The most important document at every $100M ARR company lives in an Excel file that three people know about.

It has four thousand rows.

One row for every revenue event of the past month — every contract signed, every seat expanded, every renewal processed. Each row carries a story: what the rep negotiated, what exception the CFO approved.

The person who built it is a senior accountant. She spent three days gathering the data, and then two days doing calculations. Her result was a summarized entry with exactly two lines. "Revenue, $8.33M." That’s all that’s recorded in the company’s finance software. Every story in those four thousand rows will effectively cease to exist.

Here's what it looks like:

This process is called “closing the books”, and it’s been standard practice for over 500 years. But it might be better called “destroying business context at industrial scale.” Because when the CEO asks why revenue decreased last month, the books don’t contain any of the necessary context to answer the question.

And we've been destroying this critical business context, at every company in the world, every month, for centuries.

I've spent years building these systems

My background is in software engineering. Before Quanta, I spent years building financial systems of record for many products. This included building from scratch an in-house accounting system that tracked billions of dollars moving every day, and being a tech lead for products that originated tens of thousands of loans every day. I became an expert in building these systems, and saw how powerful they were when built well.

So I was pretty surprised when I saw “modern” accounting systems for the first time.

They have a fancy name – the General Ledger – but they are effectively dumb databases. They store numbers, but force the work of calculating those numbers to happen elsewhere: across different systems and excel scratchpaper. And because the work is done elsewhere, the accounting system has zero context on the numbers it stores. It’s like if you did your math homework on a whiteboard and then erased it before turning in the final number. It’s an irreversible, completely lossy trap door.

That trapdoor creates enormous problems down the road, as the finance team is required to try and climb back through it. When the CEO wants to understand product performance, or when the CFO asks why sales efficiency increased, the finance team needs to go back to find the original source data. It's manual archaeological work, and it's also often wrong. I’ve heard hundreds of complaints from finance leaders about when the accounting system says one number but the finance team’s spreadsheet says another, and no one can bridge the two.

Sales has CRM. HR has HRIS (Rippling, Gusto, Workday). Finance has a ledger that was designed in the 1400s and hasn't fundamentally changed since. Every other function in the business has a schematized, living system that maintains context, but finance has only a google drive full of excel files.

Why this is hard

There is a reason the work is done by hand: it’s been historically impossible to beat the flexibility of spreadsheets, and recording everything that has ever happened in the business requires a lot of flexibility.

No two companies’ business models have the same shape. And even if your business model looks standard, there are always exceptions: the one strange contract where the customer reimburses you for compute; the revenue-share you're piloting with a partner who sends you deals; the new product line that starts on prepaid credits and flips to retroactive overage pricing. Building something that can express all these edge cases is extremely difficult.

In addition, this context lives across documents, external tools, product databases, data warehouses, spreadsheets, Slack threads, and people’s heads. Integrating every finance tool API is not enough. There has never been a system that could integrate deeply enough into the actual business to gather everything needed to run finance. So everyone builds it manually themselves.

How we’re solving the problem

Generative AI is an unlock for multiple reasons.

First, it can navigate the business systems that don’t have APIs, to collect the necessary data. This includes browser automation, as well as internal tools.

Second, it can read and understand all of the human-written documents and context. Every customer contract, vendor invoice, receipt, and employee-written reimbursement memo.

Last, it can solve what the current era of deterministic SaaS logic cannot: it can handle bespoke edge cases on the fly. If-this-than-that logic was never expressive enough to handle the nuances of every single customer contract. Generative AI can handle it.

But there is a key to unlocking this: agents can only solve these edge cases if they are operating on top of a rock-solid foundation of new financial primitives. If these primitives are right, LLMs can dynamically assemble them in infinite compositions: whatever is needed to handle the edge case. To achieve this, the primitives must be expressive, guardrailed, and safely composable in any permutation. Their logic and the work they did must be transparent: finance teams require audit-proof calculations they can defend.

They also must be part of a system which contains the rest of the business context, or else they won’t have enough information to make the right decision. Pointing AI at one silo’d tool leads to confident but wrong results. The glue between the tools - the assumptions, context, and tribal knowledge — is required, and we’re encoding that too.

So: AI is the unlock, but pointing Claude at the existing scattered tools and spreadsheets won't solve much. A new centralized foundation of new primitives is needed. This is what we’re building. Here's how it works, in three layers:

Bottom layer: Data ingestion & interpretation

Most AI-in-finance companies start at the top of the stack. We started at the bottom: the hard but crucial work of organizing the data.

Each customer's financial reality is spread across a dozen systems with no shared schema: billing in Stripe, payroll in Rippling, contracts in DocuSign, expense data in Ramp, custom usage logic in an internal Postgres.

LLM browser automation as well as LLMs’ ability understand human-written text is a new unlock to this goldmine of information that previously required human effort to process.

Part of Quanta’s moat is access to this data. Because we're our customers' accountants, we have actual logins to every finance tool. No competitor — not the legacy accounting systems, nor the new AI-bolt-on startups — has that level of integration. A requirement of accounting is that it needs to encompass every penny that’s moved in the business, and every penny it knows will move in the future. Our comprehensive visibility into this treasure trove of data and context is our foundation for solving incredibly hard problems.

Middle Layer: The Operational Ledger

Now that we’ve consolidated and parsed all of the data, we can organize it into the system of record that models the entire business. We’re calling this the Operational Ledger, because it has the rigor of the general ledger, while also storing all the operational context of a business.

It’s actually a system of smaller “subledgers”, one for every part of the business. We’re modeling every business operation as event, then applying state machines of rules and policies that derive its financial consequences.

As a result, every business metric is linked to its full upstream context: the source data, the policy applied, the calculation performed, the human decisions involved. We’ve captured the why behind each number.

Our vision is to replace today’s dumb and contextless general ledger with a system where each number is the output of traceable computation: versionable, replayable, policy-aware. This enables the ability to branch financial history, to replay the year under different assumptions. This has a very practical use: finance teams today can spend weeks building a “scenario” model manually. With Quanta, they can be modeled instantly. More on that in the next section.

This solves today’s lossiness problem, those four thousand rows. Our General Ledger is a projection of this richer structure, instead of a lossy compression of it. It essentially becomes a materialized view. All of the context, instead of being destroyed as it is today, is now preserved to be useful.

Top layer: Agentic intelligence that can replace today’s research projects

The job of finance teams is asking and answering questions. The job titles can be fancy, such as FP&A (Financial Planning & Analysis) — but the underlying work is answering questions about the business. Those questions are along the lines of: "what are our margins on the new product?", "who can we afford to hire?", and “how is the new pricing working?”

Companies live and die by the answers to those questions, but they currently take days or weeks of manual digging to answer. Every question requires a project coordinated across data teams, system admin teams, and analysts on finance teams.

Because the layers of Quanta described above have consolidated and organized the context of those teams, we can automate those multi-week projects by harnessing AI to run the research on our data.

This interface for these research projects is a product invention problem. Today, the only tool flexible enough to model everything required by finance teams is spreadsheets. With generative AI on top of Quanta data, we can combine the flexibility of a spreadsheet with the liveness and drill-ability of a data warehouse, to invent a new interaction layer. The interface we’re building is generative — the visualization and proof customized to the shape of each question.

Built from the work itself

For over two years, we've been closing books for real companies — including some of the fastest-growing companies in Silicon Valley like Braintrust, Browserbase, Decagon, and Paraform — and learning exactly what’s required to fully handle their business models.

We've quietly built a complete replacement for the core functionality of QuickBooks and NetSuite — and for the manual work people do around them. Many customers have turned off their Quickbooks and NetSuites and switched to Quanta.

We're figuring this out in real time, with real customers, and there's no established playbook. And we’re just getting started.

If this is the sort of work that interests you — defining the financial primitives, architecture, and interaction model this system requires— please reach out at careers@usequanta.com. We’re looking for people who are drawn to problems with no known answer, and who want an outsized say in what gets built.