Decagon
6.9/10Decagon is an enterprise platform for AI customer service agents that resolve support conversations over chat, email and voice.
Data last updated November 2026 · Our methodology
Overview
Decagon, based in San Francisco, builds AI agents that handle customer support conversations end to end. Chat, email and voice agents all share one intelligence layer. Teams write Agent Operating Procedures (AOPs) in natural language, so agents can follow multi-step workflows such as refunds and cancellations by calling backend systems. The platform also covers testing and optimization: live A/B experiments, simulation-based QA, Watchtower for always-on quality monitoring, AI-generated knowledge suggestions, and voice-of-the-customer reporting. A copilot called Duet helps teams build and improve agents. Customers include Chime, Duolingo, Notion, Rippling, Deutsche Telekom and American Airlines, which makes Decagon a fit for large, high-volume consumer and fintech support teams. It does not publish prices. Contracts are custom and use either a per-conversation fee, which Decagon says most customers pick and which falls with volume, or a higher per-resolution fee that skips escalated conversations. Its documentation and third-party reports describe API-based connections to helpdesks such as Zendesk, Salesforce, Intercom and Kustomer. Its agent-assist tool runs inside Salesforce, Zendesk and Front. Under the hood, the agents combine models from OpenAI, Anthropic and Cohere with Decagon's own fine-tuned models. The main limitation is access. You need a sales process to learn the price, reviewers put implementation at about two months, and reviewers see it as a poor fit for startups and small businesses.
Our Rating
Who Decagon is best for
Good fit
Decagon's AI agents handle support conversations end to end over chat, email and voice, and they connect to helpdesks such as Zendesk and Intercom.
The customer list is mostly large enterprises, and contracts are custom and sold through sales.
Named fintech customers include Chime, which reports 70% resolution on chat and voice.
Named customers include software companies such as Notion, Rippling and Duolingo.
Less suited
High-volume mid-sized support teams can use it, but the sales-led process and onboarding are heavy.
Support operations staff write multi-step workflows such as refunds as natural-language procedures.
Reviewers describe it as a tough fit for startups because there is no self-serve plan or public pricing.
Custom pricing and a roughly two-month implementation make it a poor fit for small teams.
Contracts are custom and enterprise-focused, and setup takes about two months.
Pros & Cons
Pros
- Chat, email and voice agents run on one shared intelligence layer, so behaviour stays consistent across channels.
- Agent workflows are written as natural-language AOPs, so support operations staff can define multi-step actions such as refunds.
- Built-in A/B experiments, simulation testing and Watchtower QA support ongoing tuning at scale.
- Large reference customers report results, for example 70% chat and voice resolution at Chime and an 80% deflection rate at Duolingo.
- Reviewers on G2 rate it highly, with an average around 4.9 out of 5, and praise how quickly support responds.
Cons
- There is no public pricing, so every quote needs a sales call.
- Under per-conversation billing you pay even when the AI fails to resolve an issue.
- Reviewers report an average implementation time of about two months.
- Third parties describe helpdesk integrations as API-based rather than marketplace apps, and some reviewers mention missing features.
- It is not suited to startups or small businesses.
Key Features
Decagon Pricing
Per-conversation (custom)
- Rates fall as volume rises
- Fixed fee for each incoming conversation
- Decagon says most customers choose this model
Per-resolution (custom)
- Escalated conversations are not charged
- Larger resolution commitments lower the rate
- Higher fixed fee charged only for conversations the AI fully resolves
Frequently Asked Questions
Decagon does not publish prices. Contracts are custom and use either a per-conversation fee or a higher per-resolution fee, and rates fall with volume or commitment.
No self-serve plan or public trial was found. Access starts with a demo request and a sales conversation.
Decagon's documentation and third-party reviews cite connections to Zendesk and Salesforce, with Intercom and Kustomer also mentioned. Its agent-assist tool works inside Salesforce, Zendesk and Front.
Decagon runs a Trust Center. Third-party reviews report that it lists SOC 2 Type II, ISO 27001, PCI DSS, HIPAA, GDPR and CCPA. Confirm HIPAA coverage directly with Decagon, because it appeared only in third-party sources.
According to Sacra, Decagon combines models from OpenAI, Anthropic and Cohere with its own fine-tuned models.
Review aggregators report an average of about two months.
Similar Tools
See all alternatives
Amazon's public cloud platform, with over 200 pay-as-you-go services for compute, storage, databases, networking and AI, used by startups through large enterprises and governments.
Anthropic's organization-wide Claude plan adds chat, Claude Code, Cowork agents, connectors, and enterprise governance such as SCIM, audit logs, and a compliance API.
Fin is Intercom's AI customer service agent. It resolves support conversations across chat, email, voice and social, either inside Intercom or on top of an existing helpdesk.