London-based scaleup Prolific has today raised £25m to scale its platform, which provides vetted research participants to AI companies that need human feedback to hone their models. The round was co-led by Paris-based VC firm Partech and Oxford Science Enterprises, the independent investor partnered with Oxford University.
Prolific, a Y Combinator alum, tells Sifted that the company has been cashflow positive since 2020, but decided to raise funding in response to the booming demand for human feedback for AI systems since the launch of ChatGPT last year.
What does Prolific do?
When Prolific was founded in 2014, it wasn’t focused on the AI sector at all. The idea was born out of the cofounders’ frustration at not being able to source reliable participants for academic research, with the company initially selling to universities and other research bodies.
The startup vets participants before matching them with organisations that need people for online research, providing them with an easily accessible, reliable and diverse pool of individuals for studies.
Prolific takes a 30% cut of the money spent by the organisation on labour, with participants ending up making $15 per hour on average.
Cofounder and CEO Phelim Bradley tells Sifted that the recent boom in the AI sector has created a new use case for the company, as companies building AI-powered products require human feedback to improve their systems’ outputs.
“Where our sweet spot lies at the moment is the process called reinforcement learning by human feedback,” he says. “How it works is that the AI will spit out multiple outputs and then you get a human to check and let the AI know which one is a better, safer response. Or one that’s more accurate, addressing some of the challenges around hallucinations and bias.”
Bradley says that, in the AI space, Prolific is working with a range of clients including companies building their own foundational models as well as those fine-tuning pre-existing models for specialised uses.
“We work with one company, for example, that’s ensuring their mental health chatbot — which is based on large language models — is safe and able to deal with a wide variety of different situations,” he says. “The diversity of our participant pool is a really critical factor when it comes to that, so that the AI can interact with a bunch of different people from different backgrounds and different ethnicities.”
It's got a blockbuster list of customers, including Google, Meta, Stanford University, the University of Oxford, King’s College London and the European Commission.
What’s the market like?
Prolific isn’t the only company offering research participants to AI companies. Amazon runs a jobs marketplace called Mechanical Turk to connect researchers with participants, while Silicon Valley-based scaleups Surge AI and Scale AI also offer outsourced AI training and feedback services.
Bradley says that Prolific’s focus on high-quality vetting and the diversity of its participant pool — the company has 120k people on its books spread across seven countries in Europe, the US and Australia — make it an attractive partner for AI companies that take bias seriously.
Mechanical Turk, meanwhile, does not vet work candidates as standard according to its own website.
What’s next for Prolific?
Bradley says that, while the company has been cashflow positive since lockdowns drove a surge in demand for online research, this new round of funding will see the company investing in growth and going into loss-making territory once again.
It’s the first time Prolific has raised money since participating in Silicon Valley’s Y Combinator programme in 2019 — Bradley adds that the company hasn’t touched the YC money thanks to its healthy revenue position.
“We were actually in an above-average position to fundraise in the current climate versus other startups,” he says. “Being a very capital-efficient model was suddenly cool again, versus the more pre-revenue, big valuation-type fundraising that was happening prior to that.”
Bradley says the money will be used to focus on growing Prolific’s pool of participants — particularly those with sector expertise, such as legal professionals, who can give more specialised feedback.
He adds that the company will also be increasing its headcount from 100 to 150 .
Given the ever-increasing investor appetite for AI-powered startups, demand for human feedback in AI development doesn’t look to be slowing down any time soon.