How AI gets built: Matthieu Hafemeister
Concourse builds AI agents for corporate finance teams. These agents allow finance teams to query information in natural language and perform quick analyses to obtain new insights and drive business decisions.
We spoke to Concourse co-founder and CEO Matthieu Hafemeister in a candid conversation on his journey building an AI company, his learnings from CFOs, and his advice to finance teams getting started with AI. You can follow Hafemeister on LinkedIn and X.
Using AI to disrupt finance’s status quo
As Jeeves, an expense management platform operating across 20-plus countries, scaled, Matthieu Hafemeister, who led growth, hired a new finance ops person nearly every week to keep up.
But this common approach of managing scaling by hiring more people didn’t work well.
"At some point that breaks," Hafemeister recalled.
More people meant more spreadsheets, more manual reconciliation, and what he calls "the messy middle" — a patchwork of disconnected analysis. "Because we were just trying to move so quickly around the decision-making, we didn't always have time to regroup all the information."
The experience exposed a fundamental tension in finance: automating to move fast and delight customers, whilst remaining flexible and grounded in data.
Hafemeister explained that since automation involves implementing software, it’s a time-consuming process that requires maintenance as shifting business needs require updating or replacing systems. Adaptability, on the other hand, still involves a manual process: people working in spreadsheets. Most finance teams rely on both approaches.
"That's great, but then the data's out of sync and the information's all over the place," Hafemeister said.
"Instead of getting the best of both worlds, you're getting the worst of both worlds."
When LLMs emerged, Hafemeister and his co-founder Ted Michaels saw an opportunity. "Every team has the same problem around data, getting access to it, and making decisions off of that information. We should just start a company in this space."
That company became Concourse.
Hafemeister at NYSE a16z Demo Day Pitch
The road to becoming a founder
Hafemeister went from fintech investor to operator to founder. At Andreessen Horowitz, he worked with late-stage companies, and then with early-stage ones, learning how to pressure‑test ideas and find product‑market fit. At Jeeves, he got further into the weeds of operations: shipping products and scaling quickly.
“Being able to just work on the same goal together, no egos, just solving problems, it’s really fun.”
Each step in his career peeled back a layer, revealing to him how companies are built. But starting Concourse showed him how difficult the founder journey can be. It was a big learning curve to build the first product, iterate, acquire customers, and pitch — and often get rejected.
"You think that it's going to be hard, but you can manage because you're smart," he said. "And then you start and think, 'This is really hard. No one wants to talk to me.'"
But Hafemeister suggests there is one way to make the journey easier: "Just work with people that you like working with."
At Concourse, he looks for people who are excited about the earliest stages of building: ex-founders and early-stage operators. He wants candidates who tackle hard, interesting problems and want more ownership than a role might offer at a bigger company.
He extends that principle to customers, choosing to work with finance teams who are excited about AI and the problems Concourse is solving.
The biggest piece of advice he’d give to someone starting a company? How important having a really great co-founder is.
“Being able to just work on the same goal together, no egos, just solving problems, it’s really fun.”
How an AI founder spends his days
While no two days look the same for Hafemeister, there are three areas he typically focuses on:
Hiring: Concourse is intentionally technical and small. Using AI internally across building, prototyping, and go-to-market helps the team maintain impact while staying lean.
“Everyone has a computer or data or finance background,” Hafemeister explained. “And engineering is our bread and butter.”
The team hires AI engineers working on model behavior and accuracy for financial data, back-end engineers for integrations and infrastructure, and front-end engineers to build interfaces.
Concourse is also building what Hafemeister calls a "forward-deployed engineer, but make it finance" — AI specialists who blend finance expertise with engineering skills to help customers encode business logic into agents.
“How do we enable CFOs to cycle through much more analysis of that information? That is the mandate we have.”
Customers: “Everyone talks to customers every day,” Hafemeister emphasized. This is to understand how customers are using Concourse, and how they could make the product better. It’s also important to Hafemeister, especially with enterprise customers, to understand how their businesses operate and encode that into the agent so it can perform on benchmarks that are relevant to them.
Product. Finally, Hafemeister spends his time with engineers, thinking through the product roadmap, ideating, workshopping, shipping, and testing.
“I try to tackle as many things as I can, and then end the day with more things than I started the day with,” he says.
“Which is why weekends are great, because I get to process a lot and do thoughtful, focused work.”
The Concourse team (L-R): Matthieu Hafemeister, Dhavan Katri, Ted Michaels, Nitya Kasturi, and Paul Sukhanov
On choosing the right customer segment
Concourse started with the treasury segment, an instinct that came from the founders' time at Jeeves. Hafemeister and his team managed roughly 80 bank accounts across regions, products, and currencies for Jeeves. Making sense of that complex flow of funds offered a concrete problem to start with.
But as they began building, he observed that teams are often hunting for operational data — financials buried in the ERP, product-level data, month-to-month metrics — more than treasury data.
This focus also helps the company deliver on the more strategic role CFOs now have. “Finance is often the arbiter of truth for business data, and CFOs increasingly own more of the stack,” he commented. That broader remit allows AI to do more for CFOs, giving them visibility into data teams, RevOps teams, deal desks, HR, and legal.
“How do we enable CFOs to cycle through much more analysis of that information? That is the mandate we have.”
On accuracy in a binary finance world
Finance is unforgiving: numbers are either right or wrong. How does Concourse meet that standard with probabilistic AI?
Hafemeister said the answer lies in how numbers are produced and how questions are defined.
Concourse doesn't let AI blindly crunch the numbers. When calculations are involved, the system writes code, using Pandas, Python, and SQL, to fetch data from source systems, compute results, and validate them. “It’s not allowed, effectively, to produce numbers that cannot be traced back, and that are not effectively accurate,” Hafemeister explained.
“The longer CFOs wait, the more their competitors will have access to AI and be doing things with it."
"That's different than a generic LLM where people might throw data into the model and it's pretty much a coin flip." In practice, this approach yields 99% accuracy or better on the numbers, though he admits "AI will never predictively and repeatedly be 100% accurate all the time.”
The harder challenge is in the definitions, since many "wrong answers" are really mismatched meanings.
"If I asked everyone what revenue is at a company, I would probably get 10 or 50 or 100 people giving a different answer," Hafemeister says. “Are we talking gross revenue or net, or including interest income?”
Concourse addresses this ambiguity through features such as memories, which understand business-specific terminology and definitions. Teams prompt the agent what key terms like revenue or contribution margin mean for their company so they don’t need to re-explain it each time.
Yet, this involves a critical skill: prompting, or what Hafemeister calls "the language of AI."
"For me it's probably the number one skill you need to have in the next 10 years to be successful," he noted.
To this end, Concourse helps customers learn how to prompt AI well. Hafemeister’s team shares examples of good prompts and writes prompts with customers for their workflows. Concourse also released a prompt library, a new feature where teams can take prompts built off of common workflows and adapt them to their businesses.
Learnings from CFOs
Hafemeister spends his days talking to finance leaders. Three themes stand out from his recent conversations:
The scope of the job is expanding: CFOs are increasingly telling him, “Now I own this part of the business,” in areas that once sat firmly outside their core job.
Skepticism is reducing: Early conversations with CFOs were dominated by doubt about AI accuracy and hallucinations. But that skepticism is shifting to curiosity. Hafemeister attributes this partly to pressure from other teams who are using AI. As finance leaders keep using AI, they realize: “I'm just using it for my emails, but I feel like it could do so much more.'"
Enterprises are more proactive: While startups are Concourse’s largest customer base, Fortune 500 companies are increasingly interested in the company. "We've seen large Fortune 500 customers start to be really excited about AI and have more dedicated budgets and core initiatives around it," Hafemeister said.
3 principles for CFOs leading AI adoption
In an X post this summer, Hafemeister predicted that software as we know it is going to die. How can CFOs prepare for this future?
- Don’t replace your stack — augment it. Hafemeister’s advice to CFOs: "Don't change anything. Keep your business software, keep your team.” Rather than replacing software or people, Hafemeister emphasized that CFOs should layer AI on top of existing workflows to 10x their capacity.
- Start now. “The longer CFOs wait, the more their competitors will have access to AI and be doing things with it," Hafemeister warned. He drew a parallel to the internet era: "Who got hired as the CFOs at all the big tech companies? Was it people who didn’t want to use the internet, or the people who were excited by it?,” he asked. “Those are the people that are getting the dynamic jobs that are exciting.”
- Invest in the learning curve. The work requires investment, but it is rewarding. "The more time you spend in the platform, the better the AI gets and the better you get at prompting. It's a compounding effect," Hafemeister noted.
The momentum notwithstanding, Hafemeister believes this is just the beginning.
"We're still very much in inning one," he said. "But the demand is there, and finance teams are realizing they need to be leveraging AI."
