Decagon

Decagon is an enterprise AI customer service platform that deploys autonomous AI agents across chat, email and voice, which the vendor calls an AI concierge for every customer. Its agents handle multi-step support conversations by following Agent Operating Procedures (AOPs), natural language workflows that support teams write and refine without engineering sprints. Named customers on the vendor's case study pages include Hertz, Notion, Rippling, Duolingo, Faire, ClassPass, Noom, Substack, Curology, Chime, Mercado Libre and Perplexity. Decagon was founded by Jesse Zhang and Ashwin Sreenivas and emerged from stealth in June 2024. In January 2026 it raised $250 million led by Coatue and Index Ventures at a $4.5 billion valuation, and said more than 100 new enterprise customers, including Avis Budget Group, Block and Deutsche Telekom, had joined in the prior fiscal year.

Around the AOPs, the platform includes a supervisor model that checks responses for hallucinations before they are sent, Watchtower, an always-on QA system that reviews every conversation against custom criteria, Experiments for live A/B testing, simulation-based testing and Duet, an AI partner that helps teams draft AOPs and analyze conversations. Proactive Agents, a spring 2026 release, added user memory across sessions, outbound voice calls and an Agent Workbench for debugging workflows, and Hertz uses the outbound agents to resolve issues before customers contact support. On 1 October 2026 Decagon released Voice 3, which runs on Chord, its first in-house voice model, in 70+ languages, alongside Agent Modules for journeys beyond support such as lead qualification, onboarding and collections, and a Personal Agent Gateway that detects and serves customers' own AI agents through a dedicated channel.

Integration covers CRM, helpdesk and ticketing systems such as Salesforce, Intercom and Zendesk, knowledge base syncs from Confluence, Contentful and Kustomer, and CPaaS voice platforms such as Amazon Connect and RingCentral, plus APIs, custom tool integrations and SIP trunking. Its industry pages also name contact center platforms including Five9, Genesys and NICE, payment and commerce systems including Stripe, PayPal, Adyen, Shopify and BigCommerce, and sector systems such as Sabre, Amadeus, Guidewire, Fiserv and Plaid. Decagon is an MCP client: its April 2026 engineering post describes introspecting a customer's MCP server and curating its tools for the agent. The Personal Agent Gateway uses PACT, a Decagon protocol built on Agent2Agent and OAuth 2.0, rather than MCP, so Decagon does not publish an MCP server for outside assistants. Decagon enforces zero-data retention with its LLM providers, including OpenAI and Anthropic, so no conversation data is stored or used for training by those providers.

Pricing is quote-only and there is no pricing page. Decagon's own pricing post describes two models, a fixed rate per conversation or a higher fixed rate per fully resolved conversation with no charge for escalations, and says most customers choose per-conversation pricing. No rates are published, so every estimate runs through a sales conversation. G2 reviewers report a Time to Implement of 2 months.

Decagon is built for enterprise support teams with high conversation volume. Teams that want published rates or a self-serve start may compare Intercom Fin, which charges per outcome from $0.99 per outcome, or Tidio, which has a free plan and paid plans from $24.17/mo billed annually. Teams already running Zendesk can evaluate Zendesk AI, with Suite plans from $19/mo billed annually, before a separate platform.

Documented outcomes come from named customer stories. Chime reports 70% chat and voice resolution and a 60% decrease in support costs. Duolingo reports an 80% deflection rate for the Duolingo English Test. Flashfood reports a 90%+ deflection rate. ClassPass reports a 95% cost reduction and 10x higher deflection at launch than it had anticipated. Hunter Douglas Group reports $1M in revenue from fully AI-handled conversations.

As of Q4 2026, the Decagon Trust Center lists SOC 2 Type II, GDPR, CCPA, HIPAA, ISO 27001:2022, PCI DSS 4.0.1 and the EU AI Act, with SOC 2 Type II reports from April and October 2025 and an ISO 27001 certificate among its resources. Subprocessors named there include Google Cloud Platform, OpenAI, Amazon Web Services and Azure. The vendor does not state what data its Chord voice model was post-trained on beyond describing it as real-world customer conversations.