Forethought
Also known as: Forethought AI Agents by Zendesk, Forethought Technologies, Autoflows, Solve, Triage, Assist, Discover, Browser Agent, Agatha Answers
Multi agent customer support platform whose Solve, Triage, Assist, and Discover agents resolve tickets end to end using Autoflows, natural language business logic instead of decision trees.
Forethought is an artificial intelligence customer support platform built around a multi agent system, not a single chatbot. Founded in 2017 by Deon Nicholas and Sami Ghoche, and winner of the 2018 TechCrunch Disrupt Battlefield, it serves enterprises including Grammarly, Airtable, Datadog, Upwork and UPS. Zendesk completed its acquisition of Forethought on 26 March 2026. The agents now sell as Forethought AI Agents by Zendesk, listed beside Zendesk's own plans as Zendesk + Forethought, and they work inside Zendesk and on any other service platform, so a customer on a different helpdesk can still deploy them.
The platform is organized as a team of specialized agents that hand off to each other. Discover analyzes a brand's historical support data to surface knowledge gaps and generate both help center articles and Autoflows. Solve resolves customer inquiries end to end across chat, email, voice, SMS, Slack and in app. Triage classifies, prioritizes and routes what needs a person by intent, sentiment, language and urgency. Assist works as a copilot inside the helpdesk, summarizing tickets and drafting complete responses using the same Autoflows.
Autoflows is the reasoning engine underneath all of it, and the thing that makes the platform agentic: teams describe business logic in plain natural language instead of building decision trees, and the agent reasons through those policies and takes action. Forethought connects to more than seventy systems spanning helpdesks, CRM, knowledge bases, contact centers and API management platforms, and it trains models individually on each customer's data, not one shared generic model.
The capability worth understanding before a demo is Browser Agent. Where no API exists, it logs in, clicks and types its way through the same browser workflows a human agent would run, inside policies approved in advance with encrypted credentials and human in the loop checkpoints. Forethought pitches it at the share of contact center work that integrations simply cannot reach.
On security, Forethought is independently audited with annual SOC 2 Type II reports, reports ISO 27001 compliance, aligns its controls to NIST 800-53, encrypts data at rest and in transit, offers a DPA covering GDPR and CCPA, runs regular third party penetration tests and has appointed a Data Protection Officer. It is not FedRAMP certified. Two gaps matter. The platform runs on AWS with no published region choice, self hosted or single tenant option, and there is no public API or SDK documentation for the platform itself, so extending it means going through Forethought. Pricing is quoted, not published, and the platform is built for high ticket volumes.
Vendor details
Canonical URL
https://forethought.ai
Category
Customer support agent
Company status
acquired
Use cases & customers
In practice
A SaaS company drowns in repetitive tickets. Forethought's Solve agent fully resolves common issues like password resets and order status across every channel, cutting deflection load by half so human agents focus on complex, high value conversations.
A global brand needs every ticket routed correctly. Forethought's Triage agent classifies and prioritizes each ticket by intent, sentiment, and urgency, then routes by product and region using a custom model, with no manual sorting.
A healthcare team handles patient questions that touch protected health information. Forethought redacts sensitive data at ingestion, routes those cases to humans with full audit trails, and uses its agent only for non sensitive intents like appointment reminders.
Agentic Index coverage score
9.5 / 14 capabilities · 68%
| Integrations & Tool Calling | Full |
|---|---|
|
Forethought connects to helpdesks, knowledge bases, CRM and API management platforms, with more than seventy integrations spanning contact centers and connector platforms including Airtable and Snowflake, and Autoflows reason through business policies to take authenticated action, not hand a suggestion back. Where no API exists at all, Browser Agent completes the workflow directly, so the action surface is not bounded by what partners expose. Sourceforethought.ai/platformread 2026-09-05 |
|
| Workflow Orchestration | Full |
|
Autoflows is Forethought's agentic reasoning engine. It takes business logic in natural language, not a decision tree, and reasons through complex policies to resolve an issue end to end. Four agents hand off to each other in a set sequence. Discover generates the Autoflows, Solve detects intent and resolves, Triage tags and routes what needs a person, and Assist surfaces the right Autoflow to the human agent who picks it up. Browser Agent executes the steps no API can reach. Sourceforethought.ai/platformread 2026-09-05 |
|
| Knowledge Grounding & RAG | Partial |
|
Forethought connects to a brand's helpdesk, knowledge base and CRM and learns from its historical support data from day one, with Discover generating Autoflow policies and help center articles from that corpus and Solve answering with the right Autoflow or knowledge base article. There is no stated index, embeddings layer, retrieval architecture, refresh behavior or source-level citation, so a maintained retrieval structure is claimed but not shown. The per-customer fine-tuned models are training, not retrieval. Sourceforethought.ai/platformread 2026-09-05 |
|
| Human Oversight & Guardrails | Full |
|
The gate before an action sits in Forethought's own product, in two forms. Browser Agent runs only inside pre-approved policies with encrypted credentials and human-in-the-loop checkpoints, with approval steps configurable per workflow, so the customer decides which browser actions the agent may take unattended. Alongside that sit configurable escalation paths and an approval workflow in which Assist drafts a response and a person approves before it is sent. Sourceforethought.ai/webinars/ai-that-clicks-for-you-automate-76-percent-of-cx-work-apis-cant-reachread 2026-09-05 |
|
| Security, Identity & Governance | Full |
|
The platform is independently audited and certified for security, availability and confidentiality, with annual SOC 2 Type II audits, controls aligned to NIST 800-53, HIPAA, GDPR and CCPA, encryption at rest and TLS in transit, production secrets managed in AWS tooling, a DPA reflecting GDPR and CCPA, security-design reviews, threat models, regular third-party penetration tests, a named Data Protection Officer, a HackerOne bug bounty, a trust center at trust.forethought.ai, and a fully published subprocessor table naming each processor's identity, purpose and location. Automated redaction removes PII, PHI and financial data at ingestion with the original securely deleted within 24 hours. Forethought reports ISO 27001 compliance and SOC 2 certification, and it is not FedRAMP certified. Role based access is claimed on the platform page but not repeated on the security page or in the legal documents. Sourceforethought.ai/platform/securityread 2026-09-05 |
|
| Observability & Auditability | Partial |
|
Discover surfaces insights across all support interactions and detects where the agent is failing, Triage gives CX leaders real-time visibility into agent performance and coaching opportunities, and audit trails and automatic redaction of personal, health and financial data at ingestion are in place. This reports on outcomes and content gaps, not on the run. There is no stated step-level decision trace, tool-call log, Autoflow execution record or retrieval log, which matters more on a product whose Browser Agent clicks and types inside third-party systems. Sourceforethought.ai/platform/triageread 2026-09-05 |
|
| Memory & State Persistence | Not documented |
|
There is no stated memory layer with a scope or lifetime. The platform is trained on a brand's historical support data from day one to generate Autoflow policies and surface knowledge gaps, which is training and retrieval over a corpus, not state the agent retains between sessions. The published agent set names Solve, Triage, Assist and Discover with no memory component among them. Sourceforethought.ai/platformread 2026-09-05 |
|
| Deployment & Data Residency | Not documented |
|
Every core processor that hosts customer data sits in the United States. Amazon Web Services, MongoDB, Microsoft Hosting Services and Google are all listed as web hosting in the United States, with Cloudflare, OpenAI, Groq, Mistral and Pusher likewise, and only Datadog for logging and Snowflake for warehousing add Germany. The subprocessor table names each location. There is no stated region menu, residency commitment, VPC, single-tenant or self-hosted option, and the security page, which covers compliance, encryption, product design, training, oversight, PII handling, FedRAMP scope and HIPAA in detail, does not address residency. A customer with an EU or UK residency requirement can see the answer and cannot change it. Sourceforethought.ai/data-subprocessorsread 2026-09-05 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Prebuilt agents come as a named set the customer adopts as units, each with its own product page. They are Solve for end-to-end customer resolution, Triage for classification and routing, Assist as an in-helpdesk copilot, Discover for gap detection and content generation, plus Browser Agent and Voice. Discover also generates Autoflows from a brand's own history, so the customer starts with populated workflows, not a blank builder. Sourceforethought.ai/platformread 2026-09-05 |
|
| Triggers & Channel Coverage | Full |
|
Work reaches the agent from any channel. Solve resolves across chat, email, voice, SMS, Slack and in-app, with email answered instantly, phone calls handled end to end, embedded chat, and internal Slack support, and Triage picks up tickets arriving in the helpdesk queue and routes them onward. Sourceforethought.ai/platformread 2026-09-05 |
|
| Model Flexibility & Routing | Partial |
|
OpenAI, Groq and Mistral are listed as subprocessors under AI Models, alongside Gladia and Deepgram for speech recognition and ElevenLabs and Cartesia for text to speech on the voice product. The platform runs across several providers, not one, and its per-customer fine-tuned models sit on top of that mix. Cloudflare, AWS, MongoDB, Microsoft and Google on the same table are infrastructure, not model providers. The routing is vendor-internal and no model selection, override or bring-your-own-model control is exposed to the customer. Sourceforethought.ai/data-subprocessorsread 2026-09-05 |
|
| APIs, SDKs & MCP Extensibility | Partial |
|
A Headless channel ships as a named product surface for embedding the agent into a customer's own systems, alongside a mobile SDK for in-app chat, an Assist browser extension delivered into the helpdesk, and connector integrations reaching API management platforms, with code and no-code deployment paths. There is no published API reference, SDK documentation site or MCP server for Forethought's own platform, so the headless surface is named as a capability but not documented as one. Sourceforethought.ai/channels/headlessread 2026-09-05 |
|
| Testing, Debugging & Optimization | Partial |
|
The Discover agent runs an improvement loop after deployment. It analyzes a brand's support data, surfaces knowledge gaps, and generates both help center content and new Autoflows to close them, so live performance feeds back into what the agent can resolve. The QA Agent, a named product at forethought.ai/platform/agent-qa, scores 100 percent of human agent tickets, not the AI agent's own behavior, and there is no stated simulation environment, scored test suite or evaluation harness for the customer's own agent. Sourceforethought.ai/platform/agent-qaread 2026-09-05 |
|
| Browser & Computer Use | Full |
|
Browser Agent drives a real interface itself. It logs in, clicks, types and completes the same browser-based workflows human agents run today, using pre-approved policies and encrypted credentials, and Forethought aims it at the work that APIs cannot reach. Sourceforethought.ai/webinars/ai-that-clicks-for-you-automate-76-percent-of-cx-work-apis-cant-reachread 2026-09-05 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Pricing
Not published. Quoted through sales, by Forethought or by Zendesk as Zendesk + Forethought.
Free / trial
None published. A Proof of Value is offered instead of a free trial.
Lowest paid plan
Not published
Support SLA / resale
Support comes in three levels, Standard support, Ask the expert and Advanced support. No SLA is published.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…
Alternatives to Forethought
The closest documented capability profiles to Forethought among customer support agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Crescendo8.0 / 14Fuller documented coverage on Testing, Debugging & OptimizationForethought vs Crescendo →
- Gradient Labs10.0 / 14Adds documented Memory & State Persistence
- LivePerson10.0 / 14Fuller documented coverage on Model Flexibility & Routing and APIs, SDKs & MCP Extensibility
- Assembled8.5 / 14Fuller documented coverage on Observability & Auditability and APIs, SDKs & MCP ExtensibilityForethought vs Assembled →
- Fini8.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Observability & Auditability
- Parahelp8.5 / 14Fuller documented coverage on Knowledge Grounding & RAG and Testing, Debugging & Optimization
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded