Filings First: Why Rexfin Isn't Racing to Build 200 Connectors

Look at how most finance software gets sold and you’ll notice the same slide in nearly every deck: a wall of logos. QuickBooks, Xero, NetSuite, Sage, Plaid, Snowflake, dozens more, implying that breadth of connections is the product. It’s an easy story to tell and expensive to keep telling: every new logo is another API to maintain, another auth flow to babysit, another edge case in how that system labels its chart of accounts. We connect to the major accounting platforms, banking feeds, and warehouses that matter most to our customers today. We are not chasing the full list, and we don’t think you should want us to.

The moat isn’t the connector count

Here’s the thing a connector logo doesn’t tell you: pulling data out of an API is the easy part. QuickBooks will hand you a trial balance without complaint. What’s hard, and what actually determines whether the number in your model matches your ledger, is what happens after the data arrives: mapping a company’s often idiosyncratic account labels onto a consistent structure, catching totals that don’t cross-foot, flagging the scale mismatch where one file reports in thousands and another in whole units, and holding anything ambiguous for a human to confirm before it’s treated as fact.

That mapping and verification layer is the same regardless of whether the raw data arrived through a live QuickBooks connection, an Excel workbook someone emailed over, or a CSV export from a system we’ve never seen before. So that’s where we’ve put the effort: a canonical ingestion and mapping pipeline that gets you to a reconciled model reliably, no matter which door the data walked in through. See how ingestion works and the canonical atom store for the structure that produces.

Upload always works: it isn’t the fallback plan

This is the part most vendors quietly treat as a downgrade: what happens when you don’t have a live connection set up yet. For us, it isn’t a lesser mode. Drop in an Excel workbook, a CSV of your actuals, or a PDF financial statement, and the same reconciliation logic runs (cross-footing totals, sanity-checking periods, flagging anything that looks off) before it reaches your model. You can start modeling with what you already have today, before anyone touches an API key.

That matters beyond convenience. A live connector assumes the other system is the source of truth and stays that way. In practice, plenty of finance teams still run large parts of their planning in a spreadsheet nobody has fully “systemized,” or need last quarter’s board deck built before this quarter’s ledger export is even ready. Treating upload as first-class rather than a stopgap means the model doesn’t wait on IT to provision a connection before it becomes useful.

Why we’re not chasing universal document parsing either

It would be tempting to swing the other way and promise we can parse any financial document from any company in any format on day one. We don’t make that claim. Building an extraction engine that reliably handles the full messiness of arbitrary filings (every issuer’s idiosyncratic layout, every regional convention) is a genuinely large undertaking, and overselling it would put us back in the same trap as the connector wall: capabilities that are technically true and practically unreliable. Guided mapping starts with the formats finance teams already have (spreadsheets and exports), with statement parsing extending from there as we harden it.

The deterministic checks that catch a bad import before it reaches your numbers are the same ones that catch a bad calculation anywhere else in the model: see why deterministic calculation. And because every mapping decision and correction is recorded, the ontology gets sharper with every workspace we onboard: a compounding asset a connector count can’t replicate.

What this means if you’re evaluating us

If your first question is “how many integrations do you have,” ask a second one: what happens to a number after it arrives. A wide connector list with a shallow reconciliation layer behind it will hand you a model that looks plugged-in and still doesn’t tie to your ledger. We’d rather you start on a spreadsheet you already trust and get a model that reconciles, than start on a live feed and discover a mapping rule quietly drifted six months in. If you’re bringing on a new team, onboarding, first week walks through what that first import looks like, and the broader inside the Rexfin platform pillar covers the rest.

The connectors will keep growing: we add them where real customer demand justifies the maintenance cost, not before. What won’t change is the order: reliable ingestion from whatever you already have, first; connector breadth, second.