Curvestone AI raises $4M seed to fix the "compound error problem" in regulated AI workflows
December 4, 2025
Curvestone AI, the London-based workflow automation platform for regulated industries, has raised $4 million (around £3 million) in seed funding led by MTech Capital, with participation from Boost Capital Partners, D2 Fund and Portfolio Ventures.
The company is going after a problem that has quietly capped real-world adoption of generative AI in regulated work: when individual AI steps run at 98% accuracy and get chained across a dozen operations, end-to-end accuracy can drop to 30–40%. In compliance, legal review or mortgage suitability work, that is not a tolerable trade-off. Curvestone calls this the "compound error problem" and has built its platform around solving it.
The product is model-agnostic and slots into existing CRMs, document management systems and loan origination tools rather than asking teams to migrate. Users configure multi-step workflows with human-in-the-loop review and audit-ready outputs. The company is ISO 27001-certified and works across legal, financial services and insurance. Co-founder Breanna Yen leads on data strategy alongside brothers Dawid Kotur (CEO) and Sebastian Kotur (CPO), who previously ran AI automation programmes at PwC, Metro Bank and GKN.
Curvestone reached profitability before raising any outside capital and grew revenue seven-fold in the twelve months leading up to the round. The platform now processes billions of tokens each quarter. Customers include Stephenson Harwood, Browne Jacobson and Walker Morris — where service agreement reviews have gone from four hours to fifteen minutes — along with mortgage network Pivotal Growth, which is moving from spot-checking around 10% of cases to reviewing all of them across its twenty advice teams.
Kevin McLoughlin, partner at MTech Capital, is joining the board. The fresh capital will go on product development and go-to-market expansion, with a focus on extending the library of validated workflows and deepening integrations with the systems regulated teams already run on. The longer-term pitch is to become the dependable automation layer for agentic AI in sectors where getting it wrong is expensive.
Sources
Thank you to our Patrons & Partners:
