Why does your close still run on headcount?

Campfire's new Ember Agents handle the recurring grind automatically: transaction matching, AP and AR processing, accruals, anomaly detection, and flux analysis — running continuously in the background and on-demand so close prep becomes a review, not a rebuild.

This is how finance teams operate once the recurring grind stops landing on people.

See Ember Agents live: join founder John Glasgow and principal AI engineer Wen Shu for a 45-minute demo and Q&A on March 17th at 10am PT

SCFO: John, you've got entrepreneurialism in the family, right? Combine that with growing up in Silicon Valley, it sounds like you were destined to become a founder eventually.

John: Yes, I have a couple of founders in the family, including my identical twin brother, who founded a high-growth tech company and a great uncle who invented the Eggo waffle. There's enough to ask, is it nature versus nurture? 

Ultimately, I think growing up in Silicon Valley was certainly on the nurture side. On the nature side, maybe there's something to it. 

SCFO: Right, like there was a fire burning to become a founder, but it needed the right problem to come along?

John: If you're going to start a company, you’ve got to ruthlessly ask yourself, “Why me?” You cannot outwork everybody - everyone's working incredibly long hours. Are you smarter than everybody? Or do you have some unique insight and you’re fueled by passion? And for me, I was fueled by passion.

Having worked in the category and knowing exactly what to build, solving my own problems, that was a unique opportunity.

When I applied to Y Combinator - they have 24,000 applicants every batch - they were like, “We've been waiting for somebody credible in this category that's actually excited.” It’s almost like the more you know about ERPs, the less excited you are to build one. 

And that was my edge.  

SCFO: ERPs have been around for about 30 years, right? You could put a room full of finance people together, and they could gripe forever about issues with existing ERPs.

What are the specific pain points in the existing ‘ERP industrial complex’ that you wanted to fix?

John: I thought about this a lot before I quit my public company executive role and started anew. There are many things, but first, time to value. It’s an incredibly painful move.

But the actual moment for me was when the ERPs - the ones that are 30 years old - pointedly acknowledged to me as a customer and partner that I had nowhere else to go. The switching costs were so high that there wasn’t a good reason to innovate because customers weren’t willing to embark on the massive journey to make a move.

Whether it was massive price increases at renewal or lack of innovation, when you're in a category with switching costs that high, your requirement to keep customers very excited about your product is very low. Then innovation suffers, customer support suffers, and massive price increases with no change in value are common.

I had a C-level at a legacy ERP tell me to my face, as one of the largest AP companies in the market, at our headquarters, “Your customers will rip you out before they rip us out and so you need to pay an eight-figure increase on the annual cost for the integration because our customers literally cannot leave us and yours can.”

And that was, very candidly, one of the founding moments. This category is so broken that they are just point-blank telling customers and partners, “People have nowhere to go.”

The pain is so acute in the category that I spent a lot of time asking myself, “Why has nobody done this? Is there a credible path to actually building an alternative?”

And really getting comfortable with those two questions allowed me to make the jump.

But there were a lot of moments of what I would almost call anger over what I experienced that caused me to go do this.

SCFO: What was the reaction of the finance community when you told them, “Hey, you know these ERPs you hate so much… I’m going to fix that?” 

John: A lot of them were like, “I've always wanted to start an ERP company. That’s cool.” But a lot of them also said, “Are you sure you don't just want to build a feature on top of an ERP?”

Having been on the invoicing side, on the AP side, and having been on the two sub-ledgers, I know the problem is the general ledger. The core is fundamentally broken, and that’s what I went out to solve. I told my wife and investors in the early days, “I am building a general ledger, or I will die trying because the core is fundamentally broken.”

SCFO: That’s interesting, so it’s the very core of the finance tech-stack that's broken? But what exactly is it about the pre-existing solutions that are broken? 

John: Well, we covered switching costs. The other one was that everybody was so frustrated with their GL that they were making it as thin as possible.

They were buying all of these solutions around the GL. You know, categories like close management have emerged, along with AR and AP solutions. There are a lot of reasons to buy third-party solutions, but ultimately, a lot of the customers were just saying, “The ERPs are so bad, why don't I do the bare minimum with them?”

And I said, “If there was an ERP with a great core, could we actually compound, from a product perspective, on top of the GL?” We can do really special things that others are not able to do. 

Nobody says, “I love the reporting in my ERP,” when you need a consultant and huge billable hour run-ups just to get a custom report. Instead, as opposed to tapping a consultant, you're just going to hit the download button and run pivot tables and VLOOKUPs to do it yourself, because you don't have time when the Board meeting is in six days, and some consultant's going to take three weeks. 

There's a consumerization of enterprise software going on. So how do we really deliver so that the end user can configure things on their own? They can build reports on their own, and then we're really actually using the ERP for what it was intended to do. That was one of the big insights.

The other insight was that since the ERPs were written 30 years ago, the core infrastructure was just not scalable. So many of our customers are saying, “We are summarizing data into the ERP because it cannot handle our volume.”

SCFO: And the term AI-native in this context is that AI-native in the tools and engineering that were used to build the product. Or is that AI-native in terms of how AI enhances the product itself? 

John: Well, just broadly, AI-native is a way to unlock budget in the current environment.

Just in terms of your pitch, I think no CEO is going to say, “Take me through AI reconciliations.” But they're going to say, “Will this 10x your team and give you superpowers because it's AI-powered?”

We'll check that box. To give you a more tactical answer, we are the only ERP, to my knowledge, with our own foundation model. So here in SF, we've got our own AI engineering team, and we've developed our own core frontier model. 

So for me, AI-native means we are vertically integrated with our own frontier model, and we own the application layer. We can do things from a performance, from an audit perspective, from a security perspective, that others just fundamentally cannot. 

SCFO: Tell me about this frontier model. Why didn’t you just build Campfire on top of an existing frontier model?

John: The second ‘L’ in LLMs means language. However, finance is not inherently language at its core. It's numbers, it's accounting, it's debits and credits.

But there’s not a lot of accounting data [to train these models on]; no one's posting their GL on the internet. There’s just not great accounting data out there, and, candidly, they haven't been trained on great data to date. 

The whole point of an LLM is that it’s fed so much data that it predicts the next character or the next word. And with numbers, if the next word isn't perfect, then you're off in the wrong direction, and you're into a large hallucination. 

So the shortest answer is that we found our performance is better than the leading commercial models. Because again, we're very narrowly focused on just this one thing. 

SCFO:  You talk to a lot of finance teams at the vanguard of AI and finance. When you think about the best examples you’ve seen, what are they actually doing? Help me understand what the cutting edge truly looks like today.

John: Because we're a modern ERP, we inherently attract the folks looking to push the boundaries. We do see some incredible teams thinking of literally every process that is repetitive in any way, whether it's a board report, transactional accounting, or a reconciliation, and they are saying, “How do I automate this?”

And they're using things like Claude Cowork. You can query Campfire from Claude with our native integration and say, “Build me a board deck. Here's what the last one looked like.” And it's going to take some iteration and prompting, just like anyone else on your team.

I always tell finance teams, “The first version is not always going to be great.” You have to learn how to work with a new hire. You have to give them a lot of feedback, give them some structure. You have to have a one-on-one with your AI.

The accounting team is building their own data pipes from homegrown internal tools. They'll build their own widgets. If there's something that off-the-shelf software cannot automate for them, they're going into Claude Code, into Replit, and building their own apps. And I think that that's the sharp end of the spear today.

SCFO: It sounds like you think CFOs should be thinking about big steps rather than small steps? Isn’t it better to start small? Test the boundaries to understand what the tech is capable of.

John: I would say if the executive level says take a big swing, then take a big swing. But I think many executives are not saying that yet.

You've got to start somewhere. And a lot of folks are saying, “I don't know where to start on AI, or we're not AI-ready yet.” Then at least start with a small swing. And for God's sake, don't call in McKinsey.

SCFO: The median CFO is still stuck on first base with AI, worried about privacy concerns and hallucinations. What do you say to these people?

John: I totally get that it's not an early adopter function. There are a lot of good reasons for that, and I think it's very fair to be skeptical. After all, do you want to be the first to go live on a new LLM and end up with your data leaking on the internet? No. 

I always tell these folks, treat AI like anybody else on your team.

As a finance leader, you always review the work. I ask for the supporting financial model, review the core assumptions that were made, make sure cells are linked correctly, ensuring there's some layers of approval before it goes anywhere.

I mean, it's the newest member of your team. Your team member has a bad first day at the company, you don't fire them. Instead, you give them a lot of feedback.

You review their work in incredible detail in the very beginning. Then you build some level of trust over time, through iterating, feedback, and input.

SCFO: I keep hearing that we should treat AI workers like human workers, but is that really true? It seems wrong to me that the role of the reviewer in an AI-powered process is fundamentally different. AI can hallucinate convincingly; your team won’t. It can replicate mistakes at a much greater speed than an AI worker. Surely that changes the review mode for CFOs?

John: Ultimately, whether you're using Excel or AI or a napkin to do the work, a human is responsible for the output.

You are fully responsible for the AI's output, just like anyone else on your team, and so you need full ownership of what you're delivering from the AI. Ultimately, it's very powerful, but it can make you incredibly lazy and worse at your job. It's a multiplier, but it could also be a decelerator.

Hallucination, error, candidly, I don't care. I tell the team, if it's wrong, that's on you. Look, if you don't know how to use Index Match, don't use it. If you don't know how to use AI, do not use it.

I think we can disagree all day, but the hill I die on is that I am responsible for what I put in front of you.

SCFO: It's very fashionable - and it's not a view I share -  to believe that we're going to automate accounting away to zero. But I've heard you say accounting can be strategic too. Explain that to me.

John: I truly believe that if you can tighten the close, if you can maintain the accuracy while speeding it up, then there is time and opportunity for accountants to pick up strategic analysis of historical data.

Nobody at the company has better visibility into the historical data than the accounting team. When I was in enterprise finance, I would circle numbers that I wanted to dig into, and then I would create a project. What happened here? Why did it happen? And what are we going to do about it? 

I think it's a very unique opportunity for more visibility internally, accounting teams meeting with other stakeholders, and I think it ultimately should lead to higher pay, along with more strategic, enjoyable work.

SCFO: I couldn't agree more. I would argue that the single most strategic thing a finance function could do in a business with a 10-day close is get that close down to 3 days.

But there's a whole bunch of other things the accounting function can do to be strategic as well, right? 

John: What we're seeing emerge is a “strategic controller”, someone who deeply understands accounting but also thinks in systems. They're great at managing software, as I noted earlier, connecting systems with automation, getting the most out of AI. But ultimately, you need someone who’s reviewing the work, ensuring things are accurate, and putting their face on the accounting data.

So it's this evolution of the role where we're not manually reconciling cash anymore, where everyone gets a promotion, and we're managing the systems that are reconciling cash.

Exceptions still come to us. We're still reviewing the accuracy, but it's almost like the specialization is moving a little bit more towards generalization, because everybody moves into a management layer of managing a system, but they're still responsible for the work.

SCFO: That's interesting. That’s going to radically change entry-level roles. There will be fewer, but they will be far more attractive and interesting. It might make accounting cool again…

John: For junior folks that really lean into being systems forward, but can still understand accounting, there's a unique opportunity for them. It has created unique opportunities for everyone if you're really willing to lean into it.

This is more about giving superpowers to a lean team and less about ‘you don't need accountants anymore.’

I think if you're very specialized, and you're reconciling cash all day, you probably want to start to rethink some ways you can step into getting more generalist or something that has long-term durability.

As John and I finished up the last of the s’mores, some important things stood out: 

Most CFOs I know feel more like hostages to the legacy “ERP Industrial Complex” rather than customers.

It’s refreshing to see folk like John tackling the problem in a novel and challenging way. By betting big on a proprietary accounting LLM and focusing on a simpler implementation, he’s giving finance teams a fighting chance to get the most out of AI. 

While we don’t quite agree on how AI will play out for finance, we do agree that we are wtinessing a big inflection point right now.

Hungry for more? Click here for a deeper dive into my conversation with John.

ChatGPT can explain flux analysis, but Ember Agents can run it

Campfire's new Ember Agents are AI workers built specifically for accounting. They match transactions, process AP and AR, manage accruals, flag anomalies, and run flux analysis continuously in the background and on-demand at close.

Every action is logged, sourced to the underlying GL, and reviewable before it posts. Nothing is a black box.

See Ember Agents in action