OneSchema

A Startup Built Inside the Gap the API Left Behind

Here is a fact that ought to be more embarrassing than it is: a great deal of the world's economically important data still moves around as spreadsheets. Not APIs, not elegant streaming pipelines, not webhooks firing in real time - spreadsheets. A file called final_v3_USE_THIS.csv gets emailed from a finance team to an operations team, uploaded into some SaaS product, and then, roughly one time in three, it breaks. The dates are formatted wrong. There's a phone number with a country code and one without. A column everyone swore was numeric contains the word "pending."

OneSchema is a San Francisco company that decided this recurring, unglamorous catastrophe was worth building a business around. The pitch is almost aggressively practical: instead of your engineers spending months writing and then forever maintaining the code that ingests customer files, you drop in OneSchema's importer, and you have a polished, validated, error-catching upload experience in about a day. That is the entire value proposition, and if you have ever shipped a "just add CSV import" ticket that quietly metastasized into a quarter of work, you already understand why it is a good one.

The founders like to point out the reason this problem refuses to die. "10 years ago," co-founder and CEO Christina Gilbert has said, "we might have predicted that all data integration would be possible via API. But it turns out there are a lot of persistent reasons why nearly all companies have to transfer data via spreadsheets." This is the sort of observation that sounds obvious once someone says it and is worth a lot of money precisely because most people assumed the opposite.

The number that does the real work here is that "20+." Consider a single column - say, a date. There is the American way and the sensible way and the ISO way and the one Excel invents when it feels like it. Now multiply that combinatorial mess across every field your customers might upload, add the customers who put headers on row 4 and a company logo on rows 1 through 3, and you arrive at the true cost of file import. It is not the happy path. It is the long, sad tail of everything that can go wrong, and OneSchema's entire product is a machine for eating that tail.

An Affinity Engineer and a Google PM Walk Into a Data Problem

OneSchema was founded in 2021 by two people who arrived at the same frustration from opposite directions. Andrew Luo had been an early software engineer at Affinity, an investor-relationship CRM whose customers - Bain Capital Ventures, MassMutual, Swiss Re - care intensely about data being correct. Christina Gilbert spent four years as a product manager at Google. One had felt the import problem from the code side; the other from the product side. Both concluded it deserved better than a pile of one-off scripts.

They took the company through Y Combinator, and by late 2022 had raised a $6.3 million seed round. The lead was General Catalyst, with Sequoia Capital, Y Combinator, Contrary Capital, Box Group and Elad Gil along for the ride. The more telling part of the cap table is the roster of operators who wrote checks: the founders of Zapier, Gem, Stytch, Podium and Athelas. These are people who have personally shipped products that had to ingest other people's messy data. When they invest in the company solving that problem, it reads less like a bet and more like a group of veterans buying insurance against their own past pain.

What You Can Actually Do With It

At the center is the Embeddable Importer: a drop-in widget that a SaaS company can put inside its own product so that its customers get a clean, guided upload flow. It validates and cleans data as the file comes in, catches the errors, and - this is the part the founders emphasize - lets a non-technical person fix them. It handles files up to 4GB and roughly 20 million rows, which is a polite way of saying it will not fall over when someone uploads their entire history of everything.

Around that, OneSchema built Pipelines (previously FileFeeds), which automates the recurring version of the problem. If a partner sends you a file over SFTP every night, or emails you a report every Monday, Pipelines lets a non-technical team member define how that file gets mapped and transformed - no engineering ticket required. And more recently the company has leaned into AI agents for autonomous data operations: software that intakes, preps and reconciles messy legacy data on its own, reaching into PDFs and database dumps and the sorts of systems that never had an API in the first place.

The thread running through all of it is the same, and it is worth stating plainly because it is the whole strategy: take work that used to require a scarce, expensive engineer, and hand it to either a non-technical user or an agent. OneSchema is not jumping between problems. It is climbing the same one - from a single import, to a repeating feed, to an autonomous migration - which is generally what good companies do instead of chasing whatever is fashionable.

The pricing follows the same logic. OneSchema charges a platform fee plus a consumption component tied to successful imports, which is a tidy way of aligning what the customer pays with what the customer actually gets - a file that made it in clean. It is a model that only works if the product genuinely reduces failure, because you are, in effect, being paid per problem solved rather than per seat occupied. That is a harder promise to make and a stickier one to keep, and it tells you something about how confident the company is in the boring part of its own machinery.

The Differentiator Is the Tedium Nobody Else Wanted

There is a category of companies here - Flatfile, Dromo, Osmos, CSVBox at one end; the big data-integration incumbents like Informatica, Talend and MuleSoft at the other; unified-API players like Merge and Finch nearby. So the natural question is what makes OneSchema's version stick. Gilbert's answer is refreshingly unromantic: rival products, she has argued, simply "haven't invested in addressing every edge case." OneSchema did. The validation layer is the moat, and the moat is made of tedium - the thousand small, boring, awful data problems that someone has to handle and that most competitors were too proud or too impatient to grind through.

This is the quietly clever thing about the business. The unsexy problem has a defensive property: your competitors do not want it, your customers hate doing it themselves, and the work is so unpleasant that once someone reliably takes it off your plate, you are unlikely to leave. The customer list bears this out - Scale AI, Vanta, Ramp, Toast, Pave and Airbase on the startup side; CreditSafe, Personio and the publicly traded Huron Consulting Group on the enterprise side. These are not companies that adopt a data tool casually.

Why "Boring" Might Be the Whole Point

It is tempting to file OneSchema under "utility" and move on. But the more interesting reading is that the company found a large, permanent, embarrassing tax that nearly every software business quietly pays - the cost of getting other people's data to arrive clean - and offered to lower it. The broader market for enterprise data management runs into the tens of billions and keeps growing, and most of that money is spent making messy data usable. OneSchema did not invent a category so much as name a fee everyone was already paying.

There is also a lesson here for anyone building a company, and it has nothing to do with data. OneSchema's founders did not pick a problem because it was exciting. They picked one because it was reliably, universally annoying - the kind of thing competent engineers grumble about at eleven at night - and then they were willing to do the unpleasant work of actually solving it end to end. Excitement is abundant and cheap. Willingness to own the tedium is rare, and it turns out to be the scarcer, more valuable resource.

Whether the newer AI-agent framing turns a good developer tool into a category-defining data platform is the open question, and it is genuinely open - autonomous data migration is a crowded, fast-moving field in 2026. But the underlying bet has aged well: the spreadsheet did not die, the API did not save us, and someone was always going to get rich cleaning up the difference. OneSchema decided it would rather be that someone than complain about the CSV like everyone else.

Figures reflect the most recent public reporting available; funding, headcount and customer counts are approximate and subject to change. Roles noted per company announcements; Andrew Luo served as CEO in the company's earliest years.