Model ML, a London-based company that builds artificial-intelligence agents for bankers and consultants, has raised $250 million in a Series B round at a $1.5 billion valuation. General Atlantic and QED Investors co-led the financing, which closed a year after the company’s $75 million Series A.
The product does work that has traditionally been given to the most junior people in a deal team: building financial models, assembling pitch materials, drafting due diligence reports. The company’s agents connect to the tools those teams already use, including Excel and PowerPoint, and update the output when something in the deal changes.
That last part is the commercial claim. A model that has to be rebuilt every time a number moves is a document. A model that updates itself when a new filing lands is closer to a system of record, and systems of record are what software companies charge for.
Model ML says it runs on open-source models rather than the frontier systems sold by Anthropic and OpenAI, which it says allows it to operate at about 2% of the cost. That figure is the company’s own and has not been independently verified, but the logic behind it is not exotic: a narrow task with a fixed output format does not need the most expensive model available.
The founders are brothers. Chaz and Arnie Englander started the company three years ago, after building and selling two previous businesses, including Fat Llama, a peer-to-peer rental marketplace acquired by Sweden’s Hygglo for $41.5 million in 2022. That background is unusual among AI founders, most of whom arrive from research labs or from infrastructure companies.
It also explains the sales motion. Model ML has gone after institutions rather than individuals, selling to firms where a managing director controls a budget and a vice president controls the workflow. The buyers are not looking for a general assistant. They are looking for a way to run a deal with fewer people on the team.
That is a sensitive proposition in an industry that has built its training model on pyramids. Banks hire hundreds of graduates, work them hard for two years and keep the survivors. The work they do is repetitive by design, and the repetition is how they learn the business. Automating the bottom of the pyramid removes the apprenticeship along with the spreadsheet.
Investment banks have been buying AI tools anyway, mostly in small deployments around research, compliance and code. The hesitation has been about confidentiality and accuracy, and both concerns are sharper in a live deal than in a back-office function. A model that produces a defensible number is useful. A model that produces a plausible one is a liability with a timestamp.
The competitive set is crowded at the top and thin in the middle. Hebbia sells search and analysis tools to the same customers. Rogo raised $160 million at a $2 billion valuation in April for a similar proposition. The large model vendors sell directly into financial services, and banks are also building internal tools on top of their own data.
The market Model ML is selling into has a peculiar cost structure. Analyst salaries are modest by banking standards, but the cost of a mistake in a model is not, and the review process that catches errors is itself made of people. Software that removes the first draft does not remove the review, which limits how much headcount a bank can actually cut.
Model ML’s answer is workflow rather than intelligence. It does not claim a better model. It claims a better fit to the specific artifacts of a deal, which are spreadsheets with fragile cell references, decks with a house style, and documents built from a checklist that nobody has written down.
Consultancies are the second market and arguably the easier one. A consulting engagement produces a slide deck and a report at the end, and both are assembled by people who bill by the hour. The economics of replacing some of those hours are more attractive to a partner than to a bank, because the partner keeps the fee.
The valuation implies a company worth roughly six times what it raised, which for a three-year-old business with undisclosed revenue is a statement about expectations rather than performance. Investors in this category have been paying for growth rates and for the belief that financial services is the largest addressable market for AI agents outside software development.
What the round does not settle is whether the buyers will change their staffing. A tool that saves a vice president twenty hours a week is adopted quickly. A tool that lets a bank cut its analyst class is adopted slowly, because the analyst class is how the bank produces its next generation of vice presidents, and no managing director wants to explain that trade to a campus recruiter.
Model ML’s own framing avoids the question. Its materials describe removing drudgery rather than removing people, which is what every vendor in this market says and what every buyer wants to hear. The $250 million will be spent finding out whether that framing survives a hiring cycle.


