You Can't Bolt AI Onto a 1998 General Ledger. Finance Is Finding That Out the Hard Way.

Every legacy finance tool now has an "AI" badge on its pricing page. Almost none of them changed the database underneath it.

That gap — between AI on the marketing site and AI in the architecture — is the most important story in finance right now. Stephen Hedlund, Head of Finance at Rillet, put his finger on exactly why it matters.

In a recent conversation on why AI-native ERP is changing finance, Hedlund made the case that you cannot retrofit intelligence onto a system that was never designed to hold it. Rillet — the AI-native ERP that raised over $100M from Sequoia and Andreessen Horowitz — rebuilt the general ledger from scratch rather than wrapping AI around someone else's.

We built JustPaid on the same conviction, one layer down the stack in billing and AR. So this one hits home.

Here's why the AI-native rebuild is the real shift, and why "AI-powered" legacy software mostly isn't.

The ERP Most Companies Run Was Built for a Different Decade

Start with an uncomfortable number: Gartner estimates that through 2027, roughly 70% of ERP implementations fail to fully meet their original business objectives. Large transformation projects routinely overrun budgets by triple digits.

These aren't edge cases. They're the base rate for the software running most finance teams.

Hedlund's framing of why is sharp. Legacy ERPs treat data the way 1998 treated data: something you key in, store in rigid tables, and pull out in static reports.

AI needs the opposite.

It needs rich metadata on the way in, deterministic logic in the middle, and a system that can read its own context on the way out. You can't get there by adding a chatbot to a 20-year-old schema.

The schema is the problem.

This is the same bet NetSuite made on cloud and the bet Rillet is now making on AI: when the platform shift is big enough, you don't patch the old thing. You rebuild.

"AI-Powered" and "AI-Native" Are Not the Same Product

The distinction Hedlund draws maps cleanly onto how AI actually has to live inside a finance system.

Think of it in three layers.

1. Ingestion — How Data Enters

AI-native systems capture deep metadata at the source — the contract, the CRM, the payment — instead of flattening everything into a few generic fields.

2. Core Processing — What Happens in the Middle

This layer has to be deterministic.

Your close cannot hallucinate. Math is math, and an AI-native architecture keeps the language model away from the ledger arithmetic.

3. Extraction and Action — How Data Leaves

This is where agents belong: surfacing anomalies, drafting the reconciliation, and recommending the next move.

Bolt-on AI usually only touches layer three — a copilot stapled to the front of a system that's still rigid underneath.

AI-native means all three layers were designed together.

That's the difference between a feature and a foundation.

The Results Show Up Where Finance Feels Pain: The Close

The proof isn't in the architecture diagram. It's in the calendar.

Hedlund points to a company that went from a 15-day close to a 3-day close in its first month on an AI-native platform.

That's not a tuning improvement. That's a different category of system.

When ingestion, processing, and extraction are built for AI, the manual reconciliation and chasing that bloat a close mostly evaporate.

We see the same physics in billing and AR.

JustPaid customers run a 3x faster month-end close and collect payments 17 days faster because the contract-to-cash flow was built AI-native from day one — not assembled from a CRM, a billing tool, and three spreadsheets held together by a controller's heroics.

The Buyer Changed, So the Software Has To

There's a generational point underneath all of this that Hedlund nails.

The CFO buying finance software today often grew up on consumer-grade tools and expects implementation in weeks, not the 12-to-18-month death marches of legacy ERP.

The data backs the urgency: in recent surveys, the large majority of CFOs plan to increase AI spend, yet most still have no generative AI in their finance function at all — even though nearly all of them believe it would help.

That gap between intent and reality is exactly what AI-native vendors are built to close.

Rillet is closing it at the ERP layer.

JustPaid closes it at the billing and AR layer with a 3-to-7 day implementation instead of weeks — because there's no legacy migration to fight.

What This Means for Your Finance Stack

If you're evaluating finance tools in 2026, Hedlund's lens is the right one to borrow.

Ask Where the AI Lives

Is it in the demo, or is it in the data model?

If the vendor can't explain how AI changes ingestion and processing — not just the chat box — it's a retrofit.

Judge by the Close, Not the Feature List

The honest benchmark for any AI claim in finance is whether it shortens the close and the cash cycle.

Weight Implementation Time Heavily

A 12-month rollout means you're buying yesterday's architecture.

Weeks-not-months is a signal the system was built for this era.

Stack AI-Native With AI-Native

An AI-native ERP like Rillet paired with AI-native billing like JustPaid beats a legacy core with a dozen bolt-ons because the data stays rich end to end instead of flattening at every handoff.

The Takeaway

The takeaway from Stephen Hedlund's argument is bigger than ERP.

The entire finance stack — ledger, billing, AR, FP&A — is getting rebuilt from the database up.

The teams that win the next decade won't be the ones who added AI to old systems. They'll be the ones who ran on systems where AI was never optional.

Frequently Asked Questions

What does "AI-native ERP" actually mean?

An AI-native ERP is built with AI embedded in its core architecture — from how data is ingested and processed to how insights are surfaced — rather than having AI features added to a legacy system after the fact.

As Rillet's Stephen Hedlund argues, this requires rebuilding the general ledger and data model, not bolting a copilot onto an old database.

Why can't legacy ERPs just add AI?

Legacy ERPs store data in rigid schemas designed decades ago for manual entry and static reporting.

AI needs rich metadata on ingestion and flexible, deterministic processing — capabilities the old data model can't provide.

That's why retrofitted "AI-powered" tools typically only add surface-level features like chatbots.

How is AI-native billing different from AI-native ERP?

They operate at different layers of the finance stack.

An AI-native ERP like Rillet rebuilds the general ledger and core accounting.

AI-native billing like JustPaid rebuilds the contract-to-cash flow — extraction, invoicing, collections, and reconciliation.

Both share the principle that AI belongs in the architecture, not stapled on top.

What results does AI-native finance software deliver?

The clearest signal is a faster close.

Hedlund cites a company moving from a 15-day to a 3-day close in its first month on an AI-native platform.

JustPaid customers see a 3x faster month-end close and collect payments 17 days faster.