AI for sales and supportDecagon at a glance
Pros
- One agent across chat, email and voice with shared logic
- Agent Operating Procedures let support teams write behaviour in plain language
- Can take actions such as refunds and account changes through connected systems
- Testing, observability and analytics tools for improving the agent over time
- Strong enterprise customer list and well-funded vendor
Cons
- No published pricing, free plan or self-serve trial
- Integrations with back-end systems usually need engineers
- Rollouts commonly take weeks to months
- Usage-based contracts make total cost hard to predict
- Few public user reviews to judge day-to-day experience
What Decagon is
Decagon is an AI customer support platform that deploys AI agents to handle customer conversations across chat, email and voice. Founded in 2023 by Jesse Zhang and Ashwin Sreenivas and based in San Francisco, it now describes its product as an AI concierge for every customer.
The aim is to resolve issues end to end rather than send customers to help articles. With access to systems such as a CRM, billing or order database, a Decagon agent can process a refund, change a subscription or update an account, and hand the conversation to a human when it should. In January 2026 Decagon raised a $250 million Series D at a $4.5 billion valuation, led by Coatue and Index Ventures, bringing total funding to roughly $481 million.
Who it is for
Decagon is built for enterprises and fast-growing consumer brands with large support volumes. Named customers in press coverage include Chime, Hertz, Avis Budget Group, Block, Affirm, Deutsche Telekom and Grubhub, and other sources cite Duolingo, Notion and Rippling. These are companies with thousands of conversations a day and engineering teams to support an AI rollout.
It is not a fit for small businesses that want to switch on an AI chatbot this afternoon. There is no self-serve sign-up, setup involves integration work, and contract sizes are aimed at enterprise budgets.
Key features
Agent Operating Procedures, or AOPs, are Decagon's core idea. Instead of building flows in a visual editor or code, support teams describe how the agent should handle a situation in natural language, and Decagon turns that into agent logic. AOPs can trigger back-end actions such as refunds or account changes, which is what separates a resolving agent from a chatbot that only answers questions.
Decagon unifies chat, voice and email in a single intelligence layer, so the same procedures and memory apply across channels. Decagon Voice handles phone calls with natural dialogue and brand customisation, and Chime's chief operating officer is quoted on Decagon's site praising cross-channel memory.
For improving the agent, Decagon offers testing, observability and experimentation tools, so teams can validate changes before they reach customers. Independent coverage mentions a trace view that shows how a procedure ran in a given conversation, and newer tools for debugging and for drafting procedures with AI help.
An analytics suite turns conversations into insight about why customers get in touch, which support leaders can use to fix root causes. Human handoff is built in, so the agent can pass complex or sensitive cases to staff with context.
Decagon's homepage highlights customer outcomes such as 70 percent combined chat and voice resolution and an 80 percent deflection rate. These are vendor case-study figures from large customers, so treat them as best cases rather than typical results.
How to evaluate Decagon
Start with your contact reasons. List the ten most common reasons customers get in touch, note which back-end system each one needs, and ask Decagon to show how an AOP would handle them end to end. The value of a resolving agent depends on how many of those reasons it can complete without a human.
Agree success measures before a pilot: resolution rate, escalation rate, customer satisfaction and cost per resolved conversation compared with your current cost per ticket. Ask how resolution is defined in the contract, how spam and misrouted chats are billed, and what happens to pricing if volume spikes. Finally, ask for reference customers in your industry and at your volume, since the headline results on Decagon's site come from very large deployments.
Pricing explained
Decagon only sells through custom contracts. You book a demo, scope the use cases and agree a price based on expected volume. There is no free plan or trial.
The two usage models are per conversation and per resolution. Per conversation is simpler to forecast from ticket volume and is the option Decagon reportedly says most customers prefer. Per resolution charges more for each conversation but only when the AI closes it without a human, which ties cost to outcomes.
Because there are no public figures, compare Decagon on total cost of ownership: the contract, integration work and the human time needed to write and maintain procedures. Request a pilot or a phased rollout with clear success measures before committing to a multi-year deal.
Ease of use and setup
Writing AOPs in plain language makes day-to-day changes accessible to support operations staff, not just engineers, which Decagon sells as faster iteration without engineering sprints or vendor tickets.
Initial setup is a bigger project. Independent reviews describe rollouts of roughly four to twelve weeks, with engineers needed to connect back-end systems so the agent can take actions. Budget for a proper implementation rather than a quick switch-on.
Integrations
Decagon connects to help desks such as Zendesk and Salesforce, knowledge sources such as Confluence and telephony providers including Amazon Connect, RingCentral and SIP trunking, according to independent coverage. Capterra's listing also mentions Slack and Claude Code connections.
The value comes from connecting order, billing and account systems so the agent can act. Those connections are usually custom work, so map the systems your top ten contact reasons depend on during the sales process.
Support and reliability
As an enterprise vendor, Decagon works closely with customers during deployment, and its Capterra listing names email, phone and knowledge base support plus training through webinars, documentation and videos.
Public reliability data is limited. Independent reviewers report concerns about visibility into why the agent made a decision and about slower responses or more escalations during sudden traffic spikes. Ask for references from customers at your volume and test peak-load behaviour in a pilot.
What users like and dislike
Decagon has very few public user reviews; its Capterra listing showed no ratings in November 2026, so most evidence comes from case studies and press coverage. Customers and analysts credit it with high resolution rates, strong voice capability and fast iteration through plain-language procedures.
The main criticisms in independent write-ups are opaque pricing, dependence on engineering for integrations, limited insight into agent decisions in some cases and performance under sudden peaks. Several of the most detailed reviews come from competing vendors, so weigh them with care.
Our verdict
Decagon is one of the strongest options for large companies that want AI agents to resolve real customer issues across chat, email and voice. Its plain-language procedures and omnichannel design are well suited to big support operations.
The trade-offs are cost, opacity and setup effort. Mid-market teams will usually get faster, cheaper results from AI features in their existing help desk. Enterprises should run a scoped pilot with clear resolution and cost targets before signing a long contract.
Decagon scores 7.4 out of 10 overall. Its strongest criterion is features (9/10) and its weakest is value for money (5.5/10).
Decagon pricing
| Plan | Price | What you get |
|---|---|---|
| Enterprise platform | Custom quote | Annual contract negotiated with sales; no public list price |
| Per-conversation model | Custom quote | Charged for each conversation the AI handles; the model Decagon says most customers choose |
| Per-resolution model | Custom quote | Charged only for conversations the AI resolves, at a higher rate per conversation |
Decagon does not publish prices and has no pricing page, free plan or self-serve trial. Every deal starts with a demo and ends in a negotiated contract, usually annual or multi-year.
Pricing is usage-based rather than per seat. Buyers choose between paying for each conversation the AI handles or paying a higher rate only for conversations it resolves. Per-resolution sounds safer, but it depends on a clear contract definition of what counts as resolved.
Independent estimates from competing vendors suggest a platform fee in the tens of thousands of dollars a year plus usage, but these are estimates from rivals, not Decagon figures. Treat them as a rough sign that Decagon is priced for enterprises, and get a written quote.
Budget for the hidden costs too: internal engineering time to connect CRMs, billing and order systems, and staff time to write and test procedures. On a per-conversation contract, short chats, misroutes and spam can count unless the contract excludes them.
Prices in US dollars, checked on Nov 21, 2026. Confirm on Decagon's own site before you buy.

