Case study: B2B SaaS with rich interface and low agent readability

Reading note
This case study synthesizes situations observed in SaaS, product and documentation environments. It is a reading model, not a promise of universal results.

A B2B SaaS company has a modern marketing site, rich documentation, a demo interface and several conversion journeys: demo request, free trial, calculator, use-case filters, comparisons and help centre.

For a motivated human, the site works. You can click, test, read and compare. But for an AI agent tasked with understanding the offer or completing a simple task, the experience becomes fragile quickly.

Initial situation

The interface has been optimized over the years for appearance, conversion and deployment speed. Several frontend components were added campaign after campaign: interactive cards, dynamic filters, capture modals, multi-step forms, customer case carousels and enriched menus.

The symptoms are not immediately visible:

• the site looks fast;
• the pages are well designed;
• the forms work for humans;
• there are many CTAs;
• the product demonstrations are visually convincing.

But below the surface, several signals are unstable:

• buttons are rendered as styled <div> elements;
• some fields have no programmatic labels;
• filters change state without a clear URL;
• modals capture focus without exposing it correctly;
• the primary CTA appears after hydration;
• important content changes depending on viewport size;
• the DOM does not always reflect visual order.

The site is not broken. It is fragile as an action environment.

What the audit reveals

The Agentic Readiness Audit reveals that the problem is not one component. It is an accumulation of divergences between what the human sees, what the DOM exposes and what the accessibility tree makes readable.

Three risk families stand out.

1. Wrong-action risk

The agent identifies a button visually, but the role, accessible name or relationship to the right content block is missing. In a list of plans, modules or customer cases, it may not know which button corresponds to which offer.

2. Context-loss risk

After a modal opens, a filter is selected or a form is partially submitted, the agent does not always receive a clear signal: success, error, next step or return to the initial context.

3. Post-hydration divergence risk

The initial HTML indicates one structure. JavaScript produces another. The site remains usable for a human, but the agent may plan its action on a page version that is no longer the one it manipulates.

Intervention trajectory

The correction is not about adding AI to the site. It is about making the interface more deterministic.

1. Prioritize critical journeys: demo request, free trial, download, plan comparison and documentation search.
2. Compare initial HTML, hydrated DOM, screenshot and accessibility tree on each journey.
3. Refactor critical actions into native elements: real buttons, real links, connected labels, explicit states.
4. Reduce layout changes around action zones.
5. Stabilize forms: field names, errors, confirmations, steps and relationships.
6. Document risky components to prevent regressions in the product roadmap.

When the site relies on heavy frontend architecture, this trajectory may lead to a more restrained machine-first website architecture: critical content rendered early, partial hydration and better isolated interactive components.

What becomes observable

After remediation, the site does not necessarily become more spectacular. It becomes more reliable.

Humans benefit too: forms are clearer, errors are more readable, buttons are more coherent and journeys are less unstable. Agents get cleaner signals for understanding the offer, choosing a CTA, filling a form and maintaining context.

The Agentic Readiness Score then becomes a monitoring tool: not to celebrate a grade, but to watch for regressions after each interface evolution.

Why this case is useful

B2B SaaS companies often have rich interfaces and abundant corpora. That richness can become an obstacle when it is not structured as a readable, stable and actionable environment.

The risk is not that agents cannot see the site. The risk is subtler: they see it, but they do not always know which action is legitimate, reliable or attached to the right context.

This case shows why agentic navigability should be treated as an interface discipline, not as a simple SEO add-on.