What it is
Forethought is an AI customer-support platform built around a multi-agent stack: Solve (autonomous resolution across chat, email, and voice), Triage (ticket classification and routing), Assist (an agent copilot inside your help desk), Discover (knowledge-gap detection), and Agent QA (automated scoring of human-agent tickets). It grew out of SupportGPT, the LLM-plus-RAG system Forethought shipped in 2023 that fine-tuned per-customer models on ticket history rather than relying on intent-matching alone.
The headline fact for any 2026 buyer: Zendesk announced its intent to acquire Forethought on March 11, 2026 and completed the deal later that month (terms were not disclosed). Forethought now ships as “Forethought AI Agents by Zendesk” — it still runs across non-Zendesk help desks, but it is a Zendesk product line, not an independent vendor, and that changes the buying calculus.
Why it shows up in Customer Success stacks
- Autonomous resolution, not just deflection. Solve closes tickets end-to-end using your policies and content, not canned answers, so the deflection it reports is real resolution rather than a chatbot punting to a queue.
- Triage feeds the CS signal layer. Auto-classification routes and tags inbound at scale, which gives CS and RevOps clean ticket data to wire into churn-risk and NRR models downstream.
- Self-improving loop. The Resolution Learning Loop detects workflow gaps, drafts new procedures, and tests them before deployment — the capability Zendesk paid up for, and the reason it shows up on enterprise shortlists.
Pricing
- Custom only — quote-based; no public self-serve tier. Third-party marketplace data (Vendr) puts annual contracts roughly in the $36K-$151K range, with a median near $75K/year.
- Pricing keys off monthly ticket/conversation volume, agent count, channels deployed (chat, email, voice), and which agents you turn on. Treat per-resolution or per-conversation usage as the variable that drives the bill.
- There is a practical data floor: Forethought recommends 20,000+ historical tickets and ~2,000+ tickets/month to train the per-customer models. Under that volume the AI underperforms and the price-per-resolution math inverts.
Best for
- Mid-market and enterprise B2B SaaS support orgs (typically 25+ agents, high ticket volume) that want autonomous resolution plus agent-assist in one platform.
- Teams already on or moving to Zendesk, where Forethought is now first-party and the integration debt is lowest.
- CS-RevOps partnerships that need clean triage and resolution data feeding renewal-risk and NRR forecasting.
Do not buy Forethought if you are a sub-20,000-ticket-per-year team or an early-stage company with thin ticket history — the model has nothing to train on and Intercom’s Fin (pay-per-resolution, no volume floor) will deliver more value per dollar in that band.
Watch-outs
- Post-acquisition roadmap uncertainty. Forethought is now a Zendesk product and the standalone roadmap is being absorbed into Zendesk’s Resolution Platform. Guard: if you run a non-Zendesk help desk, get written commitment on cross-platform support timelines before signing a multi-year deal.
- The data floor is a hard gate, not a guideline. Below ~20K historical tickets the per-customer models are weak. Guard: confirm your ticket volume and history quality in the pilot; insist on a measured resolution-rate threshold before full rollout.
- Enterprise pricing with no transparent floor. The bill is volume-driven and quoted, so it is easy to under-budget. Guard: model peak-month ticket volume, not average, and cap per-resolution exposure in the contract.
- AI summaries and QA still need oversight. Automated Agent QA scores 100% of tickets but can mis-score edge cases. Guard: calibrate QA against a human-graded sample quarterly before tying it to agent performance reviews.
For pay-per-resolution support AI without a volume floor see Intercom and its Fin agent; for the CS-platform layer that consumes this ticket signal see Gainsight, Totango, and ChurnZero.