Google Analytics and the Agentability Gap: What 100 SaaS Audits Reveal About AI-Ready Interfaces

The rise of AI agents promises to transform how software gets used, but a comprehensive audit of 100 leading SaaS products reveals that most platforms remain unprepared for agent-driven interaction. Among the products examined, Google Analytics exemplifies both the challenges and opportunities facing analytics tools in an agent-first future.

Understanding Agentability

Agentability measures how effectively AI agents can operate software interfaces, scored across eight Agent Factors Engineering (AFE) principles. These principles evaluate whether a product's interface enables programmatic agents to read state, understand options, execute actions, and handle errors—capabilities that go beyond traditional API access to encompass the full user interface layer.

The eight AFE principles assess: machine readability of interface elements, transparency about system state and available actions, avoidance of shadow-UI patterns that hide information from agents, sensible defaults that reduce decision complexity, granular control over operations, appropriate chunking of information and tasks, clear status communication, and clean handoff mechanisms between human and agent interaction modes.

The Tier Distribution Problem

The audit classified products into four tiers based on their scores. Only 22 products achieved Agent-Ready status with scores of 45 or above. The majority—54 products—landed in the Developing tier (35-44 points), indicating partial agent compatibility with significant gaps. Another 17 products fell into the Lagging category (20-34 points), while 7 products scored below 20, earning the Agent-Blind classification.

This distribution suggests that while most product teams have implemented some agent-friendly patterns, few have approached interface design with systematic attention to agent operation. The concentration in the Developing tier indicates incremental progress rather than intentional design for multi-modal access.

Critical Weaknesses in Transparency and Shadow-UI

Two principles show particularly severe deficiencies across the dataset. Transparency—the practice of exposing system state, available actions, and operational context in machine-readable formats—averaged just 5 out of 100 across all products.

This near-universal failure means that agents operating these interfaces must rely on visual interpretation and inference rather than explicit declarations of state and capability. When a user interface updates, agents have no reliable mechanism to understand what changed, what actions became available, or what constraints now apply.

Shadow-UI avoidance fared only marginally better, with an average score of 17 out of 100. Shadow-UI refers to interface elements rendered through JavaScript canvas manipulation, custom graphics layers, or other techniques that bypass standard DOM elements and accessibility structures.

For analytics platforms like Google Analytics, shadow-UI patterns often appear in data visualizations, dashboards, and interactive reports—precisely the elements an agent would need to access to retrieve insights or configure analysis parameters.

Relative Strengths in Machine Readability

The principle showing the strongest performance was machine readability, averaging 71 out of 100. This measures whether interface elements use semantic HTML, proper ARIA labels, and structured data that agents can parse. The relatively high score reflects years of accessibility work and web standards adoption, though it also highlights that basic technical readability alone does not ensure agent operability.

Other principles showed middling performance: control averaged 48, status 47, chunking 45, and clean handoffs 40. These mid-range scores suggest partial implementation—products may expose some controls in agent-accessible ways while hiding others, or may communicate status inconsistently across different interface sections.

Defaults scored 35 on average, indicating that most products require agents to make numerous explicit choices rather than providing sensible starting configurations that enable quick operation with minimal decision-making.

What This Means for Analytics Platforms

Analytics tools face particular agentability challenges. These platforms must expose complex data models, provide access to visualization configurations, enable report creation and modification, and support query construction—all activities that benefit from agent assistance but require careful interface design to enable agent operation.

An agent working with an analytics platform needs to understand available data dimensions, metric definitions, filter options, date range configurations, and visualization types. When these elements exist only as visual affordances without structured declarations, agents must resort to fragile heuristics and visual parsing.

For users, agent-ready analytics tools could enable natural language queries that get translated into properly configured reports, automated monitoring that adjusts tracking parameters based on business logic, and cross-platform analysis that pulls data from multiple sources without manual export-import workflows.

Concrete Guidance for Product Teams

Product teams looking to improve agentability should prioritize the weakest principles first. Implementing transparency means adding structured metadata that declares available actions, current system state, and interface capabilities. This might take the form of JSON-LD annotations, custom data attributes, or accessibility enhancements that serve both human assistive technologies and AI agents.

Addressing shadow-UI requires auditing rendering approaches and replacing canvas-based or graphically-rendered controls with semantic HTML equivalents. Where custom rendering is necessary for performance or functionality, teams should provide parallel structured representations that expose the same information and controls.

Teams can assess their current standing by checking the Agentability Index for comparative benchmarks or using assessment tools to run a free audit of their own interfaces.

The path to agent-readiness does not require abandoning existing interfaces or building separate agent-specific APIs. Rather, it demands extending current interfaces with structured semantics, explicit state declarations, and consistent patterns that both humans and agents can navigate. As the audit data shows, most products have started this journey but few have completed it—creating both a competitive opportunity and a usability imperative as AI agents become standard tools in users' workflows.