Make Being Right Cheap: Inside the $50M Lawsuit Economy Targeting E-Commerce

Adapted from my conversation with Zak Cassady-Dorion on the DTC Revenue Rocket podcast (Episode 11).

There's a line item on a lot of e-commerce P&Ls that nobody puts in the pitch deck.

Ask a founder doing $20M what they budget for legal, and past a certain scale you'll get some version of the same answer: we keep a reserve to settle these things. Demand letters. ADA claims. Privacy claims. It's treated as weather — something that happens to you, that you insure against, that you don't really fight.

I've spent three years building a company around the conviction that this is a broken system, and that the fix isn't legal. It's technical.

Here's the argument.

The math of the demand letter industry

Start with how these cases actually get made, because most operators only ever see their own letter and assume it's about them. It isn't.

A law firm — often the same handful of firms, often filing out of New York — recruits a plaintiff. That plaintiff's name goes on fifty, a hundred, two hundred filings. When a case settles, the plaintiff typically walks away with $500 to $1,000 for being the face of it. The firm keeps the rest: nine, ten, twelve thousand dollars, over and over.

Last year the industry saw roughly 5,000 web accessibility lawsuits. We expect more this year. Assume they settle in the $10,000–$15,000 range, and you're looking at something like a $50 million public settlement market. But filings are the visible tip. If a fifth of demand letters convert into a filing or settlement, the real volume is closer to 20,000 letters a year landing in inboxes belonging to people who have never heard the phrase "WCAG 2.2 AA."

That's the part that gets me. Accessibility exists so that a blind customer can buy your product. It's a market expansion story. And it has been almost completely co-opted into a fee generation machine that produces more anxiety than accessible websites.

The existential question: why settling is rational

Here's the trap, and it's worth sitting with because it explains everything about how this market behaves.

Suppose your site is genuinely accessible. Suppose you did the work. You still get a demand letter — plaintiff-side volume filing doesn't audit carefully before it fires. Now: do you fight?

To prove you're right, all the way through, you're hiring counsel. Call it $15,000 to $25,000 in fees, plus months of your attention. The other side is asking for $5,000 to make it disappear.

Being right costs four times more than being wrong. So you settle. Everyone settles. And the machine keeps running, because the economics guarantee it will.

That's the existential question our whole product roadmap is built around: what would it take to make the cost of proving you're right fall below the cost of settling? Not just for ADA — for every category where volume litigation has found a foothold in e-commerce. Consent and privacy under CIPA and CCPA. Strikethrough pricing and how long you're allowed to run a "sale." Subscription auto-renewal disclosure. Same structure, same incentives, same reserve line item.

Nobody is attacking that surface systematically. That's the company.

Why the two existing options both fail

When a brand starts looking for a fix, they find two things.

Option one: the overlay widget. One line of JavaScript, a little accessibility icon floating in the corner, toggles for contrast and text size. accessiBe, UserWay, AudioEye. Installation takes four minutes, which is exactly the problem — it's a visual layer sitting on top of a page whose underlying code hasn't changed.

Overlays address somewhere in the range of 40–60% of real issues. They cannot fix a broken heading hierarchy, because heading structure lives in your markup. They cannot write alt text that describes your product. They cannot turn forty links that all say "click here" into links that tell a screen reader user where they lead. Those are semantic problems, and a widget rendering over the top of your DOM has nothing to say about them.

Option two: the manual audit. You hire Level Access, Allyant, Deque, 216digital. Skilled humans test your site properly and hand you a report. This is more rigorous and dramatically more expensive — and it ends at the exact moment the hard part begins.

My co-founder lived this. He was a new grad engineer at Twitch when the ADA claim came in, and his job was to fix it. What he got was a PDF. Hundreds of findings, no tooling, no automation, no remediation path. Nobody learns accessibility in a CS program. You learn Python and SQL and data structures; ADA compliance is a forgotten topic that becomes your emergency at 2 a.m. He looked at that report and thought: what am I actually supposed to do with this?

That question is the reason TestParty exists. We wanted the installation ease of an overlay with the rigor of manual testing, delivered as fixes in your source code rather than findings in a document — reviewable, mergeable, and running continuously rather than once.

Why "continuously" is doing real work in that sentence

The obvious pushback: I fixed my site, why am I paying you monthly?

Because good e-commerce teams never stop touching their site. You're A/B testing constantly — half the time nobody knows in advance what will convert, so you ship variants and let the data decide. Perfectly accessible today, six months of color changes and template tweaks and app installs later, you've regressed. Not through negligence. Through velocity.

I think of it like cleaning a room. I would love a once-a-year model. A once-a-year model produces a room you cannot enter. So you clean on a cadence, and the cadence is what keeps you in good shape.

The part nobody expects: accessibility is a revenue channel

This is where the conversation usually turns, because compliance is a cost story, and cost stories are boring.

Consider what a screen reader actually is: a machine that parses your page structure and tries to understand what everything means. Semantic headings. Descriptive link text. Labeled form fields. Alt text on images. Now consider what a search crawler does. What an LLM does when it's answering "best waterproof duffel under $200" and deciding whose product to name.

The same code changes that make a page legible to a blind customer make it legible to a machine. Accessibility and AI search readability are the same engineering work wearing two different hats.

We ran a study across our first hundred customers — Shopify analytics, three months before TestParty against three months after. Average organic and AI search traffic lift: 14.5%.

Put numbers on it. Say you're getting 1,000 ChatGPT-driven sessions a month. Post-remediation you're at roughly 1,140. Apply a 10% conversion rate to the incremental 140, and that's 14 new orders. At a $100 AOV, $1,400 in net new monthly revenue — from work you were considering purely as legal insurance.

At the top of the market it stops being a rounding error entirely. We work with several nine-figure brands. A 2% lift in organic traffic is tens of thousands of dollars a month; the annual contract pays for itself in weeks, and the lawsuit avoidance becomes a free byproduct.

My favorite version of this: a footwear group running five separate storefronts. We do biweekly check-ins for the first couple months of a partnership, and on one call the operator says he's been trying to figure out why search traffic is up 22% year over year, because as far as he can tell they haven't changed anything. We pulled up the timeline together. The inflection point was the exact day our code changes went live.

Obvious to us. Genuinely a surprise to him. Which is a marketing failure on our part more than anything else — so consider this me correcting it.

How we actually build: agent loops, not prompts

We're an AI-native company in a specific sense that's worth unpacking, because "AI-powered" has been thoroughly drained of meaning.

Most people's mental model of AI is level-one prompting: you ask for a thing, you get the thing. "Write me this email." "Generate this image." One prompt, one output, human evaluates.

What's changed — and what I think is the actual frontier — is the agent loop. You don't specify the action. You specify the end state, and the system iterates toward it on its own: act, evaluate, act again, until the condition is satisfied.

Accessibility remediation is almost a perfect fit for this, because our success condition is unambiguous. Scan the site. Find violations. Generate source-level fixes. Rescan. Repeat until the scan comes back clean. There's no fuzzy judgment about whether the output is "good" — either the violation is gone or it isn't.

Two years of remediation data on what a correct fix looks like makes those loops precise rather than merely energetic. And every change still goes through human review before it merges. We're not at the point — nobody serious is — where you let an agent push to a customer's production codebase unsupervised. AI plus human in the loop, at a scale a human-only team could never reach.

If you're a DTC brand that feels behind on AI

I talk to a lot of operators who are somewhere between "I use it like a search engine" and nothing at all. Three patterns are worth stealing.

1. Creative velocity. If you have a design system and a brand framework, generating variants is no longer a hiring decision. Claude plus Figma for design is unreasonably good. You can produce a hundred creative tests where you used to produce three, push them into Meta, and let the analysis of what's working also be automated. We use the same approach on outbound; iterating cold email with AI took our response rates from invisible to the 5–20% range.

2. Stop drowning in your own dashboard. There are more metrics in a Shopify admin than any human will ever synthesize. Export the data, hand it to Claude, and ask the unglamorous question: what am I missing here, and where's the opportunity? For a large class of companies that is a data analyst you no longer need to hire.

3. Build institutional memory on purpose. We record every call and pipe it into Claude through an MCP server, alongside notes from in-person meetings and our HubSpot records. So I can ask for a synthesis of every conversation we've ever had with a given customer before I walk into the next one. When someone joins, I tell them to ask Claude first and me second — at this point it has more context on the company than I do. It writes our case studies, powers onboarding, and lets us query where the business stood at any point in its history.

That third one only works if your underlying data source is strong. Ours is recorded meetings. Yours might be Shopify, your helpdesk, your reviews. The AI layer is increasingly commoditized; the proprietary corpus you point it at is not.

And if you don't know where to start: ask the model. Genuinely. Once a month I ask, given everything you know about me and this market, what should I build next? Then I take the ten answers, open ten tabs, and tell each one to go build its idea. Half an hour later I have ten things to evaluate. We over-index on needing the perfect idea before we're allowed to begin; the tool is broad enough to generate its own starting points.

The one tool I'll evangelize unprompted is Clay. Think Excel where every column is an AI prompt — you feed it rows of customers and add a column that reasons over each one. Given this person's purchase history, location, and spend, what should we offer them next? We use it for outbound and enrichment; a DTC brand should be using it to build hyper-specific segments and personalized offers. "Everyone who bought last month, lives in Massachusetts, spent over $100, and whose prior purchase was six months earlier" is now a thirty-minute build, not a data project.

The end of hourly billing

A prediction, since Zak asked for one: if your agency still bills by the hour, I don't think you exist in 24 months.

The reason isn't ideological, it's that the argument becomes unwinnable. You tell a client something took three hours. The client already tried it in Claude Code and got 80% of the way in thirty minutes. Now you're litigating your own timesheet instead of your value.

Clients were never buying hours. They were buying an outcome — a site that converts at 2%, a channel that produces revenue. The move is toward outcome-based pricing, and it's genuinely harder than it sounds, because scope becomes unbounded and someone has to absorb the token costs of getting there. But the upside is a margin profile that traditional services never allowed: if the client values the outcome at $10,000 and you've gotten efficient enough to deliver it for $500, that spread is yours. That arbitrage — and the risk that inference prices move against you once you're dependent — is the central strategic question for every services business right now.

We are earlier than it feels

I'm in this every day, so I have to actively remind myself how unrepresentative that is.

My brother has over a million followers across his platforms and is enthusiastically anti-AI — he's convinced his reach depends on everything being made by hand, and he may well be right. Both my siblings are physicians. When you spend your working life with people at the end of theirs, technology is not the thing that seems important. That's a useful corrective for someone like me.

My best guess is that only 5–10% of people are using these tools daily at anything close to their capacity. Anthropic has a few thousand employees; Microsoft has north of a hundred and fifty thousand. The gap between what's possible and what's practiced is enormous, even among people who think they've figured it out.

Which is why the highest-leverage advice I have for any founder is embarrassingly simple: find a friend who is completely AI-pilled and talk to them once a week. Someone at Anthropic or OpenAI. DM them. Offer to pay them $500 a week just to screen-share how they actually work.

I had dinner with a friend who'd joined Anthropic three weeks earlier and asked how they use Claude internally. What he showed me made it clear I'm operating at level one. None of it is on Twitter. By the time these practices go public you're months behind, and months is a long time right now.

What this comes down to

Every argument above collapses into one:

Accessibility done properly is not a compliance expense. It's a revenue channel that happens to make you a bad target for opportunistic litigation.

The lawsuit economy exists because being right is expensive. Our entire job is making it cheap — by fixing the underlying code, keeping it fixed as your site evolves, and producing the continuous evidence that you did. The 14.5% search lift is what you get for free along the way.

TestParty remediates WCAG 2.2 AA violations at the source-code level for e-commerce brands — no overlay widget, no PDF report to hand your engineering team. Get a free accessibility audit of your store →

Listen to the full episode of the DTC Revenue Rocket podcast with Zak Cassady-Dorion.