Decagon vs Forethought

Decagon and Forethought both promise autonomous AI resolution of support tickets — not copilots that suggest replies, but agents that close the conversation and escalate only what they can’t. The teams choosing between them are enterprise and upper-mid-market support orgs deciding which AI layer to put in front of tier-1 volume so human CSMs stop burning hours on password resets. The core split is twofold. First, architecture: Decagon bets on one agent engine driven by plain-language Agent Operating Procedures (AOPs) that take real backend actions; Forethought ships a multi-agent stack (Solve, Triage, Assist, Discover, Agent QA) with per-customer models trained on your ticket history. Second, and more decisive for 2026 buyers: Forethought is now a Zendesk product — Zendesk acquired it in March 2026 — while Decagon is still an independent vendor. That ownership fact reshapes the decision more than any feature comparison.

Where Decagon wins

• Action-taking via AOPs. Decagon’s differentiator is that non-technical CS and support ops teams author multi-step workflows in plain language, and the agent executes real actions — process a refund, verify identity, change an order, cancel a subscription — not just answer knowledge-base questions. Forethought’s Solve resolves end-to-end using your policies and content, but Decagon’s AOP layer is the more direct path to “the agent did the thing” rather than “the agent answered about the thing.”
• No data floor. Decagon does not require a corpus of historical tickets to train per-customer models. Forethought recommends 20,000+ historical tickets and ~2,000+ tickets/month before its models perform; below that the AI underperforms and the per-resolution math inverts. If you have rich knowledge and workflows but thin ticket history, Decagon starts producing resolution without a training gate.
• Multi-channel with shared context, including voice. Decagon runs chat, email, SMS, and voice (built with ElevenLabs) off one centralized knowledge and workflow layer with cross-channel memory — a customer who started in chat doesn’t re-explain on a call. Forethought’s Solve covers chat, email, and voice too, but Decagon’s cross-channel memory and voice maturity are further along.
• Vendor independence. Decagon is a standalone product with a roadmap it controls. For a team that does not run Zendesk and does not want its AI support layer absorbed into a help-desk vendor’s platform strategy, that independence is a feature.

Where Forethought wins

• First-party on Zendesk. If you already run Zendesk or are moving to it, Forethought is now first-party (“Forethought AI Agents by Zendesk”) and the integration debt is lowest. The Triage, Assist, and Agent QA agents slot into the help desk you already operate. Decagon integrates with Zendesk but is a third-party layer on top.
• Triage and the CS signal layer. Forethought’s Triage auto-classifies and routes inbound at scale, producing clean, tagged ticket data that CS and RevOps wire into churn-risk and NRR models downstream. Decagon is built to deflect and resolve; Forethought’s Triage is purpose-built to feed the signal layer that retention forecasting runs on.
• Agent-assist in one platform. Forethought’s Assist is a copilot inside the help desk for the tickets that do reach a human, and Agent QA scores 100% of human-agent tickets automatically. Decagon focuses on autonomous resolution; if you want autonomous resolution plus human-agent copilot plus QA from one vendor, Forethought’s stack is broader.
• Self-improving loop. Forethought’s Resolution Learning Loop detects workflow gaps, drafts new procedures, and tests them before deployment. Decagon’s Watchtower (always-on QA) and Experiments (live A/B testing) cover monitoring and tuning, but Forethought’s loop is more explicitly aimed at closing gaps automatically — the capability Zendesk paid up for.

Pricing reality

Both are custom, quote-only, with no public self-serve tier — and both carry six-figure-capable enterprise floors. Decagon’s reported structure is an annual platform fee around $50,000 plus usage, charged either per-conversation or per-resolution (reportedly around $0.50/resolution), with conversation minimums in enterprise contracts; median reported annual spend lands materially above the platform fee once volume is counted. Forethought’s third-party marketplace data (Vendr) puts annual contracts roughly in the $36K–$151K range, median near $75K/year, keyed off ticket volume, agent count, channels, and which agents you enable. At comparable mid-volume scope the two land in the same band; the gap is not the headline. The cost trap is different for each: Decagon’s is the platform-fee floor (dead weight below a few thousand conversations/month); Forethought’s is the data floor (you pay enterprise rates while the model has nothing to train on under ~20K historical tickets).

Implementation effort

Neither is a switch you flip. Decagon implementation is a project: you author the AOPs, wire integrations (Salesforce, Intercom, Zendesk, Kustomer), and tune for weeks before deflection lands where the deck promised — budget CS-ops time, not just license. Forethought adds a hard prerequisite on top of the integration work: the data floor. Below ~20,000 historical tickets the per-customer models are weak, so the pilot has to clear a measured resolution-rate threshold before full rollout makes sense. Decagon’s ramp is bounded by how fast you can author workflows; Forethought’s is bounded by both workflow setup and whether your ticket history is deep enough to train on. If you run Zendesk, Forethought’s wiring is lighter; if you don’t, get written commitment on cross-platform support timelines before signing multi-year, because the standalone roadmap is being absorbed into Zendesk’s Resolution Platform.

Bottom line

• Pick Decagon if you need the agent to take real backend actions (refunds, account changes) via plain-language AOPs, you don’t run Zendesk, you have thin ticket history but strong knowledge/workflows, or you want a voice-capable autonomous layer from a vendor that controls its own roadmap.
• Pick Forethought if you’re already on Zendesk (the integration debt is lowest as a first-party product), you have the 20,000+ historical tickets to train on, or you want autonomous resolution plus agent-assist plus Triage feeding your NRR models from a single platform.
• Pick neither if you’re sub-20,000 tickets/year or doing only a few thousand conversations a month. Both carry enterprise floors that don’t pay back at that volume — Intercom and its Fin agent (pay-per-resolution, no volume floor) deliver more value per dollar in that band. And remember neither tool retains: pair the winner with a CS platform of record — Gainsight, Totango, or ChurnZero — that owns health scores, renewals, and expansion.

If you’re choosing in a vacuum without the conditions above, pick Decagon. Vendor independence plus no data floor makes it the lower-risk default in 2026, while Forethought’s roadmap is mid-absorption into Zendesk. Switch to Forethought when you standardize on Zendesk and the first-party integration becomes load-bearing.