Verdict
Forethought fits mid-market and enterprise teams with large ticket histories that want one vendor for resolution, triage, agent assist and QA, on Zendesk or another help desk. The catch is quote-only pricing, a high data bar to qualify, and a roadmap now set by Zendesk 1,7,8.
Good fit if you are
- Support teams with 20,000+ historical tickets and steady volume
- Zendesk customers who want an AI add-on with voice
- Teams on Salesforce, Intercom or other help desks that want a full AI suite
Look elsewhere if you are
- Small teams handling fewer than 2,000 tickets a month
- Buyers who need public, per-unit prices
- Organisations that require EU data residency or an ISO 27001 certificate
Pros and cons
What we like
- Takes real actions through API endpoints, plus a Browser Agent for tools without APIs 6.
- Runs on chat, email, voice, SMS, Slack and a headless API in 100+ languages 6.
- Connects to 70+ tools, including Zendesk, Salesforce, Intercom, Kustomer and Five9 9.
- One suite covers resolution, triage, agent assist, insights and QA 5.
- Still sold standalone, with no requirement to use Zendesk 2,3.
What to watch
- Every plan is quote-only; no prices are published 7.
- Needs 20,000+ historical tickets and 2,000+ tickets a month to get started 8.
- Pooled model training on de-identified customer data is on by default; you must opt out 11.
- Most teams take 30 to 90 days to go fully live 8.
- Email, voice and the Assist copilot need the Professional plan or higher 7.
Key facts
| Made by | Zendesk |
|---|---|
| Pricing | Quote only |
| Cost for ~1,000 conversations | Not published (quote) |
| Also requires | Needs 20,000+ historical tickets and at least 2,000 email or chat tickets a month 8 |
| Channels | Chat, Email, Voice, SMS |
| Languages | 100+ languages 6 |
| Works with | Zendesk, Salesforce, Intercom, Freshworks, Kustomer, ServiceNow, HubSpot, Gorgias, Gladly, Front, Help Scout, Shopify, Genesys, Five9, Amazon Connect |
| Security reports | SOC 2 Type II |
| Free trial | No free trial; Proof of Value instead [7] |
What Forethought AI Agents is
Forethought builds a set of AI agents for customer support: Solve resolves customer requests, Triage classifies and routes tickets, Assist drafts replies for human agents, Discover surfaces insights, and a QA agent scores interactions 5. Founded in 2017, the company raised $117 million before Zendesk announced the acquisition on March 11, 2026 and closed it on March 26, 2026 16,1,2.
Zendesk now sells the product as Forethought AI agents by Zendesk. Zendesk customers buy it as an add-on, and companies on other platforms can still buy it standalone, with no requirement to use Zendesk 2,3. Co-founder Sami Ghoche remains CEO 21.
What it can do
Forethought trains on a company's past tickets, so it needs at least 20,000 historical tickets and 2,000 or more email or chat tickets a month 8. Autoflows turn plain-language policies into agentic workflows, Custom Actions call APIs to process refunds or update accounts, and a Browser Agent logs in and clicks through legacy tools that have no API 6. It runs on chat, email, voice, SMS, Slack, mobile and a headless API in 100+ languages, and connects to 70+ tools, from Zendesk and Salesforce to Five9 and Shopify 6,9.
Teams set confidence thresholds, escalation logic and fallback rules, and each step is logged with its source, action, workflow and confidence score 13. Handoffs pass full context to a human, and a dashboard tracks CSAT, deflection and speed 6. Browser Agent workflows run on pre-approved policies with encrypted credentials and optional approval steps 14.
- Actions. Custom Actions connect API endpoints so the agent can process refunds or update accounts mid-conversation, and Autoflows turn policies into agentic workflows 6. A Browser Agent logs in, clicks and types in legacy tools that have no API, using pre-approved policies and optional approval steps 6,14.
- Setup. It trains on a company's historical support tickets from day one and needs 20,000+ of them 8. Most teams are fully set up within 30 to 90 days, then get a Customer Success Manager 8.
- Handoff to people. Hands conversations to human agents in real time with full context, and Professional adds custom handoff models 6,7.
- Analytics. A dashboard tracks CSAT, deflection and speed with suggestions for weak workflows, and the QA agent can evaluate every support interaction against rubrics 5,6,7.
Pricing
Forethought lists three plans, Team, Professional and Enterprise, but none shows a price; every tier is quote-only 7. The bill blends a platform access fee with outcome-based charges, and extra usage is charged when you exceed plan limits 7. Team covers chat and mobile; email, voice and the Assist copilot start at Professional, and the Solve API and Analytics API sit in Enterprise 7. There is no free trial; Forethought runs a Proof of Value on your data instead 7. Zendesk has not published add-on pricing 3.
| Plan | Price | Includes |
|---|---|---|
| Team | Quote7 | Chat and mobile; Solve, Triage and QA agents; Autoflows and custom actions; 2 brands; 3 triage models |
| Professional | Quote7 | Adds email, voice and Slack, the Assist copilot, AI-surfaced insights, up to 20 brands and custom handoff models |
| Enterprise | Quote7 | Adds the Solve API, knowledge gap detection, AI article creation, Analytics API and enterprise governance controls |
Security and data
Forethought has completed a SOC 2 Type II audit and says it aligns with ISO 27001, NIST CSF and HIPAA, with annual SOC 2 and HIPAA audits 10,11. It redacts PII, PHI and financial data on ingestion and deletes the originals within 24 hours 11. By default, de-identified customer content may be combined with other customers' data to train models, and customers must opt out to stop it 11. Hosting subprocessors are US-based, and model providers are OpenAI, Groq and Mistral 12.
- HIPAA. Claims HIPAA compliance with annual HIPAA audits; BAA terms not published 10,11
- Data residency. US hosting; no regional choice published 12
- Training on your data. By default, de-identified customer content may be combined with other customers' data to train models; customers can opt out 11
- AI models. OpenAI, Groq and Mistral 12
What customers say
Forethought holds 4.3 on G2 from about 165 reviews, 4.5 on Capterra from 11 and a single review on Gartner Peer Insights 18,19,20. Reviewers praise the support team and AI capability, but cite delays, cost, a learning curve and answers that are not always accurate 18,19. Named stories with numbers include Grammarly (87% deflection), YNAB (deflection up from 25% to 70%), YAZIO (80% deflection) and Cotopaxi (168% ROI) 15.
On G2's evaluation page Forethought scored 62 out of 100, tenth of 14, in a benchmark run by G2's partner Swept 22; it is still under evaluation on G2's General CX leaderboard 23.
| Platform | Rating | Reviews | Seen |
|---|---|---|---|
| G2 (Forethought)18 | 4.3 / 5 | 165 | 8 October 2026 |
| Capterra (Forethought)19 | 4.5 / 5 | 11 | 8 October 2026 |
| Gartner Peer Insights (Forethought Platform)20 | 5.0 / 5 | 1 | 8 October 2026 |
Independent task evaluations
| Evaluation | Result | Accuracy | Policy | Run by |
|---|---|---|---|---|
| G2 Customer Experience agent evaluations22 | 10 of 14 · 62/100 | 50% | 60% | Swept (G2 partner benchmark) |
| G2 General CX leaderboard (G2 hands-on test, 46 tasks)23 | Under evaluation | — | — | — |
Seen on G2 on 8 October 2026. G2 asks readers not to compare its own tests with partner benchmarks directly.
What Zendesk claims (not independently verified):
- Up to 98% resolution rate 4
- 15x average return on investment 4
- 55% average reduction in first response time 4
- Grammarly: 87% deflection rate and 4.2 CSAT 15
- YNAB: deflection rose from 25% to 70% 15
- YAZIO: 80% ticket deflection without lowering CSAT 15
- Cotopaxi: 168% ROI in six months 15
Concerns raised by others:
How we scored it
| Criterion | Why | Score |
|---|---|---|
| Independent tests 30% of the total | No hands-on test run by G2: a Swept (G2 partner benchmark) result only 22; under evaluation on the G2 General CX leaderboard (G2 hands-on test, 46 tasks) 23. | 2/5 |
| What it can do 20% of the total | Custom actions process refunds and account updates via API, a Browser Agent works in tools without APIs, and it runs on chat, email and voice in 100+ languages 6. | 5/5 |
| Pricing 15% of the total | Plan tiers are listed but every one is quote-only, blending a platform fee with outcome-based charges 7. | 2/5 |
| Control 15% of the total | Confidence thresholds, escalation rules, context-rich handoff and a performance dashboard are documented, but no bulk simulation suite or version control for agent changes is published 6,13. | 3/5 |
| Security and data 10% of the total | SOC 2 Type II is in place, but ISO 27001 is claimed as alignment only, hosting is US-only and pooled model training is on by default 10,11,12. | 3/5 |
| Customer evidence 10% of the total | It holds 4.3 on G2 from about 165 reviews plus named customer stories with numbers, such as Grammarly's 87% deflection 18,15. | 4/5 |
| Total | Weighted sum of the 6 levels, out of 10. | 6.1 |
Sources
- Zendesk Completes Acquisition of Forethought
- Zendesk to acquire Forethought
- Announcing Forethought AI agents by Zendesk for customers
- Forethought homepage
- Multi-Agent System
- Forethought Solve
- Forethought Pricing
- FAQs
- Integrations
- Security & Compliance
- Safeguarding the AI Frontier
- Data Subprocessors
- Why Highly Regulated Industries Need a Smarter AI
- AI That Clicks For You: How to Automate the 76% of CX Work APIs Can't Reach
- Customer Success Stories
- Zendesk announces deal to acquire customer support AI startup Forethought
- Will Zendesk's Forethought Acquisition Enable True Agentic Resolutions?
- Forethought Reviews 2026: Details, Pricing, & Features
- Forethought reviews
- Forethought Platform reviews
- The Next Chapter for Forethought
- Customer Experience Agent Evaluations
- G2 AI CX Agent Evaluations: General CX Leaderboard
