New at Cresta

AI Feedback now includes examples of specific, actionable feedback to help users describe both undesirable and desirable behavior. Users can create a test case directly from a feedback item through a dedicated flow that automatically generates the test case and a recommended evaluator based on the conversation and feedback. The generated evaluator provides a starting point for evaluation and should be reviewed and validated before use. This new flow replaces the previous create-from-scratch modal.

Test AI Agents against predictable mocked responses instead of live third-party APIs. Register an external API endpoint, review its captured invocations, and promote any invocation into a mock rule that returns a set response. Group rules into a scenario and attach it to a test case so every run exercises the same conditions, including failure paths that are hard to reproduce against a real API. Find it under System Config > APIs & Webhooks > APIs.

AI Agent now includes two new tabs to manage agent configurations across testing, staging, and production; Environments: Create and manage environment-specific variables and secrets that multiple AI agents can share. Deployments: Connect an AI agent to an environment, select revisions, configure traffic splits, and manage runtime settings. You can also duplicate deployments, deploy to another environment, and roll back to previous configurations.

Build, test, and optimize AI agents — including prompts, tools, and orchestration — from a local no-code interface, guided by Cresta best practices and your conversation data.

When you provide feedback to an AI Agent in Closed Conversations, the updated popover now shows examples of good and bad feedback. These examples help you understand what makes feedback effective and guide you on what to write.

When an agent transfers a call to another agent without context (cold transfer), a concise handoff summary is automatically generated to capture key details from the prior customer interaction, helping the receiving agent quickly understand the context and reducing the need for customers to repeat information. After the call ends, this handoff summary is available in the Closed Conversation view in Director.

Automatically translate post-call conversations from 30+ languages into English in bulk. This feature helps teams quickly review, analyze, and extract insights from diverse interactions in different languages without manual translation, making cross-language QA and reporting more efficient.

Topic Discovery V3 has updated LLM-generated clusters to produce more specific, less duplicative, and more mutually exclusive conversation reasons at every level, including drill-down views.

Regex keyword blocks match text patterns instead of exact words. Create a single pattern to identify values such as dollar amounts, confirmation codes, or ticket IDs without listing every possible variation. Open a keyword block, switch the Keyword/Regex toggle to Regex, and enter the include and exclude patterns.

Calibration provides a redesigned experience for evaluating how closely your Opera rules align with human judgment. Calibrate multiple rules in a single pass, label each conversation, and view a reliability score for every rule to identify which rules need further tuning. Access Calibration from the dedicated Calibrations page or directly within the rule builder.

Agent Operation Center (AOC) provides supervisors with a transcript of the conversation between the agent and supervisor after an agent raises their hand. Supervisors can review the complete back-and-forth conversation, including the agent’s question and the supervisor’s responses. The transcript is scrollable, and selecting a message navigates the supervisor to the corresponding hand-raise or assistance event in the AOC conversation.

All Director users will now see a default new look, a unified Search/Ask bar, and the Compass AI Agent, making it easier to navigate faster and work more efficiently.

A/B testing allows you to evaluate different versions of AI Agent/subAgent prompts, function tools, knowledge bases, and other settings using real user traffic. Use it to understand end-user preferences that offline testing alone can't reveal.

Configure scorecard publishing controls for Conversation Templates by enabling Require publish. When enabled, submitted scorecards remain hidden from agents and are excluded from Performance Insights and performance calculations until they are published by a user with an authorized role.

Added a new Scorecard editors setting under Performance Config > Performance templates. This setting lets you specify which users, teams, or groups can edit a scorecard after it has been submitted.

Admins and managers can now configure a new AI Agent feedback added notification under Notification Management > Manager notifications/Admin notifications > Conversations. This notification alerts tagged users and everyone participating in the feedback thread whenever AI Agent feedback is added to a conversation. Selecting the notification opens the AI Feedback page directly to the relevant feedback thread instead of the conversation transcript.

Admins and managers can now review the answers agents would have seen for knowledge hints they didn’t open. In Closed Conversations, look for assistance marked Ignored. Expand the section to view the hint details and the answer that would have been displayed if the agent had opened it.

Generate summaries for multiple instances within a single conversation, capturing distinct cases or subjects discussed during one call. This feature automatically identifies and summarizes each separate case — such as individual patients on a provider call or multiple member inquiries — as its own instance. Each instance includes the same summary topics, such as name, reason for call, and resolution. Example: Member 1 (name, reason for call, resolution), Member 2 (...), and Member 3 (...).

Real-time Trends is now available in Cresta Insights. The Trending Now tab surfaces emerging topics from live customer conversations, helping teams identify fast-moving issues and changes in customer behavior sooner.

Performance Insights > All scorecard now supports Cresta-modeled outcome metadata filters. You can now use Cresta-modeled outcomes alongside existing metadata to filter results and refine your Performance Insights analysis.