A 130-person company in Palo Alto is rewriting how billion-dollar banks ship software. The trick isn't a slogan. It's a scanner that sees what humans miss.
The flagged element is a button. A button no human tester would have caught - the label reads correctly on screen, the click works, the analytics fire. But the screen reader sees a blank rectangle. Six years ago, that bug would have shipped, then waited for a class-action lawyer to find it. Today it gets fixed before lunch.
That is, in a sentence, what Evinced does. The company sells AI-powered tooling that finds, clusters, and tracks accessibility defects across web and mobile apps. Its customers include Verizon, Hyundai, Progressive, Amazon, and - according to the company - six of the ten largest banks in the US and UK. It raised $55 million in December 2024 to bring the platform to Europe. Total raised: $112.5 million. Headcount: 130. The founder, Navin Thadani, will tell you the company is just getting started, which is the sort of thing every founder says, except in Evinced's case the regulatory and AI tailwinds make it hard to argue with.
Around 1.3 billion people live with some form of disability. The Americans with Disabilities Act, the European Accessibility Act, WCAG 2.1, Section 508 - the rulebooks are dense, the lawsuits are expensive, and the engineering culture has historically treated all of it as a tax. Pay your auditor, file a VPAT, move on.
The trouble is that auditors look at a release. Codebases ship every hour. A retail bank's app might have 40,000 components, half of them rendered dynamically by React, half of them gated behind login walls auditors can't even reach. Manual testing finds maybe 30% of issues, and only after the code is already in production. Which means most "accessible" apps aren't, and most accessibility programs have been performing compliance theater for two decades.
Evinced's founders, Navin Thadani and Gal Moav, had a front-row seat to this gap from their old jobs at Oracle. Thadani had spent years running products there. Moav was running an Oracle Israel R&D org. They were not, by background, accessibility people. They were enterprise software people who noticed that a deeply important problem was being solved with comically wrong tools.
The bet, in 2018, was that you could use computer vision and machine learning to look at a screen the way a human user would - not just parse the underlying HTML. Most accessibility scanners at the time read code. They flagged missing alt attributes and wrong ARIA roles, which was useful but shallow. They missed the bugs that mattered: the focus that vanished, the modal that trapped your keyboard, the color contrast that worked in the staging env and broke in production because someone changed a CSS variable.
Evinced's first product analyzed user flows - it actually moved through an app the way a person would, watching DOM mutations, comparing renders, clustering related defects into single issues so a developer wouldn't open Jira to 4,000 tickets and quit on the spot. The output looked less like an audit and more like a code review. Developers found it useful. That, in B2B software, is the only thing that matters.
Insight Partners led the Series A in 2021. Microsoft's M12 came in. Capital One Ventures wrote a check, which doubled as a customer endorsement. The company's customer base tripled in 2023. Then, in December 2024, Insight led a $55 million Series C, with Vertex Ventures joining as a new backer. The deck was reportedly thin on hype and thick on retention numbers. Insight didn't blink.
A six-year scrapbook in five frames. Most startups never get the third one.
The Evinced platform isn't a single app. It's a set of tools that plug into every stage of the modern software life cycle, from Figma to staging to production. The thesis: a designer should catch the contrast bug, the unit test should catch the role bug, the integration test should catch the flow bug, and the runtime monitor should catch whatever everyone else missed. The lawyer should never get a chance.
Walks through web and mobile apps like a user. Catches what static scanners miss.
Drop-in plugins for Selenium, Cypress, Playwright, Appium, Espresso, XCUITest.
Continuous crawl of an entire production site. Tracks regressions over time.
Live in-browser diagnostics. Pinpoints the bug and offers a fix in the same panel.
Catches accessibility defects in Figma before a single line of code is written.
Component-level checks inside the IDE. The accessibility equivalent of a linter.
So AI coding agents can plan, build, and test with full accessibility analysis.
In-IDE assistant inside VS Code. Answers a11y questions in your codebase context.
Eight products, one promise: stop shipping bugs that you'd never tolerate in any other domain.
Evinced reports that its scanners find roughly three to four times more accessibility defects than legacy static-analysis tools, with a fraction of the false positives. That's a self-reported figure, so take it with the usual salt, but the customer mix backs it up. Banks are not impulse buyers. They don't roll out the same vendor across multiple business units unless the tooling is paying for itself.
The numbers are vendor-reported. The customer list is harder to fake.
It is unfashionable to take a company's mission seriously. Most of them deserve the eye-roll. Evinced is the rare outlier where the mission is, in fact, what the company does on a Tuesday. The product directly increases the number of people who can use a banking app, an airline app, a government portal, a grocery checkout. The work isn't theatrical. It compiles.
There is also, helpfully, a strong commercial wind at their back. The European Accessibility Act came into force in June 2025, exposing thousands of companies to enforcement risk. ADA Title III lawsuits in the US continue to climb, with web accessibility cases dominating the docket. A useful pattern: the best regulatory tailwinds are the ones the founder didn't have to plan around.
Here is the interesting plot twist. The next wave of software won't be written by humans alone - it'll be drafted by AI coding agents, reviewed by humans, and shipped at a velocity that already breaks current QA pipelines. Evinced's bet on Model Context Protocol tools is that the only realistic way to keep AI-generated code accessible is to teach the AI itself. Plug an Evinced MCP into Claude, Cursor, or any agent that respects the protocol, and the agent now reasons about accessibility on every keystroke. The accessibility check moves from the test stage to the planning stage.
This is the natural endgame for the category. Static scanners told developers what was wrong after the fact. Evinced moved the check left into design, then into the IDE, then into the agent's prompt. The arrow only points one direction.
Six years ago, she'd have shipped the bug. A user with a screen reader would have hit it. A complaint would have been filed. A class-action notice would have arrived. A consultant would have been hired, an audit would have been commissioned, a remediation roadmap would have been drafted. Eighteen months later, the bug would still be there, joined by 4,000 new ones.
Today, the engineer's IDE flags it before lunch. A chatbot suggests the fix. The commit gets through review. The screen reader user, somewhere else entirely, opens the app and the button just works. Nothing dramatic happens. Which is exactly the point.
That is the work. Evinced is doing it.