Decagon

Sweet spot: a consumer brand or fintech doing 50k+ tickets a month, with clean historical ticket data and a CX leader who treats deflection rate as a first-class metric. Decagon's engineering depth pays off when you have an internal team that can iterate on policies, evaluate agent behavior, and integrate the agent with your real systems of record. The Klarna-style headlines are real, but they were earned by buyers who invested heavily on their side. Failure modes to plan for. Outcome-based pricing is honest but volatile — model your worst-month scenario before signing, and negotiate caps. AI customer service can damage brand trust faster than it builds it: a confidently wrong refund denial is worse than a slow human response, so invest in red-teaming and a strong human-handoff path. Regulated industries (banking, health, insurance) need compliance review before any go-live; do not let a vendor pitch skip this. And do not believe the "weeks to deploy" line — plan for at least one full quarter end-to-end. What to pilot. Pick one channel (usually chat) and one ticket category with clear policies (order status or refund eligibility). Set a 30-day target deflection rate and a CSAT floor. If Decagon clears both with your own data and your policies, expand category by category — never channel by channel without re-validating. Treat the pilot as a software project with eval gates, not a procurement exercise.

Decagon builds autonomous AI agents for customer support, positioning itself as the engineering-led counterweight to Sierra and Ada. The product centers on its in-house reasoning engine — an agent runtime that ingests your help-center, past tickets, and CRM data, learns the policies and tone, then resolves end-customer conversations end-to-end across chat, email, and embedded widgets. Where most "AI customer support" pitches stop at deflecting FAQs, Decagon's agents take real actions: process refunds, update subscriptions, look up order status, and escalate cleanly when policy requires a human. The competitive frame is Decagon vs Sierra vs Ada. Sierra leans brand-experience and voice; Ada is the longest-running pure-play with a polished low-code builder. Decagon's differentiation is engineering depth — APIs, observability, and a posture aimed at companies who want to instrument and iterate on agent behavior the way they would on any other production system. Reference customers include Klarna-tier consumer brands, fintech, and high-volume D2C operators where each percentage point of deflection is a measurable line on the P&L. Pricing is enterprise sales-gated and predominantly outcome-based: customers pay per resolved conversation rather than per seat or per message. That alignment-of-incentives pitch is genuinely compelling in board decks, but it also means cost variability scales with traffic in ways procurement teams have to model carefully — a viral spike or a botched product launch can produce a Decagon invoice you did not budget for. Implementation is non-trivial. Expect 4–8 weeks for a real production rollout with measurable deflection: data pipelines from Zendesk/Salesforce/Kustomer, prompt and policy authoring, evaluation harnesses, and the inevitable round of red-team testing before the agent talks to real users. Decagon's solutions team is competent, but the buyer-side investment in training data quality and human-in-the-loop QA is the single biggest predictor of outcomes.