Pave + Levels.fyi: Self-Reported Data Gets a Seat at the Enterprise Table
The news
On October 28, Pave announced a strategic alliance with Levels.fyi. Levels.fyi data now shows up inside Pave Agent and Pave's Market Pricing tool, right next to Pave's own real-time dataset (9,000-plus companies, by their count) and the survey partners already on the platform: Croner, Korn Ferry, and WTW. Mutual customers can ask Pave Agent to compare pay for a role across all of those sources in one place. Pricing and packaging were not disclosed. Existing customers are told to contact their account team, which is vendor-speak for "it depends."
On its face, this is one more data partnership in a year that has had plenty of them. It is more than that. A self-reported, employee-contributed dataset just got a seat at the same table as Korn Ferry and WTW, inside a tool that enterprise comp teams use to price jobs and build ranges. Matt Schulman's framing is that "no single source tells the whole story." True. But the interesting part is which source got invited, and why. So, let's unpack that a bit.
The oldest feud in comp data
If you have spent any time in a survey room, you know the party line. Survey data is submitted by employers, from payroll, against matched job descriptions, scrubbed by analysts, and aged to a common effective date. Self-reported data is whatever someone typed into a website. One is a dataset. The other is an anecdote with a histogram. I have said versions of this myself, in client meetings, for a long time.
There was a fair reason for the snobbery. Early crowdsourced pay data had real problems: people rounding up, confusing base and total cash, mislabeling their level, and submitting a number from three jobs ago. Anyone who tried to defend a range to a CFO using a Glassdoor screenshot learned the lesson quickly. Consultants and survey houses built a lot of credibility (and a lot of revenue) on being the grown-ups in the room.
What's different now is where the self-reported data is sitting. This isn't a separate tab an analyst sneaks a look at. Levels.fyi data is inside the market pricing workflow of an enterprise platform, as a labeled peer source next to the survey houses. Pave's own announcement calls it "self-reported" in plain language and invites customers to see "where the market views agree and where they diverge." That is the right framing. The data types are not fighting anymore. They are being asked to explain their differences to each other, in front of the client.
Why Levels.fyi earned the seat
This partnership says less about self-reported data in general and more about what Levels.fyi specifically has built. The reason a Korn Ferry client can tolerate Levels.fyi in the same dropdown is the quality review process behind it.
Start with proof. Levels.fyi asks contributors to back submissions with offer letters, W-2s, and pay stubs. Zuhayeer Musa, the co-founder, put it directly in the announcement: the data is "backed by real proof such as offer letters and pay stubs." That alone moves it out of the Glassdoor-screenshot category. A pay stub is a better source document than most survey submissions, which are a spreadsheet export someone promised was from payroll.
Then there's leveling, which is where most crowdsourced data falls apart and where Levels.fyi is strongest. Their whole product started as a crosswalk of company job ladders (they now claim 8,500-plus mapped), so an L5 at one company and a Senior II at another land in a comparable place. That is the same intellectual problem survey houses solve with job matching and level definitions. Levels.fyi solved it from the employee side, company by company, and in my experience they take it more seriously than many surveys do. Their team deeply understands how pay actually works: equity vesting, refreshers, sign-ons, the difference between a target and an actual. You can see it in how the data is structured.
A practical note for anyone reading this: Zuhayeer is a must-follow on LinkedIn for comp practitioners. He writes about pay mechanics with a level of specificity you rarely get from a vendor, and he shows his work. If you are skeptical of self-reported data (and you should be, of all data), reading him for a month is a good way to calibrate that skepticism.
What Pave gets out of it
Now the business logic, because partnerships like this are never only about the customer. Pave's native dataset is real-time and large, but it is built from the HRIS and cap table connections of Pave's own customers. Pave grew up in venture-backed tech, and its customer base still skews that way. 9,000-plus companies is an impressive number, but most of them are not Google. In tech, Pave's own data is deepest in the smaller and mid-sized organization today.
Levels.fyi has the opposite shape. Its dataset is populated by employees and candidates, so it is deepest wherever hiring happens at scale. That means big tech: the companies that make thousands of offers a year and whose employees compare notes obsessively. A Pave customer trying to price a Staff Engineer against Meta, Amazon, and Microsoft gets more from Levels.fyi on that question than from Pave's native data, and probably more than from a traditional survey, which is a year old by the time it lands.
So this is a smart move. Pave gets enterprise heft in exactly the segment where its own database is thinnest, without waiting years for enough large companies to connect their HRIS. Levels.fyi gets distribution into the comp team's daily workflow instead of being the tab the recruiter has open. Schulman's stated goal is for Pave to be "the central source for the data compensation teams use," with its own data as the foundation and "every source a client trusts available alongside it." Read that carefully. It is a platform strategy, and the survey houses are being positioned as content providers inside someone else's interface. Croner, Korn Ferry, and WTW signed up for that. It will be interesting to see who else does, and who decides they would rather not.
So what do we do about it?
If you are a Pave customer, or you use any tool that now blends sources, three things.
Treat divergence as the product, not the problem. When Levels.fyi says $310K total cash for a role and WTW says $245K, that gap is information. One is telling you what the big-tech labor market paid at the offer stage. The other is telling you what a broad cut of employers report paying incumbents. Both are true. Your job is to decide which market you are actually competing in for that job, and to say so in writing.
Know what each source is made of before you blend. Ask the questions you would ask of any survey: who contributes, how are levels mapped, what counts as total comp, how is equity valued, what is the effective date. Levels.fyi can answer those. So can the survey houses. A blended number from sources you have not interrogated is just a more confident way to be wrong.
Document your source policy now. Which sources count for which job families, how they are weighted, and when you will override. The reason survey data won its reputation was discipline, not purity. Bring that same discipline to the new sources and the old argument goes away.
At the end of the day, this partnership is a tell about where market pricing is going. More sources, more labeled, more side by side, inside the tools where ranges actually get built. Candidly, that is good for practitioners. The hard part was never getting the data. It was deciding what to believe.