Prompt Optimizer Now Supports Conversations AI
Prompt Optimizer now works with Conversations AI, giving teams a way to test and tune chat agents before applying prompt changes to production.
This update helps teams using SMBcrm AI features and customer communication workflows validate agent behavior with realistic conversations, AI scoring, and prompt comparison tools.
Build a Test Plan
Prompt Optimizer can generate contextual customer scenarios based on the agent prompt, language, knowledge base, appointment setup, and actions. Each scenario can include a customer persona, opening message, expected behaviors, and priority level.
You can also reuse scenarios from earlier runs instead of starting from scratch. Scenarios, evaluations, and optimized prompts stay aligned to the agent language, and a usage view shows daily free messages used versus remaining before a run begins.
Test with Real Chats
Prompt Optimizer sends customer-style messages and waits for the agent’s real replies, rather than relying only on simulated responses.
During testing, it can track actions such as appointment booking, human handover, workflow triggers, bot transfer, stop bot, auto follow-up, contact field updates, and knowledge base queries. Each chat includes a transcript with AI scoring that explains what passed, what failed, and why.
Runs continue in the background, so you can leave the screen and return to completed results.
Improve Prompts Before Applying Changes
One-click Improvise analyzes failed chats, identifies likely root causes, creates an improved prompt, and tests it. Auto Optimize can run multiple optimization attempts until the target accuracy is reached.
A prompt diff viewer shows exactly what changed before anything is applied. The highest-accuracy prompt is marked as the best variation, with a history of every attempt.
Safe Testing Notes
Testing runs against a cloned agent, so the production agent is not modified until you choose to use the improved prompt. Temporary test contacts are created for each run and cleaned up automatically.
Keep these details in mind:
- AI evaluations can vary between runs, so accuracy scores should be treated as directional guidance.
- Real actions, including bookings, workflows, and handovers, execute during testing. Use test calendars and workflows where appropriate.
- Long or stuck conversations time out automatically so runs can complete.
In Case You Missed It
For more on managing Conversations AI channels, read Conversation AI Channel Management.
Need Help Applying This Update?
Want help turning this update into a working part of your CRM, marketing, or automation setup?
Get Started or schedule a demo and we’ll help you map the right next step.