The "DoubleVerify" for AI
Why Customer-Facing AI Agents Need a Verification Layer
Every company that has switched on an AI agent to handle customer conversations faces the same structural problem: the platform running the agent is also the one reporting on how well it performed. Isara exists to close that gap. It is an independent AI monitoring and analytics platform that analyses customer conversations separately from the tools that generate them. The precedent for this kind of independent oversight already exists. It built a nine-figure industry in digital advertising. And it is now unfolding in AI.
The Advertising Industry Already Solved This Problem
When digital advertising scaled, brands discovered a conflict of interest at the core of the market. The same platforms selling ad space were measuring whether those ads had worked. Advertisers had no independent way to verify that their budgets were reaching real people, appearing in safe environments, or delivering any return at all.
That structural blind spot created a new category of software: ad verification.
DoubleVerify and Integral Ad Science became its dominant players by doing one thing. They sat between the advertiser and the platform as a neutral, independent layer. The data they provided was data the platforms had no incentive to surface accurately.
The numbers that resulted are significant:
The global ad verification market reached $4.82 billion in 2024, driven by rising concern over ad fraud and the need for independent measurement. Growthmarketreports
DoubleVerify alone reported revenue of $656.8 million for 2024, a 15% year-over-year increase. PESTEL ANALYSIS
By 2025, DoubleVerify measured approximately 9.5 trillion media transactions and reported net revenue retention of 109%, serving over 2,500 customers. Stocktitan
The core insight was simple. Advertisers needed a source of truth that the platform had no stake in distorting. Once that need was clearly named, a category was born.
Companies that have switched on AI agents to handle customer conversations are now in the same position. They are relying entirely on the platform running the agent to tell them how well it performed. Resolution rates, deflection rates, CSAT scores: all of it flows from a vendor with a direct commercial interest in how those numbers look.
This is exactly the problem Isara solves. By pulling conversation data independently and analysing it through its own ML and large language model stack, Isara gives leaders signal the platform has no incentive to provide. That includes Escalation and Early Warning Signals, Compliance Audit results, Customer Frustration Watch data, and Churn Signals that surface in conversations before they ever appear in a CRM.
The risk of operating without that independent layer is not hypothetical. In 2024, Air Canada was found legally responsible for a commitment its AI agent made to a customer. The tribunal ruled that the airline could not disclaim responsibility for its own AI agent's statements. Across four leading customer-facing AI agents, unsafe or inaccurate response rates ranged from 5% to 13% in 2025 testing. In an operation handling thousands of conversations per day, that is a constant, measurable liability event. SweptAicerts News
The Three Signals Platforms Will Not Show You
The most important insight from the DoubleVerify model is not just that independent verification is possible. It is that the most valuable signals are precisely the ones the platform has the least reason to highlight.
For customer-facing AI agents, those signals fall into three categories.
Compliance exposure
An AI agent can commit the company to a policy it does not offer, confirm data handling it cannot guarantee, or make a product claim that is factually wrong. Generative systems can sound confident while being wrong, and that combination is uniquely dangerous when agents are talking directly to customers. If the system treats certain customers differently, the organisation is responsible for that outcome. CX Today
Isara's Compliance Audits feature scans every conversation for exactly this kind of breach. It does not require the platform to flag it. The audit runs independently, against the full conversation record.
Frustration and churn signals
A customer can receive a technically resolved ticket while leaving the interaction significantly more frustrated than when they arrived. Standard platform metrics will count that as a success. CSAT scores can lag by days. NPS captures nothing at the conversation level.
Isara's Customer Frustration Watch tracks how frustration evolves across a conversation and across time. Its Churn Signals feature surfaces language patterns that predict disengagement before the customer has acted on them. Its Customer Monitoring and Temperature feature tags conversations with Areas of Concern and lets leaders jump directly to the interactions that require attention.
Knowledge and performance gaps
When an AI agent gives an incomplete or inaccurate answer, it is usually because the knowledge base behind it has a gap. That gap generates repeat contacts, escalations, and dissatisfaction. No platform has an incentive to surface the scale of this problem.
Isara's Knowledge Gap and Documentation Fixes feature identifies exactly where those gaps are, integrating with existing documentation to flag what is missing or unclear. Its Agent Training Recommendations feature surfaces patterns in how human agents handle escalations from the AI, creating a feedback loop that improves both.
Isara is also building toward Revenue Expansion Signals and Agent and CSM Performance tracking, extending the same independent verification logic beyond support into the full customer success function.
What Leaders Should Be Asking Right Now
Why does independent monitoring matter if the platform already provides analytics?
Platform analytics measure the performance of the platform. They are designed to show resolution rates, deflection rates, and response times. Isara analyses the underlying conversations to surface what those metrics do not capture: whether resolutions were accurate, whether compliance obligations were met, and whether customers are showing early signs of churn. Isara's Compliance Audits, Churn Signals, and Customer Frustration Watch all operate on conversation content, not platform outcomes.
Which industries face the most immediate risk?
Financial services, insurance, telecoms, and consumer finance face the sharpest regulatory exposure because AI agent statements in those contexts can create enforceable commitments or breach sector-specific rules. Isara's Compliance Audits feature is designed with these regulated environments in mind. That said, any company running AI agents at scale carries exposure. The Air Canada ruling applies to any industry where an agent makes a factual commitment to a customer.
How does Isara integrate with existing tools?
Isara integrates with the platforms companies are already using to access conversation data. It is additive, not a replacement. The platform continues to run the agent and provide its own reporting. Isara analyses the same conversation data independently and surfaces signals the platform does not. The relationship mirrors how DoubleVerify sits alongside a demand-side platform: the platform runs the campaign, the verification layer provides the honest read on what actually happened.
What does Isara surface that a manual QA review would miss?
Manual QA can sample a small percentage of conversations. Isara analyses every one. Its Proactive Service Analytics feature measures how effectively the team is anticipating customer needs across the full conversation volume. Its Escalation and Early Warning Signals feature catches heated exchanges before they become complaints. At scale, the difference between sampling and full-coverage analysis is the difference between knowing about problems after they become visible and catching them before they do.
What is coming next from Isara?
Isara is building Revenue Expansion Signals to identify accounts showing upsell or cross-sell potential within their interactions, and Quarterly Business Review preparation tools to give customer success managers a complete picture before key account conversations. Both extend the core verification model from reactive monitoring into proactive account intelligence.