Bench Accounting was once one of the most recognized names in automated bookkeeping for small businesses. It raised $135 million from investors including Shopify and Bain Capital Ventures, served over 12,000 customers at its peak, and employed roughly 600 people. On December 27, 2024, it shut down without warning. Customers found themselves locked out of their financial records, during one of the most critical financial periods of the year, with a bank reportedly calling in its loans after Bench had only a few hundred thousand dollars left in its accounts.
Ian Crosby, who co-founded Bench and led it for years, was not there for the end. The board had removed him in 2021, three months after he declined a $250 million acquisition offer from fintech company Brex. Crosby publicly blamed the collapse on the board's decision to replace him, and Shopify's COO publicly agreed. Whether or not that narrative is complete, it is the story Crosby is carrying into his new venture.
The new startup, called Synthetic, is targeting something that no one in this space has actually delivered: a bookkeeping service with no humans involved at all. Not AI-assisted humans. Just AI. Specifically, Synthetic wants to produce accrual-based financial statements — the kind that show not just what cash moved, but what was earned and owed — entirely without human review. That is a genuinely harder problem than what most tools do today.
For context on why this matters: Bench's old model relied on human bookkeepers doing most of the actual work, with software helping to organize it. That hybrid approach proved expensive to run and difficult to scale. ScaleFactor, another startup that raised over $100 million on the promise of AI-driven bookkeeping, quietly relied on outsourced human accountants in the Philippines when its AI failed to perform. It shut down in 2020 after the gap between its marketing and its actual product became impossible to hide. The pattern in this space is real: the promise is easy to sell, and the delivery is brutally hard.
Crosby is at least being upfront about that. He has said publicly that today's AI models still make significant bookkeeping mistakes, that his prototype only works for a narrow group of users, and that he cannot yet predict how it will hold up at scale. He is planning to focus initially only on AI and software startups, which have simpler, more predictable financial structures than most businesses. His plan is essentially to wait: use the early funding runway to stay operational while AI models continue to improve, and expand only when the technology is genuinely reliable.
Khosla Ventures is a firm built on exactly this kind of long-duration, technically uncertain bet. They backed OpenAI, Stripe, and DoorDash early. They have stated publicly that they expect roughly 70 percent of their investments to fail, and they price that into their strategy. Their partner Jon Chu spoke explicitly about backing founders who have been publicly written off, citing the example of Parker Conrad, who was forced out of HR software company Zenefits in 2016 amid controversy, then went on to build Rippling, now valued at close to $17 billion.
The comparison has some logic to it, but also some limits. Conrad's situation at Zenefits was different: the product worked, the company was growing, and the controversy was largely about compliance failures in sales, not about whether the core technology was viable. Synthetic is asking investors to believe both in the founder's recovery arc and in the eventual technical feasibility of something that does not yet fully exist.
What makes this worth watching for businesses outside the startup world is the direction of travel. Khosla is simultaneously investing in AI-powered bookkeeping startups and exploring buying traditional accounting firms outright and rebuilding them with AI inside. The accounting profession faces a structural staffing problem — estimates suggest that a large portion of qualified accountants will retire within the next decade. That creates real demand for automated alternatives. Whether Synthetic specifically succeeds is genuinely uncertain. That the pressure on human bookkeeping roles will intensify over the next five years is not.