CoSupport AI

Feature tested: AI Persona Setup Method

Result: Passed (8.5/10)

Verdict: Strong — two setup options covering both beginners and advanced users

Expected behavior: Choose between a guided System Prompt Builder or a manual prompt editor. The builder is ideal for most users while the manual option gives full control for experienced users.

Why it matters / Conclusion: System Prompt Builder is better for most users — during testing, the guided setup produced a fully configured StyleNova Sales agent in minutes. Manual prompt delivered equally accurate results for the NexaCore HR Assistant, but required writing detailed role boundaries and contact instructions from scratch. Note that once you select a setup method during agent creation, it cannot be changed later — so choose carefully based on your technical comfort level. See the manual prompt input and HR agent output above to compare both approaches side by side.

Feature tested: AI role

Result: Passed (9.5/10)

Verdict: Excellent — three role options covering support, sales, and custom use cases

Expected behavior: Assign a primary role to your AI agent. The role determines how the AI approaches conversations and what its main objective is.

Why it matters / Conclusion: The Sales role visibly changed how StyleNova's agent engaged with customers — it proactively offered discounts and upsold the membership plan without being prompted. Role selection is one of the most impactful decisions in the setup — get it wrong and the entire agent behavior shifts. See the Sales role output above.

Feature tested: Style & Response Settings

Result: Passed (8.5/10)

Verdict: Strong — good control over channel, style, and response length

Expected behavior: This step controls how your AI agent communicates. Three settings are configured on the same page. Channel: Chat, Email. Style: Casual (with emojis), Professional, Custom. Response Length: Concise, Standard, Comprehensive.

Why it matters / Conclusion: In testing, the Casual + Standard combination made StyleNova's agent feel genuinely conversational. Check the input and output cards above.

Feature tested: Escalation Workflow

Result: Passed (8/10)

Verdict: Good — functional but default escalation message could be more polished

Expected behavior: Defines how the AI handles queries it cannot answer. Four escalation options are available: Escalate immediately — AI logs a ticket in the helpdesk as soon as escalation is required or offers to connect with a human agent Gather details and escalate — AI collects contact or other details from the user first, then logs a ticket Forward to email — sends the full conversation transcript or summary to a specified email address End conversation — closes the conversation with a custom message

Why it matters / Conclusion: When asked a query outside the knowledge base — the agent immediately responded with a escalation message rather than attempting a misleading answer. The escalation triggered correctly every time during testing. However, the default escalation message feels generic for a customer-facing brand. See the exact escalation response in the output card above.

Feature tested: Custom Workflows

Result: Passed (8.5/10)

Verdict: Strong — effective for handling sensitive scenarios with predefined

Expected behavior: Create specific rules that tell the AI what to do in defined situations. Each workflow has a trigger condition and a predefined action — ideal for complaints, refunds, or urgent escalations.

Why it matters / Conclusion: When a customer said He received a damaged product, the Complaint Escalation workflow triggered immediately — bypassing the knowledge base entirely. Check above to see customer query & AI agent's response to it.