Our take

Goes beyond FAQs — these agents actually execute workflows and track resolution analytics, loved by modern support teams.

What it is

Decagon builds AI agents for customer support that do more than answer questions: they take actions (refunds, status updates, account changes), learn from resolved tickets, and report on metrics like deflection and CSAT. It targets modern B2B support teams and is one of the fast-rising players in AI support.

Best for

Resolve support tickets end-to-end, with actions
Deflect repetitive tickets at scale
Track deflection and resolution analytics

Pros

Agents take real actions, not just reply
Strong analytics on support outcomes
Fast to deploy on existing helpdesk data

Cons

Enterprise pricing, not for small teams
Requires clean support data to shine

Quick start

Demo + scoping call (no self-serve free tier). Decagon reads your top-20 ticket categories and 90 days of transcripts before the call.
Pilot setup: 30-day "Build" phase where Decagon's ML team trains your first agent on your knowledge base + historical tickets.
Deploy to one channel first (Zendesk, Intercom, or Salesforce Service Cloud) → measure deflection, CSAT, resolution accuracy.
Production mode: Decagon ships "AI Actions" for refunds, account changes, status updates — agents go beyond answering.
Analytics: live dashboard tracks deflection (target 50%+), CSAT (target ≥ 4.4), resolution rate (target ≥ 70%), and surfaces gap topics for retraining.
Pricing tiers: Enterprise (custom) — typically $0.50–$2.00 per resolved ticket.

Sample input / output

Input

Decagon agent config (sample):

Topic: "Subscription Pause / Cancel"
  Inputs allowed:
    - email of subscriber
    - reason (select: too expensive, not using, competitor, other)
  Actions:
    - pause_subscription (refunds pro-rated remaining)
    - offer_save_offer (3 months at 50% off) — if reason = "too expensive"
    - escalate_retention (if reason = "competitor" → tag #competitive-risk)
  Tone: empathetic, short
  Never:
    - quote competitor pricing
    - say "I can't do that" — always escalate to retention human
Channel: Zendesk + Intercom (production)

Output

30-day post-pilot metrics:
  Total tickets routed: 84,200
  Resolved end-to-end (no human): 59,150 (70.3%)
  Deflection (no human touched): 55.8%
  CSAT: 4.47 / 5
  Avg resolution time: 1m 12s (was 3h 48s human baseline)
  Top improvement areas flagged:
    - "what happens to my data after pause?" (12% of escalations) → retrained FAQ
  Outcome: 14% of subscribers who paused via Decagon returned within 60 days vs 8% historical.

Benchmarks

Founded 2023; $65M Series B in 2025 led by Bain Capital Ventures — TechCrunch 2025

Customers Public refs include Notion, Bilt, ClassPass, Eventbrite — decagon.ai customers

Action library 80+ pre-built actions (refunds, account changes, status, retention) — Decagon platform

Pricing ~$0.50–$2.00 per resolved ticket — Decagon sales briefs

Avg deflection 50–70% in published case studies — Decagon case studies

Pricing

Enterprise pricing

Underlying models

Multiple frontier models (model-agnostic)

Should you pick this?

Pick it if You run support at scale (10k+ tickets/month), want agents that take real actions (refunds, account changes), and need first-class analytics on deflection + resolution rate.

Skip it if You're a smaller team without a clean Zendesk/Intercom knowledge base to train on — the build phase needs good source data. For lighter-weight no-code, Lindy or n8n are closer fits.

Decagon