CASE STUDY 7 · AGENT READINESS
When AI Agents Could Not Use a Fintech Website — And How Fixing the Accessibility Tree Unlocked New Leads
A Mumbai-based B2B fintech company offering automated GST reconciliation software was structurally unreadable by AI agents. Their website had 62 empty buttons, 48 unlabelled form fields, and all pricing content rendered client-side in JavaScript. When Perplexity's agent tried to interact with their site, it could not complete any action. After an agent readiness overhaul, the company appeared in AI recommendations for three previously uncaptured buyer queries.
| Sector | B2B Fintech / GST Reconciliation |
|---|---|
| Location | Mumbai, India |
| Anonymised | Yes — client name withheld |
The Challenge
The company had built a technically sophisticated product and a visually polished website. Their design team had used modern React components with custom styled div elements for all interactive controls. The result looked excellent in a browser. But AI agents, which read the accessibility tree rather than the visual rendering, encountered a site full of unnamed elements — buttons that said nothing, forms with no labels, and content that only appeared after JavaScript executed.
The CEO became aware of the problem when a prospect mentioned that they had asked an AI agent to "check the pricing on your competitor's site and yours" — and the agent had successfully extracted competitor pricing but returned an error for their website.
What the Diagnostic Found
62 buttons across the website had no accessible name — no visible text and no aria-label attribute. AI agents reading the accessibility tree saw 62 interactive elements they could not interpret.
48 form input fields had no associated label element and no aria-label. The contact form, the demo request form, and the pricing enquiry form were all inaccessible to AI agents.
All pricing content was rendered by a React component that executed after page load. When AI crawlers fetched the page, they received the pre-JavaScript HTML — which contained no pricing information.
robots.txt was clean and all AI crawlers were permitted. This was not a crawler access problem — it was a structural readability problem that occurred after the crawlers got in.
What Dattva Did
The Dattva Agent Readiness Diagnostic had produced the full fix list with specific code-level instructions for each structural failure. The work was handed to the development team as a prioritised fix list. The Dattva team did not write the code — they identified every specific structural failure and specified the exact fix for each.
Priority 1: all 62 unnamed buttons received either visible text labels or descriptive aria-label attributes. The most critical fixes were the CTA buttons — "Get Started," "Book Demo," "See Pricing" — which AI agents had been reading as void elements.
Priority 2: all 48 form fields received for/id associations linking each label to its input. The demo request form was also converted from client-side validation to server-side, removing the JavaScript dependency that had been blocking agent form completion.
Priority 3: pricing content was server-rendered for the initial page load, with the React component hydrating on top of it. This meant crawlers fetching the page without JavaScript execution received the pricing information in the base HTML.
The full fix was implemented in three development sprints over five weeks. The PageSpeed score improved as a secondary benefit — faster initial render from server-side content improved Core Web Vitals.
The Lesson
AI agents do not see your website the way a human does. They read the accessibility tree — a structural model that maps every element by role, name, and state. A button with no name is invisible to an agent. A form with no labels cannot be filled. Pricing that only appears after JavaScript executes does not exist to a crawler. These are not edge case bugs — WebAIM research found 95.9% of the top one million websites have at least one of these failures.
Learn more about Agent Readiness.
FAQ
What is the accessibility tree and why do AI agents use it?
The accessibility tree is a structural representation of a web page that maps every element by its role (button, input, heading, link), its name (what it is called), and its state (disabled, checked, expanded). Screen readers use it. AI agents use it. It strips away visual design and exposes only the functional structure. A visually beautiful website can be completely unreadable in the accessibility tree if the underlying HTML is not semantically structured.
Does AI agent readiness affect SEO?
Yes, indirectly. The fixes required for agent readiness — semantic HTML, server-rendered content, named interactive elements — also improve Google's ability to crawl and index the page. The PageSpeed improvement from server-side rendering improves Core Web Vitals scores, which are a Google ranking factor. Agent readiness and SEO share the same underlying technical requirements.
How common are these accessibility failures in Indian B2B SaaS companies?
Extremely common. Modern React and Vue.js-based websites frequently use styled div elements instead of semantic HTML, client-side rendering for key content, and custom component libraries that do not pass accessible names through to the DOM. This is not a deliberate choice — it is the default output of fast-moving frontend development teams that prioritise visual design over structural semantics. An agent readiness audit identifies these gaps in under 30 minutes.
What is the business impact of agent readiness beyond AI visibility?
AI agents are increasingly used in B2B procurement — researching vendors, comparing pricing, filling out demo request forms. A website that agents cannot navigate is excluded from this procurement workflow. As enterprise buyers increasingly delegate initial vendor research to AI agents, agent readiness becomes a direct revenue issue, not a technical nicety.
See where your brand stands in AI answers today
Run a free AI Visibility diagnostic across ChatGPT, Perplexity, Gemini, and Claude — with prioritised, copy-paste fixes at no cost.