Synthetic is a $10 million bet that AI can do bookkeeping without the human labor Bench could never escape. The harder question is whether founders will trust it after Bench's collapse left customers scrambling.
Ian Crosby is back in the bookkeeping business, and this time he wants to remove the bookkeeper entirely. Synthetic, his new San Francisco startup, has raised a $10 million seed round led by Khosla Ventures to build an accounting service for software startups that runs on AI rather than a back office full of humans.
That would be a bold pitch from any founder. Coming from Crosby, it lands differently. He co-founded Bench Accounting in 2012 and helped build it into one of the best-known bookkeeping services for small businesses before leaving in 2021. Bench later became a cautionary tale, shutting down abruptly on December 27, 2024, locking customers out at one of the worst possible moments of the tax calendar, then being acquired by Employer.com days later.
According to a Business Wire announcement carried by VentureBeat on May 14, Synthetic's seed round includes Basis Set Ventures and a group of operator investors that includes Shopify CEO Tobi Lütke, Opendoor CEO Kaz Nejatian, Bridge co-founder Zach Abrams, Accrual CEO Cosmin Nicolaescu and Figure CEO Michael Tannenbaum. That is not a random cap table. It is a group of people who understand payments, software, finance operations and the peculiar pain of running a startup's back office.
Synthetic is not trying to become a general accounting firm on day one. The company says it will focus only on software, SaaS and AI businesses, starting with the smallest customers where the accounting is simplest. The service connects to banks, payroll, billing systems and inboxes, asks clarifying questions when context is missing, then produces accrual-basis books that can be handed to a tax preparer.
That focus matters because accounting is full of edge cases that look small until they affect taxes, investor reporting or cash planning. A bakery, a design agency and an AI infrastructure startup may all need bookkeeping, but they do not create the same revenue patterns, expense categories or documentation trails. Synthetic's bet is that a narrow customer base gives the AI enough repetition to become reliable before the product expands into more complicated territory.
The price is also part of the story. Synthetic says plans will start at $49 per month, roughly a quarter of what a human-staffed service costs. That is the promise investors keep circling in AI services: take a function that is expensive because it depends on trained human labor, then use software to make it cheaper, faster and always available.
But bookkeeping is not like drafting a marketing email. A wrong categorization can flow into bad tax work. A missing accrual can distort runway. A confused founder can make decisions from books that look clean but are quietly wrong. That is why Crosby's own caution is more interesting than the price. He has said Synthetic will not release until the company believes it is more reliable than a human bookkeeper, and that the timeline could be six months or six years.
Bench still shadows the pitch
Bench's collapse gives Synthetic both credibility and baggage. On one hand, Crosby has lived the operational reality of bookkeeping at scale. He knows where customers get confused, where workflows break and why even software-enabled services often become expensive labor machines underneath. On the other hand, founders who watched Bench customers lose access to financial records will not treat accounting reliability as an abstract product metric.
TechCrunch reported in January 2025 that Bench's bankruptcy filings showed $65.4 million in liabilities and just $2.8 million in cash at the end of its life. The same reporting noted that Bench had raised $113 million and owed most of its debt to the National Bank of Canada. Those numbers are a sharp reminder that back-office startups can grow large without solving the cost structure that sits underneath the service.
This is where Khosla Ventures' bet becomes clear. Synthetic is not merely another accounting tool with AI features stitched onto the interface. It is trying to replace the delivery model itself. If the work can be made autonomous, the service can be cheap enough for early founders and scalable enough for venture investors. If it cannot, Synthetic risks becoming another company that discovers accounting is stubbornly human in the places that matter most.
There is also a timing question. The AI market has been crowded with tools that promise to automate knowledge work before they have earned real trust. Founders may tolerate rough edges in note-taking, prospecting or code suggestions. They are less forgiving when the output affects taxes, banking relationships and investor updates. Synthetic will need more than a clever interface. It will need controls, audit trails and a way to show customers why the books are right.
For San Francisco's startup world, the story is part redemption arc and part test case. Crosby is returning to the same category with a cleaner thesis, a sharper customer profile and a market far more willing to believe AI can do professional work. The lesson from Bench is that demand was never the problem. Small businesses and startups clearly want bookkeeping to be easier.
The next thing to watch is whether Synthetic can prove reliability before it chases scale. If it does, AI bookkeeping could become one of the more practical examples of automation in a trust-heavy business function. If it does not, the market will remember Bench quickly, because in accounting, the cost of being almost right is still too high.
Also read: Jumia is using AI cuts to chase its first real profit • AsymFlow makes pixel-space image generation look practical again • Wirestock raises $23 million as AI labs hunt for licensed creative data
