Alex Dalyac was turned down. The programme was Entrepreneur First, the London talent-investor that backs people before they have an idea, and the reason was blunt: he did not know enough about technology.
He had been drawn to the field by a Coursera course. Six months before deep learning became the thing everyone talked about in 2013, he took Geoffrey Hinton's neural network lessons and, as he later put it, fell in love. At Imperial College London he and teammates built a plant-recognition app, and he spent spare hours on image classification.
What almost nobody realises about Tractable is where its famous insurance product did not come from.
So he did the unglamorous thing. He went away, learned, and reapplied. This time he was accepted.
What almost nobody realises about Tractable is where its famous insurance product did not come from. At an EF hackathon Dalyac met Razvan Ranca, a machine-learning researcher out of Cambridge, and the two coded through the night. In 2014 they started the company. The first paying customer was not a car insurer, or even a car park. They were a business that used the software to inspect plastic pipe welds.
That customer, Dalyac wrote, stopped working with them around the time of their first funding round. Investors, for their part, thought pipe inspection was too small a market. The founders found themselves with a technology that could see and no clear thing to look at. They tried utilities, geology, dermatology and medical imaging.
Help came from a third founder. Adrien Cohen, who had co-founded Lazada before it was sold to Alibaba for $1.5bn, joined in 2015 and taught the two engineers how to build a business and hire.
The company that came out of that search appraises damage on cars and homes from photographs, so that an insurer can settle a claim in hours rather than weeks. In June 2021 a venture round valued it at $1bn, making it Britain's hundredth unicorn, with more than 300 staff. In July 2023 SoftBank's Vision Fund 2 put in $65m.
There are limits that Tractable itself would admit: the AI sees only visible damage, not what has twisted inside a chassis or behind a wall.
For a founder, the quiet lesson is in the first two chapters. One rejection was treated as homework, and the first customer was treated as a clue rather than a destination. The business that made them rich came from the years they spent being wrong about where to point the same idea.