Pendo unveils Agent Analytics to optimise AI agents
Pendo has launched Agent Analytics for general use, with early customer Pushpay citing the product's role in identifying where users abandon an AI agent during search tasks.
The software focuses on how people interact with AI agents once they are in day-to-day use. Pendo said the product measures performance, usage, and business impact across an organisation's agent deployments.
Early usage
Pushpay used Agent Analytics during a beta phase for a conversational search agent. The agent lets church leaders query membership and donation data using natural language, according to the companies.
Pushpay said it saw a recurring pattern in early user behaviour. Users would enter a small number of prompts and then stop.
"We started noticing users were prompting only three or four times, then quitting. Agent Analytics made that pattern obvious, so we could pinpoint and target exactly where users were getting stuck. Now we can get them past that drop-off point and into real value," said Paul Frank, Staff Product Manager, Pushpay.
Pushpay said the findings fed into product decisions. The company referenced changes across its roadmap, governance model, and wider AI approach.
Product focus
Pendo positioned the release against a backdrop of faster agent roll-outs inside large organisations. It said many teams still rely on simple user ratings as a primary signal of whether an agent performs as expected.
"Companies are launching agents faster than ever, but they're relying on a simple thumbs-up or thumbs-down to gauge their success which is insufficient," said Todd Olson, Chief Executive Officer and Co-Founder, Pendo.
"Agent Analytics fills that gap, giving teams a reliable way to monitor usage, detect friction, and improve the user experience, ensuring their investment in AI delivers meaningful business value," said Olson.
Pendo said the tool examines what happens after deployment rather than focusing on pre-launch testing. The company drew a distinction between evaluation tools used during development and monitoring in real workflows.
Workflow tracking
Pendo said Agent Analytics tracks hybrid workflows that include both AI agents and traditional software. It said the feature maps how users move between tools and how agents fit into existing processes.
The company also said the product surfaces patterns around agent interactions, including activity before, during, and after a conversation. It said teams can view these behaviours through visual replays.
Pendo said it analyses conversations for prompt themes and emerging use cases. It also said teams can track the impact of subsequent updates on outcomes.
Quality signals
The release includes features aimed at detecting frustration and failure modes. Pendo said the product identifies "rage prompts" and sends alerts when issues affect trust or performance.
Pendo also said the system flags "off-script" behaviour, hallucinations, and cases where an agent cannot respond. It said teams can use those signals to address issues.
The company said the product can embed guidance and surveys inside an agent interface. It also said it maps agent activity to task completion and overall platform engagement.
Customer learning
Pushpay described how usage data shaped its understanding of customer needs. It said it used the system to determine what customers were attempting to find out through the agent and how that should affect development priorities.
"Every day we're learning something new with AI. Agent Analytics helps us understand what our customers want to know-which has informed how we prioritise prompt inferences, fine-tuning, filter enhancements, and experience redesigns based on real usage, not assumptions," said Frank.
Pendo said it has made Agent Analytics free to get started for its customers. The company said the product is now generally available following an initial launch earlier in the year.