Built to break: why ERPs fail at transaction scale
NetSuite architect Nathan Garing on why finance teams with a world-class ERP still reconcile in spreadsheets, and why the fix sits before the ledger, not inside it.

The business runs on a well-known ERP. Still, every morning someone in finance clicks export, opens the file in Excel and does the real work there. By lunch, the dashboard and the spreadsheet no longer agree.
Nathan Garing has seen that pattern from both sides. He is a NetSuite Architect and Co-Founder of Salora ERP. He has more than a decade of accounting experience and seven years of leading ERP implementations. Before he built systems, he was a finance user himself, managing more than 250 million dollars in accounts payable. In this episode of High Volume Accounting Live, he tells Peter Engel where standard configurations break down. He also explains why more customisation is not the answer.
Two jobs, one condition
Nathan starts with what an ERP is for. "The ERP number one is a data tool. It’s a unifier of the client’s data. Number two, it’s a communication tool."
Both jobs depend on the same thing: the data underneath has to be right. And the ERP cannot make it right. It records what it receives and passes it on.
The export is already out of date
The habit he meets most is the spreadsheet that becomes the real workspace.
"Every time you click that export button for data, that data is immediately stale data. It’s no longer the most accurate up-to-date data because so much is happening in the real world."
While the team works on the file, orders ship, payments land and refunds post. Once the spreadsheet is the source of truth, the system can do neither of its two jobs.
Volume does not wait
At high volume, payouts show where it hurts first. Take a store with 50,000 orders a day. In Nathan’s words, "you get one settlement deposit from Shopify, one from Klarna, one from Afterpay, one from PayPal."
Each deposit bundles thousands of orders, net of fees and refunds. Matching that back by hand is not a staffing question. Nathan’s verdict on growing without fixing it is blunt:
"How are you gonna grow and scale up to 150,000 orders a day if you can’t get it right at 20,000 orders a day? You’re just not."
People scale in a straight line. Transactions do not. A workaround that just holds at 20,000 orders a day will not hold at seven times that volume.
Validate before the ledger
The fix sits before the ERP, not inside it. Nathan calls unvalidated data in the ERP exactly what it is.
"It’s a ticking time bomb. The data needs to be clean and accurate. Number two, we need to validate it and reconcile it. Number three, the data needs to be actionable."
Skip the validation step and the ERP processes a wrong number with the same confidence as a right one. Most companies cannot even see where the wrong numbers come from.
"Rarely does a client have a good flowchart of their architecture. You look under the hood and you’re getting 5,000 script and integration errors a day. That’s data that’s inaccurate."
This is why he sees ERP work as architecture, not IT. The tool is the easy part. The hard part is deciding what good data looks like and where it gets validated. After that comes the question which systems in the stack should still exist.
AI needs a human check
Nathan uses AI tools every day and builds automations with them. He is clear about the limit. "AI also just hallucinates and makes stuff up. No matter the data source or how sophisticated the tools, we need human validation and a human element."
When Peter asked whether broken finance operations come down to technology or to people, he refused to choose. "It’s always a people mindset problem. And it’s always a system problem too. It’s always a little bit of both."
What to take from it
Three checks from this conversation for your own team:
- Where does your team still work on exports instead of in the system?
- Can you tie each PSP deposit back to the orders it covers, every day?
- Can anyone draw how your systems connect, and where data is validated before it reaches the ledger?
If the honest answer to the last one is no, start there. The Actuals scan takes 15 minutes and shows where your data first stops matching before it reaches your books.
In practiceWhat breaks first when your volume doubles?


