Thesis

The starting point wasn’t payroll. It was a broader frustration: finance teams at growing companies – CFOs, FP&A managers, treasury leads – were drowning in manual work while AI tooling was exploding everywhere around them. The founding observation was that this specific cohort had been left behind. They were still reconciling spreadsheets, chasing approvals over email, and making decisions from data that was always slightly stale.

The initial concept was an AI agent that could act as a real financial co-pilot – not another dashboard, but something that could read, reason, and act. Early conversations with finance leaders across Israel, the US, and Germany confirmed the gap. Close to 100 CFOs and finance managers all said the same thing: yes, the pain is real, and no, there’s nothing trustworthy out there yet. The trust problem with early LLMs – data privacy, accuracy, auditability –  was part of the reason the category was still wide open.

The team iterated through several product directions – FP&A automation, cash flow management, treasury tools – before finding the sharpest wedge: payroll. It was a domain expert who was brought onto the team as the third co-founder that introduced the insight. Payroll is where finance pain is most acute, most recurring, and most underserved by modern tooling. The vision crystallized around an AI agent that could automate the most labor-intensive parts of the payroll cycle, from data collection to reconciliation to compliance checks, while integrating natively with existing systems.

---

“Every finance person we spoke to lit up. No one has built this properly. Design partners were talking about paying $1,000–$1,500 a month for mid-market – and multiples of that for enterprise.”

---

Validation

Validation was a wild west effort. Before the era of LLM-driven research, I performed hundreds of hours of manual competitive analysis, digging through TechCrunch archives and YouTube demos to map the global fintech landscape. I conducted dozens of discovery calls with industry experts while simultaneously running 8 The payroll pivot came with a significant accelerator – the newly introduced third co-founder had a vast network across the Israeli finance ecosystem, and could basically reach any company, entity or finance professional or decision maker in Israel in a manner of one to two phone calls. Design partners were recruited quickly (with Letters of Intent in place), including companies with 100+ employees in both Israel and the US. Early conversations moved fast into specifics: real workflows, real pain points, real Excel files handed over willingly.

Willingness to pay validated from two directions simultaneously. Bottom-up, prospects named numbers unprompted, and top-down, independent cost-savings modeling converged on the same range. The unit economics penciled out clearly. The team also made contact with an American payroll company as a potential design partner, a signal that the idea had legs beyond the local market.

---

“Every finance person we sat down with gave us their actual Excel files. They wanted to show us the problem. That only happens when the pain is real.”

---

Why it didn’t progress

1. A co-founder relationship that was never fully aligned: The venture started with 2 co-founders which later expanded into 3, with the addition of a highly credentialed finance veteran with exactly the domain knowledge and local relationships the company needed. But from the start, there were signs of misalignment, on various subjects. The 3rd cofounder struggled to prioritize the startup over other commitments, and the vision presented for the business diverged from that of the other founders’ from the beginning (startup vs. business; local market focus vs US & global customers; product vision and more). Needless to say, neither vision was wrong – they were simply incompatible.

2. Product direction as a proxy for deeper disagreements: Debates about what to build – which agent first, which market, which customer segment – were really debates about what kind of company this was. Each product decision became a flashpoint. The same external input (feedback from a VC, a design partner’s reaction, a competitor move) would be interpreted differently depending on which vision it seemed to support. Reaching alignment felt temporary; the underlying tension kept resurfacing.

3. The fundraising trigger that broke things open: After months of friction, the team finally got everyone aligned on starting a first funding round. Alas, external feedback questioned the group dynamics, responsibilities and roles. This then resurfaced and enhanced previous trifles and negative feelings about the potential of collaboration, and the long term feasibility of the partnership. The impasse was unresolvable.

4. A dissolution that burned the bridges: When the company folded, the damage wasn’t contained. On the day the decision was made, emails went out to design partners and potential investors announcing the closure. Relationships that had been built over months were severed in a single move. Even if one founder had wanted to continue with a restructured team, the investor relationships and the design partner pipeline were effectively gone. The startup didn’t just stop – it was publicly ended.

---

“We were almost never aligned. On the product, on the market, on how to raise money, on what success looked like. Every time we found a patch of agreement, something came along to rip it open again.”

---

Insights

1. The product was right. The team dynamic was a failure: The payroll AI thesis remains largely unbuilt. Willingness to pay was validated. Design partners were real. The market was moving in the right direction. Nothing about the external conditions killed this – it was entirely an internal story. That’s both the hardest thing to accept and the most important thing to understand.

2. Domain expertise is blinding as well as enabling: Bringing in a specialist with deep knowledge and strong relationships is a genuine accelerant. It’s also a risk. Someone who has spent decades in a field has strong intuitions about how it works – which can make it harder, not easier, to sell a vision of how it could work differently. The most valuable domain expert isn’t someone who knows the current state better than anyone. It’s someone who can hold that knowledge lightly enough to imagine its disruption.

3. A pilot period isn’t optional when the relationship is new: The team moved quickly to formalize the third co-founder relationship without a real working period first. In hindsight, a structured 4–6 week trial – a small discrete project under NDA – would have surfaced the misalignment before equity and expectations were on the table.

4. Founders’ agreements should protect the originating founders: Standard vesting schedules weren’t enough to handle this situation cleanly. When a co-founder brings unique assets to the table – customer relationships, domain reputation, a professional network – the founders’ agreement needs to explicitly address what happens to those assets if the relationship ends. Ambiguity here is the vector for the hardest conflicts.

5. A company dies the day its bridges burn, not the day it closes: The formal dissolution of a startup is rarely the moment it actually ends. This one ended when emails went to design partners and investors before any internal resolution had been reached. Anyone building in a fragile co-founder situation should have a protocol for what gets communicated externally – and who controls it – before a crisis makes it necessary.

---

“The idea is still there. The market is still waiting for it. A founder who wakes up every morning thinking about payroll, with the right team around them, could still build this. We just weren’t that team.”

---