Decagon

Decagon positions an enterprise AI concierge platform spanning chat, voice and email, with AOP workflow building, optimization, analytics and current aggregate customer-result cards.

The About page names Jesse Zhang as co-founder and CEO and Ashwin Sreenivas as co-founder and President, identifies enterprise buyers and investors, and publishes aggregate customer scale and outcome cards.

The unified platform covers AOPs, Duet, memory, guardrails, integrations, simulations, versioning, A/B tests, traceability, diagnostics, Watchtower and chat, voice and email deployment. Operators author logic while technical teams retain integration, version and guardrail control.

AOPs define inbound and outbound multi-step workflows in natural language, integrate with support tools, pull live data and trigger workflows. Git-based tracking, reasoning traces, runtime and latency, testing, versioning and alerts support control and observability.

Current integrations include Salesforce, Intercom, Zendesk, Confluence, Contentful, Kustomer, Amazon Connect and RingCentral, plus self-serve APIs, custom tools, MCP, SIP trunking, data retrieval, action triggers and escalations.

Voice agents support inbound and outbound calling, 70+ languages, automatic language switching, interruptions, hundreds of voices, guardrails and human transfers with concise summaries.

Chat operates across web, mobile, SMS, WhatsApp and other messaging, supports structured inputs, multi-brand experiences, product discovery, WISMO, subscription changes, upsell, cross-sell, loyalty and end-to-end inbound resolution.

Email agents handle contextual threads, customize tone, route and respond through Intercom, Zendesk and Salesforce, and escalate through tags and ticket integrations with context for human agents.

Duet generates AOPs from conversations, detects gaps, generates simulations, explains real-time traces, incorporates human QA and produces reports and targeted recommendations. Updates are staged for human review with traceability.

Autopilot identifies production issues, proposes agent-logic updates, iterates against original and golden test sets and expands tests automatically. Every proposed update requires human approval before production, and reviewers see diffs, issues and validation results.

The launch combines persistent cross-session and cross-channel user memory, outbound voice and Agent Workbench. Outbound AOPs handle appointment coordination and informational calls with follow-up actions; Workbench analyzes logs, reasoning, traces and latency.

User memory stores unstructured history and structured preferences or signals, exposes full APIs, supports import/export, opt-in storage, redaction, expiration controls, AOP-scoped memory calls, audit logs and customer consent flows.

Guided Discovery handles open-ended product discovery, retention and expansion conversations by asking follow-up questions, using context and guiding customers to an outcome instead of requiring a predefined workflow for every path.

Simulations validate actions, tools, data and business logic across generated scenarios; scheduled runs, CI/CD integration, root-cause suggestions and PagerDuty alerts support regression management.

The current A/B testing suite splits live traffic and measures tone, AOP, refund-logic and onboarding variants against CSAT, deflection and statistical significance.

Trace View exposes step-by-step agent execution, rationale, model and tool activity, errors and latency to diagnose, audit and refine workflow behavior.

Watchtower continuously reviews AI and human conversations against natural-language criteria for compliance, sentiment, CSAT and upsell signals, with filters and real-time surfacing.

Dashboards track CSAT, deflection, AOP and subtopic performance; natural-language analysis explains refund requests and filters by channel, escalation, ratings and other segments.

Suggestions detects knowledge gaps, generates article drafts from strong human resolutions, groups drafts for review, refreshes suggestions monthly and supports export; it does not prove automatic knowledge-base publication.

Decagon states Google Cloud hosting, perimeter controls and vulnerability scanning; TLS 1.2 in transit, AES-256 at rest, production audit logs, SSO, 2FA, least privilege and RBAC. No certification report or scope is published on this page.

The policy names Decagon AI, Inc., gives a San Francisco mailing address, describes enterprise-only services and separates website data from customer-controlled service data governed by customer agreements.

Rituals confirms that Ray is developed and operated by Decagon AI, Inc., integrates with backend systems for FAQs and order tracking, operates autonomously only within defined boundaries, makes no independent decisions affecting rights or service access and escalates human-judgment cases through supervision.

Decagon states it offers per-conversation fixed pricing with volume flexibility and higher per-resolution fixed pricing with no charge for escalations and lower rates at larger commitments. Numeric prices, minimums and standard contract values are not disclosed.

Decagon says it is available through Google Cloud Marketplace, is natively built on Google Cloud and can apply committed cloud spend, consolidate billing and use Partner Essentials training.

The accredited Five9 integration uses Five9 for engagement, routing and workforce infrastructure while Decagon executes voice workflows such as refunds and account updates, handing transcripts, data and intent to humans when required. Joint customers are named.

Decagon announced a $5M seed led by a16z and $30M Series A led by Accel, with A*, Elad Gil and named angels. It then described the founders as CEO Jesse Zhang and CTO Ashwin Sreenivas and direct actions such as refund requests and shipment postponement.

Decagon announced a $65M Series B led by Bain Capital Ventures with total funding then at $100M. The article gives named Bilt deployment claims and notes some customers handled millions of annual conversations at 90% resolution.

Decagon announced a $131M Series C at $1.5B valuation co-led by Accel and a16z, with existing and new investors. It claimed near-70% average deflection, named customer outcomes, tens of millions of end users helped and 0-to-eight-figure ARR in the prior year.

Decagon announced $250M led by Coatue and Index at a $4.5B valuation after more than 100 new global enterprise customers in the preceding fiscal year, and enumerated participating investors. Current fully diluted ownership remains undisclosed.

The filing identifies Decagon AI, Inc. as a Delaware corporation incorporated in 2023, headquartered in San Francisco, names Jesse Zhang and Ashwin Sreenivas, and reports $61,524,976 sold of a $64,999,978 equity offering to nine investors. Revenue range was declined.

Eventbrite disclosed a May 2025 one-year agreement with Decagon AI, Inc. for a $0.3M fee and described Decagon as a customer-support AI platform. It also disclosed A-Star Partners Fund I's minority stake and related-party connection.

The London regional team reached 50+ and serves EMEA. The article reports named Rituals, Hunter Douglas and Oura outcomes, including Oura's 3x CSAT, 93% quality and tens of millions in hiring and support savings.

Retail workflows cover product recommendations, loyalty, upsell, WISMO, returns and account or order updates across channels. Decagon says agents integrate with Shopify and take actions under business policies; listing, inventory and fulfillment authority are not shown.

Travel workflows include reservation changes, rebooking, upgrades, loyalty and proactive disruption communications using connected systems, with voice, chat and email support.

The current hub links 22 live case-study URLs. One URL labeled Dutchie currently renders the Rippling title and story, leaving 21 distinct named hub cases after conservative deduplication.

The company publishes anonymous deployment examples: five workflows built in three weeks versus nine months with a prior vendor, an airline under three weeks, a music platform in six business days at 71% handled and 4.2 CSAT, a social network at 80% in two months and a health insurer with five use cases in one month.

A 45-day live head-to-head pilot closed near 93% CSAT, described as double-digit points above prior automation. Decagon later supported three named brands, with wider portfolio expansion planned.

The current official page identifies Away as a Decagon customer and describes fast, reliable and memorable support, but retained page text exposes no quantified result, baseline or workflow detail.

The story reports 70%+ chat resolution, nearly 70% voice resolution, more than one million calls monthly, hundreds of thousands of messages automated in two weeks, 60% lower support cost and doubled member satisfaction; it also shows card guidance, deposit status and SMS subscription workflows.

ClassPass reports 95% cost reduction, a 32% deflection increase and chat expansion from 16 hours on five days to 24/7. A testimonial says launch deflection was 10x the team's prior expectation, not 10x its baseline rate.

Curology says chat's ticket share rose from 5% to 80%, support operations cost fell 65%, surge volume was handled without added headcount and agents could update accounts, replace orders, change shipments, request refunds and check shipment status.

Duolingo reports 80% chat deflection, one-month implementation and hourly FAQ synchronization. A prior vendor had about 30% email deflection but never launched chat after a year, so that rate is not used as a like-for-like baseline.

Despite its Dutchie URL, the live page title, body, speaker and metrics are the Rippling case. It is retained as a URL anomaly and is not counted as a distinct Dutchie case.

Faire describes streamlined returns initiation, fewer repetitive contacts, lower operating costs and faster decisions, but provides no quantified Decagon outcome. Hundreds of thousands of brands and retailers describes customer scale, not impact.

Flashfood says Decagon handles all incoming support tickets and resolves more than 90% without human intervention, including complex and time-sensitive inquiries.

Fourthwall reports more than 70% of incoming tickets resolved without human review. Unresolved cases escalate through its ticketing system with context preserved.

The current official clip page identifies Hertz as using proactive outbound agents to anticipate needs and resolve issues before inbound contact, but retained page text exposes no quantified result or workflow baseline.

Hunter Douglas reports more than $1M revenue from fully AI-handled conversations, 85% higher AOV for customers who engaged with agents, 40% average deflection and scale from 10% to full traffic in one week across localized ecommerce deployments.

Mercado Libre confirms progressive Decagon voice rollout, Portuguese QA, explicit escalation guardrails and Watchtower use, but publishes no quantified Decagon outcome. Two billion deliveries, 120M buyers and 18 countries are customer scale.

NG.CASH reports deflection rising from 13% to 70%, avoiding more than 35 hires, absorbing 3–4x inquiry volume and reaching six million accounts after an acquisition without expanding support headcount.

Noom reports deflection moving from 59% to 68% within 30 days while handling a major seasonal support-volume surge.

Notion reports up to 34% better ticket-resolution time, a twofold deflection increase and a 3.4% average ask-for-human rate across roughly one million annual inquiries.

Rippling reports chat deflection rising from 38% to more than 50%, a stated 32% increase, 7% better routing, 75+ tags and support across 12+ product lines for 400,000+ users. The duplicate Dutchie URL renders this same story.

Rituals reports expansion to 19 countries and 15 languages within two months, zero backlog from Black Friday through December 26 while handling up to 9,000 daily cases, and a historical need for 300 seasonal agents. Planned resends and compensation are not current proof.

SimplePractice reports 85% deflection within one month, above a 66% end-of-year target, for a platform serving 250,000+ practitioners. The story also describes redaction testing and soft human handoffs.

Substack reports more than 90% resolution without human intervention and no team-size increase. Connected APIs execute refunds and cancellations end-to-end; integrations identify and report feature requests and bugs.

Valon reports 50%+ inbound voice deflection, about 90% CSAT and under-one-minute answer time while onboarding 80,000 loans in one month and handling call volumes 2–3x prior highs.

Whop reports 65–70% deflection against 5,000–10,000 weekly tickets and 50% fewer payment-related tickets after Decagon insights led to product fixes. The headline rounds deflection to 70%.