Decagon Product Foundations

Manually reviewing thousands of conversations is time-consuming and unsustainable. Watchtower automates this process, helping you monitor and assess conversations at scale by flagging only what matters.

With Watchtower, you define flag criteria,  the conditions you want to track, and the system highlights conversations that match. Flags can be set for abandoned chats, upsell or cross-sell opportunities, regulated complaints, or negative sentiment that may hurt CSAT. You can then group flagged conversations into categories to simplify investigation and trend tracking.

Start by creating a Watchtower job. You define detection rules such as keywords, sentiment signals, or combinations of conditions. Watchtower can run once on historical data or continuously on new conversations. Filters let you focus on specific channels, products, or CSAT ranges, making detection faster and more relevant.

To further refine results, you can add rubrics that evaluate or score flagged conversations, grouping them into categories for deeper analysis over time. Two features support nearly every setup: View Flagged Conversations, which shows why each conversation was flagged and what action to take; and Test Watchtower, which runs your configuration on a small sample of conversations so you can validate accuracy before scaling.

Best practices include writing clear flag rules with positive and negative examples, testing early with small samples, applying relevant filters, and reviewing flagged cases regularly. Audit logs can also help pinpoint operational issues, such as when a refund tool was called but no refund was issued.

In summary, Watchtower makes QA scalable, precise, and proactive, ensuring your AI system and human agents stay aligned with both customer needs and business standards.