What it is
Decagon is an AI customer support platform that deploys autonomous AI agents to resolve customer conversations end-to-end — refunds, identity verification, subscription cancellations, order changes — across chat, email, voice, and SMS. It is not a copilot that suggests replies for a human to send; it is an agent engine that handles the conversation itself and escalates only what it can’t close. The closest reference point is Intercom Fin or Sierra, but Decagon is aimed squarely at enterprise volume (Notion, Duolingo, Eventbrite, Rippling, Chime, Affirm, Riot Games are public customers) rather than SMB self-serve.
Why it shows up in Customer Success stacks
- Agent Operating Procedures (AOPs). Decagon’s differentiator: non-technical CS and support ops teams define multi-step workflows in plain language instead of coded decision trees. This is what lets the agent do real actions (process a refund, verify identity) with consistency rather than just answering FAQs.
- It deflects the ticket volume that buries the CSM. Tier-1 “where’s my order / reset my plan” volume gets resolved by the agent, so human CSMs and support reps spend their time on retention-bearing accounts instead of password resets. That deflection is the load-bearing reason it shows up next to Gainsight, Vitally, or Totango rather than replacing them.
- Multi-channel with shared context. Voice (built with ElevenLabs), chat, email, and SMS run off one centralized knowledge and workflow layer with cross-channel memory, so a customer who started in chat doesn’t re-explain on a call.
- QA built in. Watchtower (always-on QA) and Experiments (live A/B testing of agent behavior) let ops teams monitor and tune resolution quality without exporting transcripts to a separate tool.
Pricing
Custom only — no public list price. Reported structure as of 2026: an annual platform fee around $50,000, plus a usage charge that is either per-conversation (every interaction the agent handles) or per-resolution (only when the agent successfully closes the issue, reportedly around $0.50/resolution). Enterprise contracts typically carry conversation minimums; the median reported annual spend is materially higher than the platform fee once volume is counted. You will not get a number without a sales call.
Best for
- Enterprise and upper-mid-market support orgs doing tens of thousands of conversations a month, where tier-1 deflection at scale pays back a six-figure annual contract
- CS/support ops teams that need the agent to take actions in backend systems (refunds, account changes), not just answer knowledge-base questions
- Companies standardizing one agent layer across chat, voice, email, and SMS rather than buying a separate voice bot
Watch-outs
- Pricing opacity and floor. With a ~$50K platform fee before any usage, Decagon is not viable for SMB or low-ticket-volume teams — under a few thousand conversations a month, Intercom Fin’s per-resolution model or a lighter tool will cost far less. The guard: model your actual monthly resolution volume against both Decagon’s per-resolution rate and Fin’s before committing to an annual.
- Implementation is a project, not a switch. AOPs do real work but you have to author them; expect weeks of workflow definition, integration wiring (Salesforce, Intercom, Zendesk, Kustomer), and tuning before deflection rates land where the sales deck promised. Budget CS-ops time, not just license spend.
- It deflects, it doesn’t retain. Decagon resolves support contacts; it does not own health scores, renewals, or expansion. Pairing it with a dedicated CS platform (Gainsight, Vitally) is the norm — treat it as the deflection layer, not the CS system of record.