top of page
Business & Enterprise
AI Technology
Customer Support Software
Decagon
Replaced fragmented support automation by deploying enterprise AI agents that authenticate users, retrieve data, take actions, and resolve issues end to end, despite Intercom’s embedded helpdesk advantage and Sierra’s bespoke-agent positioning.
1
MODEL
BUSINESS MODEL
SaaS
- Founded in 2023 by Jesse Zhang and Ashwin Sreenivas; emerged from stealth in 2024 with enterprise AI customer-support agents.
- Built Agent Operating Procedures, allowing CX teams to define agent behavior in natural language rather than rigid flow builders.
- Connected agents to ticketing systems, customer databases, authentication, order systems, and backend tools so they could retrieve data and take actions.
- Expanded across chat, email, voice, and SMS, targeting enterprises with high-volume support operations.
- Signed more than 100 global enterprise customers in 2025, including Avis Budget Group, Block, Deutsche Telekom, Affirm, Chime, Duolingo, Notion, Rippling, Eventbrite, and Bilt.
- Raised $131 million Series C at a $1.5 billion valuation in June 2025, then $250 million Series D at $4.5 billion in January 2026; reached approximately $35 million ARR by October 2025 and an estimated $100 million annualized revenue by July 2026.
1. Identify a high-volume operational workflow where humans repeat the same data lookup, decision, and system action.
2. Build an agent that can both understand intent and complete the action inside the customer’s systems.
3. Make behavior configurable in business language, so domain owners—not only engineers—can govern the workflow.
4. Start with a narrow, high-volume, low-risk use case where success is measurable.
5. Add authentication, permissions, audit trails, and escalation before expanding autonomy.
6. Price around the completed outcome or managed interaction, not seats alone.
7. Avoid promising autonomy before the agent can act safely within permissions and recover from failure. The common mis-build is building a conversational layer before solving system integration, identity, and action reliability.
DISTRIBUTION MODEL
Enterprises faced an urgent support problem: high-volume, repetitive inquiries consumed agent capacity, while older chatbots could answer FAQs but could not authenticate users, inspect orders, process returns, or update systems. The existing workaround was large support teams, rigid decision trees, and human escalation. Decagon’s Agent Operating Procedures let companies define agent behavior in natural language while connecting the agent to backend systems; ClassPass reported a 95% decrease in support-conversation cost, making the pain and payoff measurable.
HOW THEY BUILT IT
HOW TO ARCHITECT IT
Enterprise Sales, Direct Sales, Product-Led Growth, Partnerships
HOW THEY OPERATIONALIZED
HOW TO REPLICATE WHAT WORKED
dm
- Enterprise sales targeted CX, support, and operations leaders at high-volume consumer and financial-technology companies.
- Product-led demonstrations showed agents resolving real customer scenarios, including authentication, order lookup, returns, and account updates.
- Integrations with ticketing systems, CRMs, and customer databases reduced the need to replace the entire support stack.
- Reference customers such as Duolingo, Notion, Rippling, Eventbrite, Bilt, Affirm, Chime, Block, and Deutsche Telekom created enterprise credibility.
- AOP-based configuration lowered implementation friction relative to bespoke agent builds, making enterprise rollout more repeatable.
What worked: Demonstrate the agent completing a high-value workflow on the customer’s real systems, then expand from one measurable use case to adjacent actions.
Trap if copied blindly: Agent demonstrations fail if the vendor cannot guarantee permissions, auditability, and failure handling; a compelling demo that cannot operate safely in production becomes an enterprise-sales liability.
| PATTERNS OF THIS MODEL
PATTERNS IN ENTERPRISE AI AGENT PLATFORMS:
1. ACTION COMPLETION CREATES PRICING POWER. Agents that retrieve data and complete transactions command higher willingness to pay than chatbots that only answer questions.
2. NATURAL-LANGUAGE GOVERNANCE REDUCES IMPLEMENTATION FRICTION. Letting business owners define behavior lowers dependency on engineering, but only if guardrails and testing are built in.
3. OUTCOME PRICING ALIGNS VENDOR AND BUYER. Per-resolution pricing transfers performance risk to the vendor and forces the company to measure true resolution quality.
THE WEAKNESS: Outcome-based AI pricing can become margin-volatile when resolution definitions are ambiguous, escalations are frequent, or model and infrastructure costs rise faster than the agreed price.
What companies with this model reveal
| OPPORTUNITY INTELLIGENCE
GOLDMINE 1 — AGENT ASSURANCE LAYER. Enterprises need simulation, regression testing, permission boundaries, and audit trails before autonomous agents touch production systems; this assurance layer is a separate, durable product opportunity.
GOLDMINE 2 — RESOLUTION-QUALITY MEASUREMENT. “Resolved” is often self-reported; a vendor that independently verifies customer outcome, compliance, and satisfaction can price on trusted outcomes.
GOLDMINE 3 — VERTICAL AGENT OPERATING PROCEDURES. Banks, insurers, travel companies, and healthcare providers need prevalidated workflows for identity, refunds, claims, and compliance; vertical templates can reduce deployment time without sacrificing control.
THE PIT — PER-RESOLUTION PRICING WITHOUT A SHARED DEFINITION. If vendor and buyer disagree on what counts as resolved, the model creates disputes and revenue unpredictability.
THE SECOND PIT — CHATBOT RELABELING. A conversational interface that cannot authenticate users, update systems, or complete workflows will not sustain enterprise pricing.
MOVE WITH CAUTION — Do not give an agent write access to production systems before sandboxing, role-based permissions, rollback, and human override are proven; one erroneous refund or account change can erase the trust that justifies enterprise adoption.
Untapped Business Model / Gaps / Goldmines / Pits
Patterns & Insights
2
MARKET
mkt mt es
Decagon entered AI customer support against Intercom Fin, Sierra, Forethought, Ada, Zendesk AI, and legacy chatbot vendors. Its wedge was not another chatbot but an enterprise agent that could authenticate users, access systems, complete actions, and resolve issues end to end. AOPs changed the comparison from “which bot answers questions?” to “which agent can safely execute operational work?”
MARKET TYPE
Red Ocean
ENTRY STRATEGY
Brownfield Entry
WHY THEY WON
AI customer support is a Red Ocean with helpdesk incumbents, AI-native entrants, and foundation-model providers all converging. Decagon’s opening was architectural: most incumbents added AI to existing ticketing workflows, while bespoke vendors built highly customized agents. Decagon occupied the middle with configurable Agent Operating Procedures and system actions. The lesson is that in a crowded AI market, the winner must own a repeatable way to turn enterprise intent into safe, production-grade action.
EXECUTION
Decagon entered by integrating with existing support systems rather than requiring companies to replace their helpdesk. It connected to ticketing platforms, customer databases, authentication systems, and backend tools, then deployed agents across chat, email, voice, and SMS. The critical bet was that enterprises would adopt a standalone agent platform if it could act safely within their existing stack. The prerequisite was reliable integrations, permissions, and evaluation tooling; the approach would fail if the agent could not operate within enterprise security and data-governance requirements.
FOOTHOLD STRATEGY
Wedge Strategy
Decagon’s initial wedge was high-growth technology companies with high-volume, repetitive support workflows—account questions, subscription issues, order status, refunds, and authentication. These companies were reachable through product-led demos and founder networks; they could pay because support cost and response time were material. Decagon solved the immediate job of resolving repetitive requests without human intervention. The path outward was from tech companies to financial services, telecom, travel, and other regulated, high-volume enterprises.
GROWTH CAMPAIGN
CAMPAIGNS THAT WORKED
SO WHAT: Decagon’s strongest campaign was proof of completed action: it showed the agent authenticating a user, retrieving an order, processing a return, and updating systems in one interaction.
KEY LEARNING: In AI markets, buyers do not buy intelligence; they buy a safely completed operational outcome.
HOW TO USE: Choose one workflow where the AI can complete an end-to-end action, then demonstrate it using the customer’s actual systems, permissions, and failure paths.
KEY LEARNING
THE STANDARD: In a Red Ocean, AI entrants win by changing the unit of value from conversation to completed, governed action.
RULE 1 — MOVE FROM ANSWERS TO ACTIONS: A support agent becomes strategic when it can update systems, not merely respond.
RULE 2 — MAKE CONFIGURATION ACCESSIBLE: Business owners must be able to define behavior without engineering dependency.
RULE 3 — SELL SAFETY BEFORE AUTONOMY: Authentication, permissions, auditability, and escalation determine enterprise adoption.
RULE 4 — FAILURE MODE: A crowded AI market punishes vendors whose agents cannot reliably complete or verify the outcome.
MARKET TYPE: Red Ocean (category).
gc
Market Context
| MARKET INTELLIGENCE
PATTERNS IN ENTERPRISE AI-AGENT DISTRIBUTION:
1. REAL-WORKFLOW DEMOS SHORTEN EVALUATION. Buyers trust agents more when they see them operate within actual systems and permissions.
2. REFERENCE CUSTOMERS REDUCE AI RISK PERCEPTION. Enterprise logos matter because buyers are purchasing operational reliability, not just model performance.
3. CONFIGURABLE AGENTS BEAT FULLY BESPOKE BUILDS ON SPEED. Natural-language procedures reduce implementation time, but require strong testing and governance.
4. THE WEAKNESS: Enterprise AI distribution can create a custom-integration treadmill if every customer requires new backend connections and unique failure handling.
| MARKET ENTRY PLAYBOOK
THE STANDARD: Brownfield entry is appropriate when the customer has existing systems of record and the AI can add an action layer without forcing replacement.
RULE 1 — Integrate with existing systems before asking the customer to change their stack.
RULE 2 — Start with read-heavy or low-risk actions before granting write access.
RULE 3 — Build simulation, permissions, and rollback before production autonomy.
RULE 4 — Reject brownfield entry if the customer cannot provide secure system access or if the incumbent helpdesk can nativize the agent capability.
How to enter
| FOOTHOLD STRATEGY PLAYBOOK
THE STANDARD: A wedge strategy works when one high-volume workflow produces measurable cost savings or revenue lift.
RULE 1 — Choose a workflow with clear success criteria and frequent repetition.
RULE 2 — Make the first autonomous action safe, reversible, and observable.
RULE 3 — Expand only after resolution quality and escalation behavior are proven.
RULE 4 — A wedge fails when the AI handles simple questions but cannot complete the operational action that creates value.
How to get the first strong position
MARKET PATTERNS & PLAYBOOK
3
MONEY
money rev pri
Decagon chose interaction- and outcome-based pricing rather than seat-based pricing. It offers per-conversation pricing, with volume discounts, and per-resolution pricing, where the vendor charges more only when the agent fully resolves the issue without escalation. This aligns revenue with the customer’s support workload and the agent’s actual performance, while creating a direct incentive to improve resolution quality.
REVENUE MODEL
Usage-Based, Subscription
PRICING MODEL
Usage-Based Pricing, Performance Fees, Custom Enterprise Pricing
WHY THEY WON
Customers pay an annual platform fee plus per-conversation or per-resolution usage charges. Revenue expands as conversation volume grows, more channels are enabled, and more workflows are automated. Per-resolution pricing carries a higher rate because it transfers outcome risk to Decagon; escalations are not charged under that model. Gross margin depends on model inference, voice and messaging costs, integration maintenance, and the share of conversations successfully resolved.
Decagon uses quote-based enterprise pricing with an annual platform fee, reported by third parties at around $50,000, plus usage charges. Per-conversation pricing charges a fixed rate for each incoming conversation; per-resolution pricing charges a higher fixed rate only when the agent fully resolves the issue, with no charge for escalations. The meter aligns price with support workload and agent performance.
TARGET AUDIENCE
High-growth technology companies (enterprise contracts; want to automate repetitive support at scale); consumer internet and fintech companies (enterprise contracts; want authenticated account and billing support); large enterprises in telecom, travel, retail, and financial services (enterprise contracts; want secure, multichannel AI agents integrated with legacy systems).
CUSTOMER BUYING BEHAVIOUR
A CX or operations leader typically begins after support costs, response times, or ticket volume become unsustainable. They evaluate Decagon against Sierra, Intercom Fin, and legacy chatbots; procurement involves CX, IT, security, legal, and data teams. Adoption expands after a scoped workflow proves resolution quality and safe system access.
| PRICING INTELLIGENCE
What makes this model effective & make customers pay
Customers are not buying AI conversations; they are buying lower cost per resolved customer problem without losing control or trust.
RULE 1 — PRICE THE RESOLVED OUTCOME: Per-resolution pricing ties vendor revenue to the buyer’s desired result, but only when “resolved” is objectively defined.
RULE 2 — CHARGE FOR ACTION, NOT CHAT: The value comes from authentication, data retrieval, refunds, updates, and workflow completion.
RULE 3 — USE A PLATFORM FEE TO COVER GOVERNANCE: Enterprise buyers pay for permissions, auditability, security, and integration reliability.
RULE 4 — WATCH THE OUTCOME-DEFINITION TRAP: If resolution is ambiguous, per-resolution pricing creates disputes and margin volatility.
The buyer is really buying support capacity and resolution quality, evidenced by CX operations and cost-reduction budgets.
PRICE & REVENUE
| Revenue Risk - The biggest threat to revenue stability
Resolution-definition disputes: Per-resolution pricing becomes unstable if vendor and customer disagree on what qualifies as a complete resolution.
Inference and voice-cost pressure: AI, voice, SMS, and integration costs can rise faster than negotiated per-conversation rates.
Escalation concentration: If complex cases escalate frequently, the vendor may provide value without capturing proportional revenue.
Helpdesk nativization: Intercom, Zendesk, and Salesforce can embed AI agents directly into their platforms.
Trust and compliance risk: An erroneous refund, account change, or data exposure can damage enterprise credibility.
Funding/valuation: Decagon raised $231 million across four rounds, including $131 million Series C at a $1.5 billion valuation in June 2025 and $250 million Series D at $4.5 billion in January 2026; it reached approximately $35 million ARR by October 2025, with later $100 million annualized revenue estimates from third parties rather than audited company disclosures.
Where the model can break
4
MOTION
motion ge cs
Decagon sold high-value AI-agent deployments through enterprise sales involving CX, IT, security, legal, and data stakeholders. Its motion combined product-led demonstrations, AOP-based configuration, system integrations, and reference customers. The company signed more than 100 global enterprise customers in 2025, including Avis Budget Group, Block, Deutsche Telekom, Affirm, and Chime, indicating a repeatable enterprise-sales system rather than founder-led selling alone.
GROWTH EXPANSION MODEL
Land & Expand, Horizontal Expansion
HOW THEY EXPAND
Decagon first landed high-growth technology companies with high-volume support workflows, then expanded horizontally into financial services, telecom, travel, retail, and other enterprise sectors. Within accounts, it expanded from one automated workflow—such as order status or authentication—to refunds, returns, account updates, billing, and multichannel support. Expansion across chat, email, voice, and SMS increased conversation volume and account value, while AOPs made new use cases easier to configure.
COMPETITIVE STRATEGY
Differentiation
HOW THEY COMPETE
Against Intercom Fin, Decagon differentiated as a standalone enterprise agent platform that could operate across channels and systems rather than being embedded only in one helpdesk. Against Sierra, it differentiated through configurable AOPs and faster deployment rather than highly bespoke agent builds. Against legacy chatbots, it differentiated through authentication, system actions, and end-to-end resolution. Its constraint is that helpdesk incumbents can nativize agent capabilities within their installed base.
GROWTH ENGINE
GTM
Content Flywheel
An enterprise identifies high support cost or slow resolution → Decagon demonstrates an agent completing a real workflow with appropriate permissions → the customer deploys one scoped use case → successful resolution reduces cost and increases automation coverage → additional channels, languages, and workflows are added → conversation or resolution volume expands revenue. The loop compounds when each new workflow produces more integration, evaluation, and governance capability. It requires reliable system access and clear outcome definitions; it breaks if escalations remain high, integrations become bespoke, or the helpdesk incumbent nativizes the agent.
Decagon runs a sales-led enterprise GTM supported by product-led proof. Direct sales targets high-volume support organizations; solution engineers demonstrate agents on real workflows; security and IT validate permissions and integrations; customer success expands use cases after initial deployment. The primary motion is enterprise sales, with product-led demonstrations reducing evaluation risk. The transferable implication is that an AI-agent company should sell one safely completed workflow first, then expand autonomy only as governance and outcome measurement mature.
SUSTAINING MOATS
High Customer Lock-In
moat
The moat is Decagon’s accumulated enterprise interaction, workflow, and resolution data. Each conversation, escalation, action, and customer outcome improves agent evaluation, routing, and behavior. AOPs turn this learning into reusable operational procedures, making new deployments faster. The moat weakens if customer data remains siloed, integrations are shallow, or competitors achieve equivalent resolution quality with lower cost.
| MOAT INTELLIGENCE
THE STANDARD: An AI data moat lasts when the platform captures the full feedback loop from intent to action to verified outcome.
RULE 1 — ACTION FEEDBACK: Agent actions, escalations, and customer outcomes improve future behavior.
RULE 2 — WORKFLOW REUSE: AOPs convert customer-specific learning into repeatable enterprise deployment patterns.
RULE 3 — MULTICHANNEL CONTEXT: Chat, email, voice, and SMS interactions enrich the agent’s understanding of customer intent.
THE SIGNAL: Decagon added more than 100 enterprise customers in 2025 and reached approximately $35 million ARR by October 2025, with third-party estimates placing annualized revenue near $100 million by July 2026. The moat does not protect against ambiguous resolution definitions, poor escalation quality, or a helpdesk incumbent with native distribution.
Why this company remains defensible
ARR & TAKEAWAY
ARR Journey - what to do at each stage
PRE-$1M ARR — PROVE ACTION-COMPLETE AI. From 2023, Decagon built agents that could authenticate users, retrieve data, and take actions rather than merely answer FAQs. WATCH: system integration and safety.
$1–5M ARR — WIN HIGH-GROWTH TECH WEDGE. Duolingo, Notion, Rippling, Eventbrite, and Bilt adopted Decagon for high-volume support automation. WATCH: resolution quality and escalation handling.
$5–10M ARR — MAKE CONFIGURATION REPEATABLE. AOPs allowed CX teams to define agent behavior in natural language, reducing engineering dependency. WATCH: governance and testing.
$10–50M ARR — ENTER REGULATED ENTERPRISE. Decagon expanded into financial services, telecom, travel, and retail with customers including Affirm, Chime, Block, Deutsche Telekom, and Avis. WATCH: security, compliance, and integration depth.
$50–100M ARR — SCALE MULTICHANNEL AUTONOMY. Decagon expanded across chat, email, voice, and SMS; it raised $131 million Series C at $1.5 billion in June 2025 and reached approximately $35 million ARR by October 2025. WATCH: inference cost and outcome definitions.
$100M+ ARR — CONSOLIDATE THE AGENT PLATFORM. Third-party estimates place Decagon at $100 million annualized revenue by July 2026, after a $250 million Series D at $4.5 billion in January 2026. WATCH: helpdesk nativization and per-resolution margin.
Rule: Prove one safe, completed workflow, make configuration repeatable, then expand autonomy only as governance and outcome measurement mature.
COPY PLAYBOOK : What Worked → What Failed → What to Replicate → What to Avoid
THE STANDARD: Build an AI agent that completes a high-value workflow safely, then expand autonomy and channels only after outcome quality and governance are proven.
SEQUENCE:
1. Choose a high-volume workflow with a clear success definition and measurable cost or revenue impact.
2. Build system access, authentication, permissions, and auditability before granting autonomous action.
3. Make behavior configurable by the business owner, but constrain it with testing and guardrails.
4. Price around managed interaction or verified outcome, with a platform fee covering governance.
5. Expand to adjacent workflows and channels only after resolution quality is proven.
WORKED: Decagon’s AOPs and system actions let enterprises move beyond FAQ chatbots to end-to-end resolution, while enterprise references accelerated adoption.
CAUTION: Do not define “resolution” loosely; also avoid giving agents write access to production systems before sandboxing, rollback, and human override are reliable.
bottom of page