Fable has gained a reputation as the go-to startup for helping companies build digital products that are more accessible to people with disabilities. After raising $25 million in new funding, the Toronto-based startup is now expanding the communities it supports and working to make AI training data more inclusive.

Fable started in 2020 as a way to consult more easily with accessibility experts and people with disabilities, so anyone building a product could have the best advice on making it as accessible as possible. Since its A round two years later, it has been building out a more robust product offering, with content, testing, and tools that integrate more directly with developers’ workflows.

“Accessibility is no longer just the responsibility of the accessibility specialists or product managers — it’s shifting to everyone in the product development process: researchers, designers, product managers, engineers,” said CEO and co-founder Alwar Pillai in an interview with TechCrunch. “It’s historically been another group that’s been responsible for it and doing the work. Now they’re taking that shared accountability and responsibility, and Fable is a platform that’s allowing them to do it by themselves.”

Companies building and maintaining large platforms have come to accept to the fact that you can’t sprinkle accessibility on at the very end like Salt Bae. It has to be baked in from the start — and making the product better for people with disabilities usually makes it better for everyone else, too.

Fable initially supported vision and motor disabilities to get a warm start, Pillai explained:

“We wanted to find a way just to get companies comfortable with engaging with this population, because it has historically been excluded, and these were communities that organizations are a bit more familiar with. over the course of time, we’ve observed a couple of trends within our customer base that made it like this is the right time for us to expand our community to represent people with cognitive disabilities and with hearing disabilities.”

One in six people has some form of disability, she noted, though not all are visible or even something a person may mention. There are lots of assumptions built into user experiences, about what a user can see, hear, do, and understand. Finding and improving them is not always easy if you don’t have, say, a deaf person or someone with dyslexia on your testing team.

“If you get the insights from these communities, you are going to, at the end, make products that work well for everyone. But it has been historically challenging for enterprises to engage with this community easily, and on demand. And that’s when Fable jumps in,” she said.

Over the years, Fable has built out assessment tools as well as advisory ones, so product managers and engineers can track accessibility the way they do other standard function and quality milestones.

A new area of tech that deserves special attention, and which Fable is hoping to improve, is the data powering AI models. Bias in data translates to bias in models, and that’s as true for people with disabilities as it is for other categories.

One reason for this is that these AI models tend to aim for the most aggressively average answer or response — dead center on the bell curve. But people with disabilities tend to fall outside that average need or experience.

“We’re excited and cautious about the proliferation of AI; there’s a huge opportunity to make experiences better for people with disabilities,” said Pillai. “But at the same time, it also has the ability to amplify the digital divide that exists. We’re seeing AI getting baked into so many things, but because people with disabilities have not been taken into account there, and the data that you probably collect is smaller, so they get excluded from the large models, we think they have the ability to exclude the experiences of people with disabilities because it deviates from the ‘normal.’ “

This can be mitigated with fine-tuning or prompt engineering, but only so much; a model necessarily pulls from the datasets it’s trained on, so if disabilities are not adequately represented in there — and they aren’t — the models simply are not equipped for accessibility. Fable has been working with the community, as well as researchers and governments, to create resources and best practices.

“Our goal is that, in the near future, we’re able to introduce these inclusive data sets and offer testing for accessibility in AI. Our customers are already coming to us for it,” Pillai said.

She emphasized that, as before, it’s about empowerment to include these methods in a company’s own development processes — Fable doesn’t do the work for them.

“Our platform now has become this dashboard where you can monitor all of your digital properties and products against these accessibility metrics. We invested in integrations, because we wanted the insights and data to live in the products that product teams are using on a daily basis,” Pillai said. “We’ve gone from just being able to get one piece of insight to really being able to observe your performance across products, and when you think about enterprises, they have, like, 500-plus digital products. The goal for them is, how do I know if I’m getting better or not? And finally, they have metrics to prove it.”

The $25 million B round, led by Five Elms Capital, will be put towards standing up the new teams and products around cognitive and hearing impairments, and of course AI expertise as well. Pillai said that, while the investment climate isn’t as open as it was a few years ago, they were pleasantly surprised while they were raising.

“It was so different from the last few times we raised,” Pillai recalled. “I remember when we raised our seed and series A it was very much, you know, trying to convince investors about the opportunity around the accessibility space. But this time around, investors had a very strong understanding of the space, the growth opportunity. It was more about, you know, how much value are you adding to customers, and how are you growing? I think that stood out.”