AI ERP: A Startup Founder's Guide

The general ledger is the least glamorous piece of software a company runs and one of the most important. It is the record of every dollar that moves: the source of truth accountants close each month and auditors check each year. For most of the last two decades the rest of finance modernized around it, expense tools, billing tools, payment rails, while the core ledger stayed slow, rigid, and dependent on consultants to configure. AI ERP is the attempt to rebuild that core around a language model. Instead of storing numbers a person keys in, the software reads the documents, drafts the entries, reconciles the accounts, and assembles the reports. Campfire, an AI-native ERP out of Y Combinator, is betting the whole company on that idea and winning startups off the incumbent NetSuite to prove it. This piece explains what an AI ERP is, why the opening exists, and what founders in any category can copy from how Campfire is attacking it.

What an AI ERP actually is

Enterprise resource planning is a heavy phrase for a simple job: the accounting and financial reporting that every company has to do. Taxes, the general ledger, the monthly close, the reports investors and auditors ask for. Because every company that makes money has these obligations, the market is enormous and no startup gets to skip it.

A traditional ERP is a system of record. You configure the workflows, your team keys in the transactions, and the software stores and organizes them. An AI-native ERP moves the model from a bolted-on feature to the engine. It ingests invoices, contracts, and bank feeds directly, proposes the journal entries, matches transactions during reconciliation, and drafts the reporting, with a human reviewing and approving rather than typing. The work that used to be manual entry and cleanup becomes review of what the software already did.

That shift is possible now for a specific reason: language models can finally read and understand messy documents and unstructured text, which is most of what accounting actually is. This is why finance keeps showing up on lists of the next industries AI will reshape, the raw material of the job is documents and language, exactly what the technology got good at. It is the same pattern we cover across the AI for startups pillar, where AI stops being a feature you sprinkle on and becomes the thing that does the work, an idea we go deeper on in software that does the job instead of handing you a tool.

Why the general ledger got left behind

The interesting question is why this opening exists at all, given how mature accounting software is. The answer is that the incumbents are genuinely deep and broad, and that depth became the problem.

Systems like NetSuite can do almost anything a large enterprise needs, which is exactly why they are hard to run. Configuring them is a project, keeping them running often means paying consultants, and many technology companies end up using a fraction of the features while carrying all of the complexity. The tools around the ledger, billing, expenses, payments, got modern and pleasant to use. The ledger at the center stayed a heavy implementation. That mismatch, a modernized finance stack wrapped around a legacy core, is the gap an AI ERP aims at.

There is also a middle ground that nobody served well. Below the enterprise incumbents sits QuickBooks, which is fine until a company starts scaling and hits its ceiling. Once you need approval workflows so the right people sign off on spending, or multi-entity accounting to consolidate several legal entities, QuickBooks runs out of room. The traditional next step was a painful jump straight to a heavy enterprise system. That awkward middle, too big for the starter tool, not ready for the consultant-heavy incumbent, is precisely where a new ERP can plant itself.

The Campfire case: winning startups off NetSuite

Campfire is the clearest working example of the AI ERP thesis, and its story is useful precisely because it is narrow. The company was founded by John Glasgow and went through Y Combinator's Summer 2023 batch. Rather than trying to serve everyone, it aimed at one profile: technology companies that had outgrown QuickBooks but were not well served by a legacy enterprise system. By being the best tool for that specific customer, Campfire has been pulling companies off NetSuite, the deep incumbent it competes against, according to reporting from TechCrunch.

Investors have backed the bet heavily. Campfire raised a $35 million Series A led by Accel, then a $65 million Series B co-led by Accel and Ribbit, as the company reported in its own funding announcement. The founder's background is part of why the bet reads as credible: Glasgow spent his career in finance, strategy, and business development roles at Adobe, Invoice2go, and Bill.com before starting the company, so the pain of running legacy financial systems is something he lived, not something he read about. He talks about Campfire as a decade-plus commitment rather than a quick swing, which is the kind of founder-market fit that makes a slow, trust-heavy category winnable.

One detail from Campfire's origin is a lesson on its own: the first version was prototyped in a spreadsheet, usable in a browser, before it became a product. You do not need a finished platform to test whether the core idea works, which is the whole argument for prototyping cheaply before you commit engineering.

What founders can copy from the AI ERP playbook

You do not have to be building finance software for the Campfire approach to be useful. A few moves generalize to almost any AI-native product going after an entrenched incumbent.

• Pick a painfully narrow customer, not a broad one. Campfire did not sell to every company that does accounting. It named one profile, tech companies that outgrew QuickBooks, and became the best option for exactly them. A few similar customers compound: one insight serves many, and the roadmap stays focused. This is the same discipline behind writing a narrow ideal customer profile before you chase demand.
• Attack the middle the incumbents ignore. The opening was not the enterprise top or the small-business bottom, it was the underserved band between them. Look for the customers who have outgrown the cheap tool but dread the heavy one.
• Use consumer-grade usability as the wedge. Legacy ERP needs consultants because it is hard to use. Making it configurable by the finance team itself, without an implementation project, is a real advantage, not a cosmetic one. Ease of use is a moat when the incumbent's product is genuinely painful.
• Sell it yourself until it clicks. The early motion here is founder-led sales, especially in a top-down, trust-heavy purchase like a company's financial system. The founder carries the conviction and the domain credibility that no early sales hire can fake, which is the argument for a top-down enterprise sales motion run by the founder until you have real product-market fit.
• Make AI the engine, not the sticker. The claim that lands is not "we added AI." It is that the software does the accounting work now. If the model is not doing a real share of the job, an AI label is just marketing, and buyers in a serious category can tell the difference.

The security and control questions to ask

I write this as a CISSP, and a finance system is where the security angle stops being optional. The general ledger holds a company's most confidential data and, through approval workflows, it moves real money. Handing more of that work to a model raises the stakes of getting the controls right, whether you are building an AI ERP or buying one.

Ask the hard questions before you trust it. Who can approve a payment, and does the system enforce that separation of duties? Is every AI-drafted entry reviewable, with a human sign-off before it hits the books? Is there a complete audit trail an auditor can follow, showing what the model did and who approved it? How is your financial data isolated, and is it used to train anything outside your walls? The point of an AI ERP is to remove manual keying, not to remove human judgment from the moments that matter. A well-built one makes the review faster and the trail cleaner. A poorly-built one hides mistakes inside automation, which in accounting is the expensive kind of mistake.

What to do this week

• Write down which finance system you run today and where it hurts: manual entry, a slow monthly close, a consultant dependency, or features you have outgrown. That list is your real buying signal.
• If you are on QuickBooks and hitting its ceiling, evaluate an AI-native ERP against the legacy incumbent on the specific things you now need, approval workflows and multi-entity accounting, not on the length of the feature list.
• If you are building an AI-native product in any category, name the one narrow customer profile you will be the best in the world for, and write it down before you widen.
• Put the control questions in writing for any finance tool you consider: approval separation, human sign-off on AI entries, a full audit trail, and data isolation.
• Keep founder-led sales for now if this is your own product. Carry the pitch yourself until the market tells you clearly that it fits.

Choosing or building the finance core is one decision inside a much larger question: how to run an entire company on AI without losing control of it. That operating picture, from the tools you adopt to the governance around them, is what we teach in the AI Operating System for Startups.

Sources

• The ERP for the AI Revolution is here (Y Combinator, Root Access), the interview this article distills, in which Campfire founder John Glasgow explains the AI-native ERP thesis, the narrow tech-company focus, the QuickBooks-to-NetSuite middle, and the founder-led sales approach.
• Reporting and background on Campfire and its founder: TechCrunch on the company winning startups off NetSuite and its Accel-led Series A, Accel on the investment thesis, Campfire's own $65 million Series B announcement, CFO.com on Glasgow and the company, and the Y Combinator company profile. Founder profile: John Glasgow (LinkedIn).
• Cross-source context from our video knowledge base: This Is The Next Industry AI Will Disrupt (Y Combinator, Root Access) on why accounting automation is now possible, as language models can read and process documents.