Why a CFO copilot should be model-agnostic

The question comes up in almost every conversation with a founder thinking seriously about AI tools for their books: which model are you using? It is the wrong question. For a product that reads messy financial data and turns it into a plain-language answer, the model matters less than the read. And one model at one tier is the wrong architecture for that read.

Reading messy money is language work

A bank export is not a database query. Vendor names are misspelled, truncated, or aliased by the payment processor. A single charge might map to any of a dozen cost categories depending on the vendor. A contractor appears under their personal name one month and under their LLC name the next. A rules engine breaks the moment any of these ambiguities appear. A language model reads intent, and it does that reliably in a way that no lookup table can.

Why one model is not enough

A clean three-line ledger and a 400-line mixed-currency export from a startup with overseas contractors are not the same problem. Routing both through the same model at the same cost is like sending every package by overnight courier regardless of weight or destination. You overpay on the simple reads and sometimes underserve the complex ones. A routing layer solves this cleanly.

• Clean, well-structured ledgers route to a fast and cost-efficient model
• Messy, contradictory, or multilingual exports route to a stronger one
• A single-provider outage does not stop the analysis
• The routing table improves as model capabilities shift, without touching the product

The figures are grounded in code, not inference

The language model reads the ledger and reasons about it. But the headline figures, runway, net burn, total leak, are not produced by the model's inference. They are computed in code from the model's structured output, checked for internal consistency, and tied back to the specific line items that produced each number. The model is the reader. The arithmetic is a machine.

Why this is a copilot, not financial advice

A licensed financial advisor has a duty of care, a regulatory framework, and liability for what they tell you to do. Vael does not have any of those things, and is clear about that. What it has is a careful read of your numbers, with the reasons shown and the sources cited. You might share that read with your accountant, your fractional CFO, or your board. The decision is always yours.

We built model-agnostic routing not because it is clever, but because it is the honest answer to a problem of varying difficulty. Committing to one model is committing to one difficulty setting for every ledger.Mara Whitfield

The result is a product that gets stronger as models improve, without a rewrite. When a faster or more capable model becomes available, the router updates and every read benefits. The user never picks a model or needs to understand what changed. That is what model-agnostic means in practice: the product improves on its own.