Skip to content

Decagon

Enterprise AI agents for chat, voice and email, defined in plain-language procedures and sold to large consumer brands.

Reviewed by Andrew Miller · Updated

Our rating

4.4/5

Decagon is the enterprise support agent for buyers who would rather pay per conversation than argue over what counts as resolved.

What it is

Decagon builds AI customer service agents for large companies. Instead of a visual flow builder, teams describe how the agent should handle each situation in Agent Operating Procedures (AOPs), written in natural language. The same agent works across chat, email and voice, with memory across channels, testing and simulation tools, and an analytics suite.

At its Dialogues event on 1 October 2026, Decagon announced Voice 3, running on its own speech model called Chord, which slows down for phone numbers and confirmation codes and talks through a customer's 'mhm' while yielding to a real interruption. It also launched Duet Apprentice, which learns from internal documents and escalated cases inside Slack and Microsoft Teams; a Personal Agent Gateway and an open protocol for AI agents that customers send on their behalf; and Agent Modules for lead qualification, onboarding and collections in five industries.

Decagon does not publish prices. Customers choose per-conversation or per-resolution billing, with no seat fees. In January 2026 it raised $250 million at a $4.5 billion valuation; named customers include Chime, Affirm, Block, Hertz, Avis Budget Group and Deutsche Telekom.

Our review

Decagon competes with Sierra for the largest consumer brands, and the clearest difference between them is commercial. Decagon offers a per-conversation rate as well as a higher per-resolution one, and says most customers take the former to avoid renegotiating what a 'successful outcome' means. That gives finance a predictable bill, at the cost of paying for conversations the agent fails to close.

AOPs are the practical core. Support leaders change agent behaviour by editing plain-language procedures rather than filing an engineering ticket, and Duet Apprentice now drafts those edits from escalated cases. Decagon's claimed average deflection rate of above 80%, quoted in SiliconANGLE's funding coverage, comes from the company and should be read as a best case.

On data, the security page says Decagon runs zero data retention with model providers such as OpenAI and Anthropic, and redacts personal data from logs and transcripts with Google's DLP service once a conversation ends. It shows SOC 2, HIPAA, ISO and PCI badges without stating report types, and says nothing about where data is stored, so regional residency is a question for the sales call.

The October releases push voice further than most rivals. Chord, its own speech model, handles barge-in and pacing, and the Personal Agent Gateway addresses a problem few vendors have touched: a customer's own AI assistant calling in to cancel a subscription. What Decagon still lacks is a way for a mid-size team to evaluate it. There is no trial, no self-serve plan and no public price, and independent contract data is scarce enough that buyers mostly hear figures from Decagon's competitors.

Where it earns its place

  • The VP of support at a US neobank with millions of customers

    Automate card disputes, balance questions and account changes on chat and phone

    One agent covers chat, email and voice, with AOPs written by the support team. Per-conversation billing keeps cost predictable at high volume; Decagon says volume lowers the unit rate, but rates are not published.

  • A support operations analyst at a travel company

    Keep agent procedures current as policies change

    Duet Apprentice reads internal documents and escalated cases in Slack or Teams and drafts AOP updates for a person to approve, which shortens the loop between a policy change and the agent's behaviour.

  • A security engineer reviewing the vendor

    Confirm how conversation data is handled before go-live

    Decagon states zero retention at model providers, Google DLP redaction after each conversation and AES-256 at rest; SOC 2 and other reports are requested via trust.decagon.ai. Ask directly where data is hosted.

Before you commit

Setup
Demo and enterprise contract first; no trial or self-serve plan, and rollouts need internal time to write AOPs and connect systems.
Integrations
Chat, email and voice channels; connections to CRM and billing systems for actions such as refunds; Duet works in Slack and Microsoft Teams.
Data and privacy
Zero data retention with AI providers; PII redacted with Google DLP after each conversation; AES-256 at rest, TLS 1.2+ in transit; data location not published.
Cost at scale
Not published; billed per conversation or per resolution with volume discounts and no seat fees, under annual contracts.
Limits
Supported languages, latency and rate limits are not published.

What's good

  • Choice of per-conversation or per-resolution billing, no seat fees
  • Own voice model (Chord) with natural turn-taking
  • Plain-language procedures support teams can edit
  • Zero data retention with model providers; automatic PII redaction

What's not

  • No public pricing, trial or self-serve plan
  • Performance figures come from the company
  • Certification types and data residency not stated publicly
  • Rollouts need internal time and resources

Best for

Large consumer companies with high chat and phone volumes.

Alternatives

Decagon: common questions

Is Decagon worth it?
Decagon is the enterprise support agent for buyers who would rather pay per conversation than argue over what counts as resolved. We rate it 4.4 out of 5.
How much does Decagon cost?
Enterprise contracts only; per-conversation or per-resolution billing, no seat fees; no public rates.
What are the downsides of Decagon?
No public pricing, trial or self-serve plan; performance figures come from the company; certification types and data residency not stated publicly; rollouts need internal time and resources.
Who is Decagon best for?
Large consumer companies with high chat and phone volumes.
How much does Decagon cost for a team or in production?
Not published; billed per conversation or per resolution with volume discounts and no seat fees, under annual contracts.
Is Decagon safe for company data?
Zero data retention with AI providers; PII redacted with Google DLP after each conversation; AES-256 at rest, TLS 1.2+ in transit; data location not published.
What are the best Decagon alternatives?
The closest alternatives we have reviewed are Sierra, Intercom Fin and Ada.

Answers checked Oct 22, 2026.

Make Decagon? Show this rating on your siteDecagon rated 4.4 out of 5 on ReviewRamble