Why Decagon at $4.5B exposes Sierra's pilot trap

The math that matters: 128× revenue for Decagon vs. 66× for Sierra. Coatue and Index Ventures aren''t paying a brand premium — they''re paying a velocity premium. Decagon''s customer base is Duolingo, Notion, Webflow, and Eventbrite. Internet-native companies that ship agents to production in 60-90 days. Sierra''s flagships are WeightWatchers, SiriusXM, Sonos, and ADT — enterprises that take 9-18 months to clear procurement, security review, board sign-off, and a voice-of-customer review committee. Same product category, two completely different time-to-revenue curves, and the secondary market just told you which one it values.

The conversion problem is the real story under the funding round. IDC says 88% of AI proofs of concept never reach production. A March 2026 enterprise survey put it tighter: 78% of Fortune 1000s have at least one agent pilot running, but fewer than 15% reach production by month 12. Customer service is the cleanest exposure to that failure mode in the entire AI stack, because CX is the one workflow where ROI is unambiguous — deflection rate × ticket volume × cost-per-contact. If a CX agent doesn''t ship, it''s not the model''s fault. It''s the buyer''s procurement velocity. Decagon''s premium says the smart money has stopped betting on model quality and started betting on which customer base actually deploys.

Bain ran a regression on 200+ agentic deployments and found four factors explain 71% of the variance between top-quartile and bottom-quartile payback: (1) eval spend above 15% of total program budget, (2) named executive sponsor at C-1 or higher with the budget authority to fire the vendor, (3) one success metric defined at kickoff and not revisited mid-flight, (4) live integration with the system of record — Salesforce, Zendesk, ServiceNow, or Snowflake — within the first 60 days. Internet-native buyers default to all four. Legacy enterprise buyers default to zero, then wonder why their pilot stalled in week 14. Vendors live or die by which side of that buyer split they sell to.

The unit-economics divergence is already visible. Sierra hit $100M ARR in 7 quarters; Decagon will hit it in 4-5 on current pace. Sierra''s reported NRR is roughly 121%, Decagon''s 118% — functionally identical product retention. But Decagon''s customers expand 3× faster because they instrument agents from day one and ship the second one within 90 days of the first. Sierra''s customers run a 14-month review committee before authorizing agent #2. That gap compounds quarterly. Three years from now Decagon will have 4× the deployed agent count even if Sierra stays ahead on ARR — and deployed agent count is the metric that determines who gets bought, who buys, and who gets re-priced.

This is the playbook signal you can act on this week. If you''re picking a CX agent vendor in Q2 2026, the question isn''t which platform is technically superior — Decagon, Sierra, Cresta, Crescendo, Parloa, Kore, and Ada all benchmark within 4 points of each other on autonomous resolution rate. The question is whether you can run your buying process like Duolingo, not like ADT. If you can clear procurement in 60 days, define one metric at kickoff, put a VP-level sponsor on the deployment, and write the system-of-record integration into the SOW, any vendor in the top six wins. If you can''t, no vendor wins, and you''re the 88% statistic the LPs are quietly pricing into the next round.