seven, but, in my opinion, seven is already a lot.

Peers look at what has been done, give it a rating and write comments; moreover, they have the opportunity to write a comment, accessible to all, and private. Further, all this is collected, and the manager gives his assessment.

Calibration

But that is not all. Ahead is an important process - calibration. In Badoo, it works like this: almost all managers gather in one room and begin to consider evaluations. There are key points (too high a score, too low, it was high, and it became a low one), from which it is immediately obvious that there is a need to discuss and understand the point of view of one or another manager. During this meeting, information is exchanged, we decide what is good for us, what is bad, what is good, what is expected, what is a super result. When a person grows, we say that he needs to raise wages, give another level, put another title within our system, and so on.

Personal conversation

And this is also not all! The main thing is a one-on-one conversation that a manager must hold with each member of his team: tell him what happened and why, what is expected, what needs to be changed.

This is how our framework works in general.

By the way, Badoo has two offices: one is in Moscow, the other is in London, both are engaged in development. In the Moscow office since 2011, when we moved and grew strongly, the number of teams does not change, but the London one is growing. And at first we launched a review in Moscow, ran in this process: we agreed on what and how, we achieved synchronization between the managers, decided for ourselves what this or that estimate meant, and so on. Only after that the London office was connected.

Now it is quite large, the review is going on in parallel in two cities, and, of course, we can’t put them all together. We are trying to divide our entire large team into teams that work among themselves, and calibrate separately. Usually, Moscow is obtained separately, London is separate, but a significant number of teams are distributed between two cities and it is necessary that they all be present at the calibration.

Review problems

Problem 1: bonuses

The first problem we encountered is a different attitude towards bonuses. There are two approaches here. One approach (it is especially popular in the West) relies on an almost guaranteed annual bonus: as a percentage of the annual it makes you so much, and if everything is normal, it is paid in full. If there are any questions or any KPIs have been established, then, as they say, options are possible ... In general, with this approach, the bonus is usually paid either fully or almost completely - that is, changing the bonus is definitely a “punishment” . But the harder to try to play them, the worse because the employee expects that this part of the income is guaranteed to him.

We chose a different approach. We say that if, in general, the employee meets the expectations of the company, then he gets a certain annual bonus. But he may exceed these expectations, or vice versa, a little bit, will not reach them. Therefore, we use a framework that makes it possible to get a very good grade, and worse.

What will people do who believe that you need to “punish” only those who frankly cannot cope with their work? They will try to give maximum marks in this framework. The maximum score doubles the bonus. So it is very important for us to convey to them that if they like everything and everything is in order, this is a usual assessment, everything meets expectations. In the same vein, we have to talk with employees. And in general, it is very important - the bonus must be managed, the lazy manager will simply distribute the maximum bonus, and thus the system will not only not improve efficiency, but, on the contrary, will work to increase fading.

By the way, in addition to text descriptions, we also have numbers that we use for a shorter record. And this is a problem for us. If a person thinks in categories of school grades, he thinks: “Why is not my maximum score? It should be the maximum! I have always been an excellent student, I am the coolest. Why is my rating not the coolest? ”Everybody needs to work on changing such thinking: managers and eycharam.

Problem 2: search for an estimation algorithm

The next problem is an attempt to find a simple “estimation algorithm”. Most programmers, by their nature, are analysts and are trying to build a clear scheme for obtaining the maximum evaluation. Moreover, managers adopt this approach and begin to invent their own stories about how and why they gave this or that assessment, thereby disclaiming responsibility (it's not me! - this is the car you thought).

Another example of this “analytical” behavior. Every time we review salaries, some manager will definitely say: “This employee is not on the median. Unfair - you have to pull it up to the median! ”. But if you follow this approach, as a result, the salary of employees may increase disproportionately to success. In fact, in this median, a very big mistake was originally laid. Therefore, the median is a very approximate guideline, there is no need to “pull up” anything to it! But in reality you need to analyze a lot of factors in the complex. First of all, the success of a person and how the salary of this person has been rising for several years if he has been working in the company for a long time.

A lot of problems are connected with the fact that programmers try to use analytical tools to give one or another assessment. If you have very complex KPIs that are directly expressed by the formula, then there is a high probability that people will just start pulling up their performance to get a higher score: “Eh, now how will I break this bureaucratic system! ..” and rub their hands together. That is why the framework should give as few opportunities as possible to use analytical tools.

I will say a commonplace: the larger the company, the less interaction between different departments. And the more people believe that they are not obliged to tell anyone anything - they just move "their own", they have formed some small states. Problem? Of course. And the only way to fight this is to explain to all parties the need for communication. If someone thinks that they do not intersect at all, it is worth considering the possibility of conducting their review separately with someone else, just do not calibrate in the office.

Moreover, the larger the company, the harder the calibration process, especially during the promotion. People may just not be aware of this. And partly, this suggests that, in general, the explanation of the activities of the company's structural divisions is not very effective and must be dealt with.

Problem 3: inflation estimates and levels

The next moment is inflation. Distinguish between inflation estimates and inflation levels.

With inflation estimates everything is quite simple - it occurs, as a rule, in the normal range. What is exceeding expectations and how does it differ from the “Fully meets expectations” rating? My position is that if there were a certain number of projects and they were all implemented on time or with a slight delay (up to 25% - but there is no single rule), then this is an excess of expectation. It’s not for me to tell you that the timing is usually broken. :) Some managers take a different approach. But in any case, as soon as it comes to deadlines, this means that some document should be included in the review procedure, containing information on what deadlines were set and when everything was completed. This in itself is very cool - it teaches accountability for deadlines.

Fighting inflation levels is more difficult. In short, the upper levels are usually not very specific formulated. For example, an expert. By itself, without explanation, this is a rather vague wording. It turns out, managers see that people grow, develop. Here they are already worthy of the level of an expert - it means you need to promote them. It is very important to compare some experts with others and to constantly improve the descriptions or at least tell each other what this or that level implies.

I can not say that Badoo in this regard is an example to follow. On the one hand, we try to formalize this system, on the other - we understand that as soon as you begin to touch it and describe the levels in a bit more detail, employees start to make more mistakes in expectations, ask: “But I do all this, I deserve it! What is wrong? ”Therefore, the fight against inflation of the levels is most often solved not by a more detailed description of each of them, but by better calibration.

In this sense, the dispute resolution procedure is very important. When calibrating, situations may arise when people are rested - and that’s all, but the process should work. I'm not sure that we use the most efficient way, but now the rules are as follows. If we just give a grade for the work during the reporting period, then we trust the manager who gives out this grade, even if someone thinks that this is not very fair. And if we are talking about moving an employee to the next level, and there are a sufficient number of people who are against it, then we trust this majority more. The logic is clear: if a person works in a team, then the manager knows better how to motivate him. But when the manager promotes his employee, he begins to influence the entire system and show others that this person is, roughly speaking, well done. – .

, , - , .

: . , , – , . , - , “” , – - . : - , , , .

. . , , , . , , , — . How it works? , , , . , , , – .

,

, ? . , . , «» , , , .

, , . , , . ? , – .

. , . , , . “ ” – - . – : , , – .

– . , , . , . , , , , … : , , .

(, , ), , “” – . .

, , , ( ). , – , HR-. , .

Conclusion

, .

, , – , . "" “” , .

KPI? , ? R&D ? , – , , .

, , ? , “ ” . “” .

, , . – . . – “” . .

, , “ !”, Google ( Google).

? .