Compensation is really, really hard to get right. Very little directly-comparable data exists to make a truly objective decision, and what data is available, is easily sliced in such a fashion to support whatever conclusion you want to draw. This intersection of “high importance” and “high ambiguity” makes the whole topic stressful, where even if you feel great about your personal compensation, you can’t help but be suspicious that others are being paid more for doing less. This stress and suspicion erodes the trust necessary to maintain an open and collaborative environment: you can’t pull your oar the hardest if you’re looking back over your shoulder.
How It’s Typically Done
There are a variety of ways to solve (or at least manage) this problem. The most common of those is to create a series of formal titles, and then create “compensation bands” around each, where every role has a range of possible salaries and equity grants. Then when you are hired, whoever is hiring you assigns you to one of those — and then picks a place in that band for you.
This technique has the benefit of being very simple: it’s usually pretty clear which title most applies to which people, so in theory it makes the entire compensation process straightforward and objective. Where it fails in practice, however, is that the compensation bands are incredibly wide even for the same role, and this system provides very little guidance for whether to pick the high or low end of the band for a particular person.
Furthermore, because there is typically no defined process for selecting the exact number within the band, the decision comes down to one person making a decision based on some internal measure that will never be known. This is a recipe for abuse, where one person has an enormous amount of influence over your compensation without any recourse.
Even worse than abuse (or at least more common) is laziness. If your compensation depends on one person to evaluate all the data and make a fair decision — despite them not benefiting in any way — odds are they simply won’t do it. Rather, they’ll likely pick somewhere in the middle, and see how strenuously you respond. If you guffaw loudly, they’ll raise it. And if you don’t, they won’t.
This dynamic is what has created the general expectation that everybody should negotiate hard during the interview process, as the easiest time to increase your salary is before you take the job. After that, you lose your leverage, and your compensation is largely up to the invisible whims of your manager, who is given a wide band to work with, and who is likely going to take the easiest possible case of just sorta winging it and doing the minimum that they think you’ll accept.
The next best time is during the performance review time. But startups are notorious about having erratic performance review periods, making those opportunities few and far between. Similarly, despite the general wisdom of discussing compensation at that time, it’s still enormously stressful and awkward — meaning most people lose their nerve even if they do get the chance.
Playing this out over time ends up with the “squeaky wheel getting more oil”, and those who negotiate the hardest and complain the most are unfairly compensated — even though those are rarely the folks who are actually performing the best. (After all, it takes time to negotiate and complain, and that’s not time doing what you were hired to do.)
Nearly every company works this way, and you can tell if yours does by asking for a raise — just out of the blue, don’t wait. If it works, it means they’ve been underpaying you the whole time, and are counting on your silent acceptance of the raise to foster an environment that is fundamentally unfair to all those who aren’t brave enough to ask.
On the other hand, if they deny your request for a raise, it might be due to them being fair. But only if they explain that they’re denying you because they deny everyone. If they don’t, then any fairness is the result of sheer coincidence, and not by taking a process that achieves fairness by design.
Fairness By Design
There are two competing philosophies on how to “design for fairness”. On one end of the spectrum would be Buffer, and their legendarily radical approach toward transparency. In short, there is a published formula for exactly how everybody’s compensation is determined — both salary and equity — and the results are clearly on display for all to see, both to employees and the general public alike. They take the approach that transparency alone is enough to create fairness, and that the more eyes look at the results, the fairer those results will be.
I have a huge amount of respect for Joel and Leo at Buffer, and commend them for this bold approach. But Expensify takes a different path.
Our view is that the world is too nuanced and dynamic to be fully captured with any formula. In reality, there are a thousand contributing variables, and any two people would pick and weigh variables differently (as well as assess a different value of each variable) when assessing the fairness of any compensation scheme. Accordingly, locking in a single formula for everybody is essentially adopting one person’s opinion of what it means to be fair, which rarely achieves the best results.
Granted, there are a series of basic “yes/no” questions that can just be decided by fiat and enforced uniformly, such as:
- Should you be paid less if you live in a lower cost of living area?
- Should you be paid more if you have a larger family or more expensive habits?
- Should you be paid less if you are junior, even if you are outperforming someone more senior?
(We answer those “no” to all three: we feel you should be compensated at Silicon Valley standards, entirely by your contribution, regardless of your costs or where you live.)
But at some point you get to fluid questions that aren’t quite so easy to answer:
- How much more should people be paid who take on hard work that others don’t like, or who spend time helping the people around them?
- How much less should people be paid who are easily distracted, and distract those around them?
- What premium do we place on those who are growing quickly and show capacity for continued growth, versus those who are plateauing?
- What penalty is there for creating or contributing to a toxic work environment, versus those who tackle daily challenges with a positive collaborative attitude?
These highly subjective variables matter, a lot, and the inability to quantify them is no excuse to ignore them — that just leads to a formula that codifies unfairness, even if transparently on display for all to see.
Expensify’s Compensation Review Process
Our approach toward fairness depends on two assumptions:
- Everybody should be paid generously with respect to the market. We aim to pay “comfortably above” whatever the data suggests others get paid in similar roles — with the caveat that good data is surprisingly difficult to come by.
- Everybody should be paid fairly with respect to their peers, which means you should be paid more than those creating less value than you, and less than those creating more value than you — with the caveat that “value” is a subjective measure in the eye of the beholder.
Both of these should be true even if you don’t ask. That sounds obvious, but by necessity it also means:
- We don’t negotiate during the hiring process. Our first offer is our last and best offer. If it’s basically in the right ballpark, then we encourage you to take it and we’ll adjust it later if we were off. But if we’re so far off that we’re not in the right ballpark, then clearly this isn’t going to work anyway. There’s no need to fight over a few percent: let’s just get started and adjust later.
- “Later” comes reliably — twice a year — and we follow a process that ensures both of the above principles are thoroughly evaluated on your behalf, without you needing to ask or negotiate. Indeed, asking for a raise outside of this process won’t work, and the process isn’t influenced by negotiation, so there’s really nothing you can do to improve your compensation but excel in your role every single day.
The process itself is very, very involved and time consuming, and works as follows:
- We start with an alphabetized list of every employee in the company, worldwide. This list has no titles, no indication of current compensation, etc. Just names in a spreadsheet.
- We ask our directors and team leads (16 people) to bucket people who they think should be paid about the same, and then sort those buckets. This is an extremely difficult, stressful, and time consuming process, as not every lead works with everybody in the company. This involves a lot of searching your inbox and GitHub to get a sense of how this person has done, asking around to people who know them better, and mostly just a lot of introspection.
- The “sorted bucket” technique is surprisingly flexible to handle the range of scenarios between:
- People you don’t know well should be grouped into larger buckets, indicating that you don’t really have good insight into their performance individually, and are instead making a more aggregate assessment about the group as a whole.
- People you do know well should be put into smaller buckets (or even sorted individually) to recognize that you do have useful insight into this person’s performance that you want to contribute to the process.
- Outside the mechanics of how to actually do the bucketing, leads are given the following guidance for assessing relative value:
Factor in everything you can imagine: productivity, quality, collaborative skill, enjoyability to work with, inspiration of peers, dedication to do the hard work nobody else wants to do, title, seniority in the company, long term potential, etc. Choose whatever factors you care about, and weigh them however you like.
This process recognizes that value itself is subjective and shouldn’t be left up to any one person to decide. Accordingly, the only way to produce an output that captures the full nuance of reality is to place the fewest constraints on the input.
- This is normalized and aggregated to create a matrix of about 1500 datapoints. In a way, you can think of this like the first layer of a distributed neural network, where each lead is a separately trained net processing input from their unique perspective, the output of which feeds into a second layer for aggregate processing. Accordingly, this is as far from a simple calculation as one can get: the very first step involves dozens of hours of work by our top people in the organization. And similar to a neural net, the results are too nuanced and sophisticated for any single person to understand — the quality of that result is better than any single person could achieve.
- This matrix is visualized a number of ways to highlight outliers (i.e., where one lead assigns a value to someone wildly different than everyone else), as well as to calculate a “confusion coefficient” for each employee (measuring how much disagreement the team had about an individual). This is shared with me (CEO), our directors (sales/success, engineering, and strategy/marketing), and our HR lead for further processing.
- Before we continue, we remind ourselves of the decisions made in previous cycles around how to handle various edge cases (eg, “How do we factor in employees who are undergoing a ‘performance improvement plan’?”) and how to answer various philosophical questions (eg, “Do we give a cost-of-living adjustment based on location?”) to make sure we’re all on the same page.
- Then we begin a laborious process, starting at the bottom of the list and working our way up, of sanity checking each employee’s relative position in the list. This means reviewing how different leads assessed the same person, and discussing why they might have come to different conclusions. We factor in unique information we have that the leads might not (especially regarding sensitive matters that they aren’t aware of), placing equal weight on each of our two rules.
- If we disagree with how the team sorted someone, we’ll take a look at the “confusion coefficient” for that person — the more confused the team was on an individual, the more comfortable we are in trusting our own judgement and overriding the team’s results. However, we make every attempt to honor the team’s judgement unless we have some compelling theory for why it was off, typically based on information they didn’t have, or some other bias inherent in the system. This process takes about 8 hours, and grapples with impossibly subjective questions that need to be answered in the most coldly objective ways.
- In particular, “silent contributors” are identified in this stage, i.e. employees who just work hard on something critical that everybody else takes for granted, as these are the people most vulnerable to unfair compensation and thus deserving of extra attention.
- Similarly, this is the stage where incredibly difficult comparisons are made about how to fairly compensate across wildly different parts of the organization. As convenient as it would be to give up and say it’s impossible to compare sales, engineering, marketing, and HR employees in an apples-to-apples basis, the reality is that this comparison needs to be made and this is the stage where that happens — and not according to abstract titles or job descriptions, but based on an individual case-by-case basis.
- Once this is done, we go back through the list and identify those individuals whose team-assigned position we overrode, and document our reasoning for each. The main reason behind this is to just double-check that we had good justification for each decision made.
- Next we choose a “cost of living adjustment” that is applied uniformly to everybody, generating a baseline raise on a daily compounded rate (going back as far as your last raise, which might have been more recent than the last comp review if you are a new hire).
- Then we go through the list again from bottom to top, and give the minimum raise necessary to establish “fairness” based on the new ordering, which means you are paid no less than the person beneath you, and no more than the person above you.
- In practice, this is a difficult process fraught with exceptions. For example, there are often cases where somebody is paid substantially more or less than their peers, due to some oddity about their unique history. We take each on a case-by-case basis and try to come up with the fairest outcome to all, without being slave to any particular formula.
- Next we compare anybody we can do with external data available to sanity check that we’re paying our goal of “comfortably above market”. This works nicely because while it’s impossible to find data on every single role, we don’t actually need to: once everybody is sorted correctly, if we determine that market rate for someone “beneath” you has gone up, you benefit from that because the market adjustment “trickles up” to maintain fairness.
- Next, we go through and assign any individual performance based raises we feel are appropriate — again, trickling up the effects on anybody above them.
- Between each of these steps we typically “sleep on it” and review the results with a fresh mind the next morning. Every decision is up for reconsideration at any point in the process.
- Finally, the new compensation is revealed in a 1:1 setting, typically with an explanation of why it’s so big (or not) and feedback on how to make it bigger next time around.
- After all is done, the individual picks how much of the raise to take in the form of additional cash salary, or in the form of additional equity.
And that’s it. It takes probably 100 hours each time, spread across the top members of the team, and we do it twice a year. It’s a huge investment, but it’s one of our most important investments because the team is our most important asset. If your company isn’t taking your compensation as seriously as this, it means they just don’t value you as highly.
To learn more about what Expensify can do for you, please check out http://we.are.expensify.com and I hope I can convince you to consider a switch!