Before I ran product at Pilot, I was one of Pilot’s target customers. I ran a tech company, and I remember the first time I went looking for a way to do the books myself. I had taken accounting classes, but I had never run books end to end in a real business, and I was frustrated that no platform offered it. That was a number of years ago. When I heard about Pilot and saw what it offered, it made immediate sense. I had lived that problem, I knew how important financials were, and here was a company delivering something close to heart.
These days I’m the head of product at Pilot. No two days look the same, but the core of the job is understanding what our customers need and building products that serve them. I work closely with our engineers and designers on what to build and how it should work, from the technical side all the way up to the things that delight people.
Having been on the other side gives me some perspective on individual features. If we’re building a new insight report that shows how a line item varies month to month, I can picture needing that number for a board meeting next quarter. It’s nice to think from that seat. It’s another data point for building the best product possible.
Meridian is the operating system for a firm delivering books at scale
Meridian is the operating system of a firm delivering books at scale. We can say that with confidence because Pilot has spent over a decade as exactly that kind of firm. In the process of making our own teams more productive, we created something that does more than enhance the work. It delivers a fully automated set of books. We saw how transformative that was internally, in our metrics and in how quickly we could serve customers, and because we’d lived it, we knew it was missing from the market. Meridian is our chance to bring that end-to-end close to the rest of the world.
The moment we knew this was big came early, when one of our founders put together the first proof of concept several months ago. Seeing that demonstration was a light bulb moment. It was fundamentally different from the usual road of development, the incremental progress of chasing something a little better. This was a leapfrog.
When a piece of technology fits the ethos of what you’re building and is that transformative, it’s clearly more than business as usual. There’s still a lot of work to build an entire platform around it. But when you’re sitting on something that exciting, you don’t want to keep it internal. You want everyone else to benefit.
The 10-year recipe
Could someone take an AI algorithm and just say, go do the books? Technically, yes. The difference between that and Meridian is thousands of hours, tens of millions of transactions, and 10 years of data across over 7,000 customers.
One reason an agentic system works so well for us is that we’ve been building the guardrails for 10 years without quite realizing it. The close checklist is a recipe guide for shipping a great set of books. Follow it for a neighborhood flower store or bakery, or for the early days of OpenAI, as we did, and you get an auditable set of books on the other end.
Paired with that recipe are the automations tied to each step: the tools and engineering that handle reconciliation or apply a set of rules to a category. We built all of that for humans. It turns out it doesn’t matter whether a human, a set of automations, or an agentic AI system does the step. The guardrails are built in. That hardened, decade-old recipe is what’s different about Meridian, and it’s why this is more than taking a model out of the box, handing it some data, and seeing what you get.
A great demo is easy. The last 40% is the hard part
If another firm tried to do this in-house, I think they’d find something we’ve heard a lot: take a model out of the box, give it a month of financials, and 60% of the time it looks really good in 10 or 15 minutes, maybe a few hours. That’s extremely promising. You think, if I can spin that up this fast, how amazing is it going to be later on?
What you’d find is that the last 40%, the last mile of getting the work done for every type of customer and every irregularity, is the hard part. Every business has exceptions every month. It’s just a matter of finding where. Ten years of understanding, learning, and codifying those exceptions is what separates an exciting demo someone could spin up today from doing this at scale for every customer, managing every exception, firm-wide.
Describing the work in words
One of the more exciting things we’re rolling out now is open-ended work in Meridian AI. We rolled it out first for the customers Pilot serves. Thinking as a founder again, I could ask: “I’m doing X, Y, and Z in a board deck, and I know they’ll ask me about this specific thing. Can you pull in my numbers and help me figure it out?” Or, “I’m building this forecast. Can you help me structure it?” No two businesses are the same, and I’ll never build enough reports to serve every need, so being able to ask an open-ended question for my specific need is great.
Now we’re releasing that same open-ended, user-defined outcome for accountants. Long, complicated tasks, like splitting out across categories, running a more advanced revenue recognition, or making a batch of journal entry changes, can be described in words and then done, without the hours of manual ledger work they usually take. Having a tool that flexible is really exciting.
Still, what carries the most weight in deciding what to build is the conversation with customers. Just today we were talking to a customer about how they use the product, and every time I ask a question like that, something I couldn’t have thought of comes up. Those are the magical moments that tell us what to build, what to prioritize, and how to structure it. The human element separates an okay product from a great one. I highlight that because it matters even more in a world where it’s tempting to over-rely on AI and agentic services. The connection to a human, and how we facilitate it internally and help our partners facilitate it with their customers, is what ultimately wins the day.
Building trust in an agentic system
Trust is a little personal. What I trust might differ from what someone else trusts. But ultimately, trust is built over time and by understanding how something works. I trust a person more once I know their goals, what they’re interested in, how they function. Something similar is true here. It’s weird to talk about goals and functions for an AI, but understanding the guardrails, what it looks at and what it doesn’t, how it treats my data, how it handles privacy and security, how it’s been audited: those are the machine proxy for meeting someone and learning what they care about. They’re what we go through with the firms we work with on a demo and testing basis. We had to explain them to ourselves first, because we need to trust this too.
The goal is to move the work and the thinking into higher-level conversations, and provide the underlying basis so that as a user I don’t have to think about the rest. Once I have the financials, I can start asking: what if this changes? What if I hired that person? Those stop being abstract questions. They’re informed by data. So I think demand for everything we’re doing is going to go up. That’s the other side of the coin from the risk people see in agentic systems. It lets us have higher-level conversations about value.
If you’d like to see what Meridian can do for your firm, schedule a call with the team.