While everyone is still competing over “which LLM is smarter,” Decagon is doing something more fundamental: redefining the category of customer support software.
Over the past two years, at least 30 companies have entered the AI customer service agent market. Zendesk has AI. Intercom has Fin. Cresta, Kustomer, Forethought, and many others all say the same thing: our AI understands customer issues better.
But one company founded only in 2023 has quietly won Rippling, Chime, Oura, ClassPass, Substack, Avis Budget Group, 1-800-FLOWERS, and many other well-known brands in roughly 18 months.
That company is Decagon.
Decagon does not publish a $29/month pricing page. It sells through enterprise sales. But customers are willing to wait for sales calls.
Why? Because Decagon asks a different question from most competitors.
Not “AI Conversation,” but “Agent Operations”
Most AI customer support products start from the question: how can AI answer customer questions more accurately?
Decagon starts from a different question: how does an enterprise team manage AI customer support?
That subtle difference produces a very different product shape.
Look at Decagon’s product matrix and you find more than a smart chatbot. It is an agent operations system:
AOPs, or AI operating procedures. Decagon turns complex routing, escalation, tagging, and support workflows into a visual configuration panel. Support leaders can define when AI should resolve an issue, when it should hand off to humans, and when it should escalate to a manager, without needing engineers.
Watchtower. This monitors every agent’s performance in real time. If an agent’s accuracy drops on a class of issues, the system alerts the team. Enterprises no longer need to fear “AI going out of control” without visibility.
Experiments. This enables A/B testing in production. Companies can run two versions of an agent configuration and see which one produces higher deflection. AI optimization becomes data-driven rather than mystical.
Together, these features do something important: they move the purchase from “AI capability” to “AI operations system.”
The former is a tool. The latter is a management capability. Once a customer learns the AOP workflow, depends on Watchtower monitoring, and uses Experiments as an optimization loop, leaving Decagon is not just switching AI models. It is replacing the operating system for support agents.
That is Decagon’s real moat.
ROI Transparency Is the Most Efficient Sales Strategy
Open Decagon’s case-study pages and you see the same pattern:
- who the customer is, industry, and scale
- what problem existed before
- precise numerical outcomes
Examples include:
- Rippling: 32% deflection lift, supporting 400,000+ users and 12+ product lines
- Curology: 65% reduction in customer support operations cost
- Valon: more than 50% deflection in voice channels
- ClassPass: 10x deflection improvement
- Oura: 3x CSAT improvement
There is no vague “efficiency improved significantly.” Every case has specific percentages tied to business scenarios.
The logic is simple. When a customer support VP at a financial services company sees that Chime achieved 70% conversation resolution with Decagon, they immediately benchmark against their own number. If their team’s current resolution rate is 40%, they have a reason to contact sales.
ROI transparency turns the customer’s own pain into a sales lead.
That is why Decagon does not need to publish pricing or run broad ads. Each case study functions as customer acquisition.
Forward-Deployed Engineering: Using Human Effort to Offset AI Uncertainty
One phrase appears repeatedly in Decagon’s customer cases: forward-deployed engineering support.
In Rippling’s case, Decagon engineers reportedly helped build 75+ tagging systems in one week and replaced a deprecated product’s support tooling in one day.
For enterprise AI sales, this is underrated and crucial.
The biggest concern in enterprise AI procurement is not price. It is uncertainty: will your AI really handle our business scenario?
Decagon’s answer is to send engineers into the customer’s environment and solve concrete problems with the team. The first months of engineering investment may not maximize immediate profit, but they build trust.
Once trust is established, the system goes live, and ROI data appears, the customer enters a flywheel of renewal, expansion, and referral.
This model has scaling pressure. But for an early AI product, forward-deployed engineers may be one of the highest-ROI investments for winning the first landmark customers.
Four Transferable Lessons for AI Founders
1. Build an operations platform, not just AI capability.
If your product is only an API that answers questions, your moat depends on how long that API remains best. But if you build an operating system around the AI capability, including workflows, monitoring, alerts, and experiments, your product becomes part of the customer’s operating process. Switching costs become much higher than changing APIs.
2. Build ROI measurement into the product.
Decagon can publish precise deflection and cost-reduction metrics because the product itself includes analytics. From day one, an AI product should answer: how will the customer prove our value with data?
3. Use case studies instead of ads.
Decagon appears to do little brand advertising. Its marketing ammunition is customer stories. If an AI product cannot produce case studies that make prospects contact sales, the problem may not be painful enough.
4. Expand deep before broad.
Decagon started with chat, expanded to email, then voice, and launched Proactive Agents in 2026. Each new channel builds naturally on the last. Existing chat customers are far more willing to try voice than cold prospects.
What to Watch
Decagon’s biggest challenge is that the category is crowding quickly. Zendesk, Intercom, HubSpot, and other established SaaS companies are improving AI capabilities, while Cresta, Kustomer, and other startups compete for large customers.
For Decagon, the next 12 to 18 months hinge on whether it can maintain product depth while moving from purely sales-led growth toward some product-led motion, such as self-serve versions that accelerate acquisition.
But if it continues to combine product depth with ROI transparency, it will be difficult to replace.
Closing Thought
Decagon’s lesson is not that customer support needs a smarter chatbot.
It is that enterprises do not buy “AI conversation” for long. They buy systems that let them operate AI safely, measure it, improve it, and trust it.
In that shift from conversation to operations, Decagon found its category.
Data note: Customer data and ROI metrics come from Decagon’s website and public case-study pages and are marked by the company as not independently audited. Customer brand information comes from Decagon’s website and public company materials. Product descriptions are based on Decagon’s official site.