John Glasgow rebuilt the general ledger from scratch. Here’s what that means for the teams who depend on it most.
Click for Takeaways: AI General Ledger
- The general ledger bottleneck: legacy ERP code written in the 1990s compresses data before AI ever sees it, making sophisticated analysis structurally impossible no matter how good the model is.
- Six-day close, mostly wasted: the median organization spends more than six business days closing the books each month, with a large part of that time consumed by reconciliation work that better architecture could reduce.
- The 18-month analysis problem: at Adobe, a product-level LTV-to-CAC analysis that shut down an underperforming product line only ran every 18 months because data prep made anything more frequent impractical.
- Security is the real AI adoption barrier: routing general ledger data through a third-party AI system reintroduces the exact data sensitivity concerns that made finance the last enterprise workload to move to the cloud.
- What closing 50% faster actually buys: Campfire customers cutting close time in half aren’t just saving days on the calendar, they’re reclaiming the capacity to bring working capital strategy to the board instead of reconciliation status updates.
Every CFO has seen the pitch: add AI to your ERP and watch the insights flow. What the pitch usually omits is the part where your general ledger has been summarizing data for 25 years, and no AI system layered on top of compressed, context-stripped entries is going to produce anything worth acting on.
John Glasgow has lived both sides of that problem. As a strategic finance leader at Adobe and later as CFO of Invoice to Go through its $625 million acquisition by Bill.com, he spent years as a downstream customer of accounting data. He knew what the ledger was supposed to deliver and how consistently it fell short.
After a run-in with a legacy ERP executive that made the market’s dysfunction impossible to ignore, he founded Campfire, an AI general ledger built from the ground up for the data volumes and dimensional complexity that modern finance teams actually deal with. The company has raised over $100 million and now counts public companies and pre-IPO high-growth businesses among its customers.
His argument is structural, not philosophical. The ceiling on AI-powered finance insight isn’t the AI. It’s the data architecture sitting underneath it.
The Ledger That Legacy ERPs Built
Most of the code running today’s dominant ERP systems was written in the 1990s, when cloud infrastructure didn’t exist and storage constraints made summarization a practical necessity. Finance teams adapted by compressing data before it ever reached the general ledger. Stripe revenue got rolled up. Vendor spend got collapsed by category. Subledger detail got flattened into something the system could process.
That workaround was reasonable in an era when analysis meant human analysts working with what the system produced. It’s a structural liability in an era when AI is doing the analysis.
According to APQC benchmarking data on monthly close cycle time, the median organization still takes a median of 6.4 calendar days to close the books each month, with a significant portion of that time spent on manual data reconciliation that better-architected systems could eliminate.
“If you slap AI on top of an ERP with summarized revenue data, you’re essentially going to get no insights that are of any value.”
The mechanism is straightforward. AI systems generate insights proportional to the richness of the data they can see. Ask one to analyze 100 summarized rows representing a month of Stripe revenue and it will produce answers calibrated to 100 rows. Give it 10 million transactions at record level, with customer, product, region, and timestamp intact, and the analytical surface area expands by orders of magnitude.
At Adobe, Glasgow ran into this ceiling directly. A product-level LTV-to-CAC analysis that ultimately led to shutting down an underperforming product line took weeks to complete and only ran every 18 months. The bottleneck wasn’t analytical capacity. It was the data prep time that summarization upstream created.
What Subledger Detail Changes
The practical unlock from eliminating the summarization tax is that FP&A teams gain access to dimensions they previously had to reconstruct outside the ERP through spreadsheet workarounds. Customer-level contribution margin. Product-level gross margin. Regional revenue with intra-period visibility. The work that used to live in a tangle of VLOOKUPs and pivot tables lives in the ledger instead.
For AI-native companies, where compute and token costs can move gross margin meaningfully within a single month, that visibility is operationally significant.
McKinsey research on the finance function of the future identifies boosting finance’s role in managing data as one of four critical imperatives, finding that leading organizations cutting wasteful data-cleaning efforts are the same ones enabling the function to guide better decisions throughout the enterprise. The constraint has historically been the data layer, not the finance team.
“The dimensionality is all in place: business unit, customer, product level, geo. You have through-the-month, daily, even hourly visibility. You can look at seasonality through a month. You can even course correct through a month.”
Glasgow describes customers who moved from quarterly performance reviews to monthly or weekly cadences without adding headcount, simply because the data was already organized and the reporting automated. The question of signal versus noise at daily granularity remains a judgment call for finance leaders. The ability to make that call, rather than being constrained by what the system can handle, is itself a strategic capability that legacy ERPs do not offer.
The FP&A implications extend beyond reporting speed. Only 35% of FP&A professionals’ time goes toward high-value analysis, with 45% consumed by data collection and validation, according to the 2024 FP&A Trends Survey of more than 2,400 finance practitioners over eight years. Subledger detail flowing directly into the system of record doesn’t just make analysis richer. It removes the upstream work that prevents analysis from happening at all.
Why Bolting AI onto a Legacy ERP Doesn’t Work
The market response to AI in finance has largely followed a predictable pattern: take an existing ERP, surface a natural language query interface, and route questions through a frontier model. Glasgow’s critique of this approach is technical, not rhetorical.
“If you’re feeding data from your ERP to a third-party AI system, it’s like feeding data through a straw.”
Campfire says it built an ERP-native accounting model (LAM) rather than relying entirely on third-party general models. The rationale comes down to three things: security, attribution, and performance.
On performance, Glasgow is direct. General-purpose frontier models will always outperform purpose-built systems on broad tasks. But Campfire positions LAM as more reliable for accounting workflows than general-purpose models. The surface area of the problem is narrow enough that focus becomes an advantage.
On security, financial data was among the last enterprise workload category to move to the cloud precisely because of sensitivity concerns. The data quality and complexity challenges slowing AI adoption in finance are well documented: Gartner’s 2025 AI in Finance Survey of 183 CFOs and senior finance leaders found that adoption has plateaued, with complexity and data challenges cited as key barriers. Routing GL data through a third-party AI system reintroduces a version of the concern finance teams thought they had resolved by selecting a compliant ERP. When data stays within the Campfire environment, the customer retains control over what happens to it, including whether it is ever used for model training.
On attribution, the auditability question matters as much as the accuracy question. When an AI system makes a coding or classification decision that ends up in front of a Big Four auditor, the documentation requirement isn’t just whether the decision was correct. It’s whether the finance team can demonstrate why the system reached it. Owning the model makes that explanation tractable. Routing to a third party makes it nearly impossible.
The hallucination concern that finance leaders raise most often is real but addressable. Glasgow’s position is that hallucinations are meaningfully reduced when context is rich, data is well-structured, and the model was trained specifically for the task. The residual risk is managed the same way competent finance leaders manage risk from human analysts: by building in review, showing the work, and treating AI output as a draft that requires sign-off rather than a conclusion that doesn’t.
“Treat AI like anybody else on your team. Review its work. We give you all the data to do that.”
What Continuous Close Actually Unlocks
The phrase continuous close has been in circulation long enough that most finance leaders have learned to be skeptical of it. The gap between what vendors promise and what most teams experience has been wide for years. Glasgow’s framing is more grounded. The goal isn’t a mythologized real-time close. It’s compressing the feedback loop enough that finance teams can act on data before it’s obsolete.
The mechanism is architectural. Because Campfire eliminates summarization and connects directly to upstream data sources throughout the month, the accounting period’s smallest unit doesn’t have to be a month. Weekly or daily views become practical, not aspirational. For businesses with meaningful intra-month variability, that shift from lagging to leading visibility changes the nature of the decisions finance can support.
Campfire and partners report customers reducing close time by about three days per month on average. That number matters less as a headline than as a description of what it frees. A team that recovers two or three weeks of close-related work per quarter doesn’t just close faster. It recovers the capacity to do the work that the old system’s demands were crowding out.
One customer brought a rethought working capital strategy to the board, not a close status update. The strategic initiative emerged directly from the time and data visibility that the faster close created. That is the version of continuous close worth pursuing: not a real-time dashboard for its own sake, but a finance function with enough runway to bring something to the board other than a variance explanation.
“Someone just shared with me that a customer went to the board and showed how they rethought working capital now that they’re on Campfire. They rethought the entire invoicing process and used Campfire to automate a lot of the tasks, and so they’re able to actually go out to customers and chase invoices in a completely different way. That’s what gets me excited.”
The general ledger has always been the foundation of what finance teams can see and say. For three decades, the constraints baked into that foundation shaped what was possible upstream in FP&A, forecasting, and board-level reporting. Glasgow’s argument is that those constraints were never inevitable. They were architectural decisions made in a different era, preserved by switching costs and market concentration long past the point where they made sense.
The AI general ledger is, at its core, a bet that removing those constraints changes what finance teams can do. The early evidence, from faster closes to weekly board-ready analysis to working capital strategies that didn’t exist before the data did, suggests the bet is worth taking seriously.
Where Datarails Fits
Datarails is the AI-powered FP&A platform built for Excel users, giving finance teams the data consolidation, reporting, and AI-driven analysis they need without leaving the spreadsheet environment they already work in. More than 1,500 companies use Datarails to accelerate close cycles, build board-ready reports, and turn raw financial data into the strategic narratives that drive decisions. Learn more at datarails.com.
This article is based on John Glasgow’s appearance on the FP&A Today podcast.
John Glasgow is the founder, CEO, and CFO of Campfire, an AI general ledger built for mid-market and enterprise finance teams. Before founding Campfire, he served as CFO of Invoice to Go and led its $625 million acquisition by Bill.com, where he subsequently led business development and partnerships. Earlier in his career he held strategic finance roles at Adobe. Glasgow holds the CFA designation and is a Y Combinator alumnus. Connect with him on LinkedIn or learn more about Campfire at campfire.ai.
