Decagon is an enterprise conversational AI platform that hands customer support over to AI agents. Founded in the United States in 2023, it brings the main support channels — chat, voice, and email — onto a single foundation. Its defining feature is a mechanism called Agent Operating Procedures (AOPs), which lets you write handling steps such as “when a return request arrives, confirm the order number and, if the conditions are met, proceed to the refund” in ordinary prose rather than a dedicated configuration language. In other words, the people who actually run support can define how the AI behaves in their own words. More than 100 companies use it, including Deutsche Telekom, American Airlines, Duolingo, Notion, Ticketmaster, and Rippling, and in March 2026 the company announced funding at a $4.5 billion valuation.
Key Features
- Procedures defined in natural language with AOPs: Describe the handling flow in everyday language and the AI agent’s behavior follows from it. Support leads can add and revise procedures without going through engineering
- Omnichannel coverage across chat, voice, and email: Automated web chat, voice handling for phone calls, and automatic replies to email inquiries all run on the same knowledge and the same procedures. There is no need to build a separate bot per channel
- Continuous quality monitoring with Watchtower: Real conversations are monitored on an ongoing basis to surface cases the agent handled poorly and gaps in the procedures. Quality is maintained without relying on manual review of every ticket
- Simulation and A/B testing: Run large volumes of expected scenarios before going live to verify behavior, and compare multiple versions of a procedure. The impact of a configuration change can be measured in advance
- Voice of the Customer analytics: Accumulated conversation logs make inquiry trends and sources of dissatisfaction visible. Beyond automating support, the data becomes material for product improvement
- Browser Actions and Assist: Features added in 2026. Browser Actions lets the AI operate web-based internal systems directly to complete a task. Assist supports the case where the AI helps a human agent handle a conversation
Pricing
| Plan | Price | Highlights |
|---|---|---|
| Enterprise | Contact for quote | All features. Individual quote based on deployment size, channels, and conversation volume. Starts with a demo request |
Decagon publishes no price list; pricing is quoted individually. The official site describes billing concepts such as conversation-based and resolution-based models, but no specific unit prices or annual costs are disclosed. Third-party media publish estimates, but these are not figures that can be confirmed officially, so the actual cost must be checked through a quote. Contracts assume a mid-size to large enterprise scale, and there is no self-service free plan that individuals or small teams can start on immediately. Deployment proceeds with dedicated support staff, and a build period of several weeks before going live is generally expected.
Pricing is as of August 2026. Please check the official site for the latest pricing.
Pros & Cons
✅ Pros
- Because procedures are written in natural language, the support organization can run operations on its own without depending on engineers
- Chat, voice, and email are handled on one foundation, so duplicated management per channel is less likely
- Designed for high inquiry volumes, with deployments in high-ticket-count industries such as airlines, telecom, and finance
- Continuous monitoring and simulation come as standard, allowing the quality of automated responses to be verified over time
- Conversation analytics make it easier to feed results back beyond support, into product and operations improvement
⚠️ Cons
- Pricing is undisclosed, making it hard to gauge cost early in the evaluation process
- Because enterprise contracts are the premise, it is not a realistic option for individuals or small teams
- There is no free plan or self-signup, so you cannot simply try it first
- Deployment requires integration design with existing CRM, help desk, and internal systems, taking time and effort before go-live
- Japanese-language official documentation and domestic case studies are limited compared with English-language resources
Comparison with Similar Services
| Criteria | Decagon | Intercom Fin | Sierra | Zendesk AI agents |
|---|---|---|---|---|
| Main audience | Support teams at large enterprises | SMB to enterprise | Customer experience at large enterprises | Companies already on Zendesk |
| Channels | Chat, voice, email | Mainly chat and email | Chat, voice | Channels within Zendesk |
| How procedures are defined | Natural language (AOPs) | Based on knowledge articles | Natural language plus developer configuration | Configuration in the admin console |
| Pricing disclosure | Undisclosed (quote required) | Published (per-resolution billing) | Undisclosed (quote required) | Published (plan-based) |
| Ease of adoption | Assumes a guided build | Can start relatively quickly | Assumes a guided build | Easy within an existing Zendesk setup |
Who Is It For
- Support leaders at companies where inquiry volume makes staffing costs and wait times a business-level problem
- Customer success teams running chat, phone, and email on separate tools who want them unified
- Support teams where depending on engineers to configure automated responses has become a bottleneck
- Businesses that want the AI to complete actual transactions such as refunds and booking changes, not just answer questions
- Operations teams that want to manage automation rate and quality by the numbers and improve continuously
Summary
Decagon treats support automation not as “installing a chatbot” but as transferring business procedures. Natural-language procedure definition through AOPs, integration across channels, and the combination of continuous monitoring and simulation let the frontline team grow the AI’s behavior themselves. On the other hand, pricing is undisclosed and enterprise contracts are the premise, so this is not a service to try casually. If your organization is large enough that inquiry volume is high and the return on automation can be explained in numbers, a good starting point is to request a demo and check how far your own handling flows can be expressed as AOPs.