Forethought

Most support AI is sold as one thing: a bot that deflects tickets. Forethought's structure is more interesting, because it splits the job into four agents that operate at different points in a support organisation — and only one of them talks to customers.

That split is worth understanding before any demo, because it determines where the value actually lands. Deflection is the headline; the knowledge-gap analysis and the automated quality scoring are the parts that change how a support team runs.

The other thing to weigh in 2026 is ownership: Forethought has been acquired by Zendesk. That changes the buying question from 'is this the best point solution?' to 'where does this sit in the Zendesk roadmap?', and it is the first thing to raise with a salesperson.

Ownership And Claims, Checked

Last checked 8 September 2026. Read from forethought.ai during this check, with the screenshot above captured at the same time. The acquisition is stated on the company's own site.

• Acquisition by Zendesk, announced on Forethought's own site
• Prior venture funding of $17 million, as reported on the site
• The named agents — Discover, Solve, Triage, QA and the Agentic AI Copilot — and what each does
• Solve channel coverage across chat, email, voice, Slack and headless API deployment
• QA described as scoring 100% of human agent tickets
• Named customers including Acorns, Grammarly, Fetch Rewards, Upwork, Redis, UPS, Corsair, WordPress, Cohere and Scale AI
• The 15x ROI, 55% first-response reduction and up-to-98% resolution figures are Forethought's own published claims and are described that way throughout this page
• Solutions, platform, company, resources, compliance and careers sections present in the site footer
• No plan prices published here, in line with this site's pricing policy

We have not independently verified the ROI, response-time or resolution figures. The observations about acquired point solutions describe a general pattern in enterprise software and are not a statement about Zendesk's intentions for this product.

Forethought, And The Zendesk Question

• Acquired by Zendesk. If you are evaluating it standalone in 2026, that is the first fact to weigh, not the feature list.
• Four separate agents, not one bot: Discover (knowledge gaps), Solve (answering), Triage (routing), QA (scoring human tickets).
• QA is the underrated one — it scores 100% of human-handled tickets, which no support team does manually.
• Omnichannel Solve covers chat, email, voice and a headless API, so it is not a chat widget.
• Headline metrics are the vendor's own. 15x ROI, 55% faster first response, up to 98% resolution — treat as marketing, not measurement.

Four Agents, Four Different Jobs

Naming them separately is not marketing tidiness — they solve genuinely different problems and you may only want two of them.

Deflection is what gets bought. Full-coverage QA is what actually changes how a support team is run.

Solve, The One Customers See

The omnichannel agent: chat, email, voice, and a headless deployment for teams building their own front end. It learns from past tickets and help-centre content rather than a hand-built decision tree.

The headless option is the detail worth noting. Most support AI assumes you want its widget; an API-only deployment means the AI can sit inside a product experience you already control.

Triage, The Quiet Saver

Classification and routing. Unglamorous, and often where the fastest real gain is — a misrouted ticket costs a full handling cycle before anyone starts helping, and large teams misroute constantly.

It also degrades gracefully. A routing suggestion that is wrong is a minor annoyance; a customer-facing answer that is wrong is a complaint.

Discover And QA, The Ones Aimed Inward

Discover reads support interactions to surface knowledge gaps — the questions customers keep asking that your documentation does not answer. That is a content backlog most teams know exists and never quantify.

QA scores 100% of human agent tickets. Traditional support QA samples perhaps 2% of tickets, because a human has to read each one. Moving from a 2% sample to full coverage is a different kind of visibility, not just more of the same — and it is the feature most likely to change management behaviour rather than cost.

Reading The Numbers On The Page

Forethought publishes 15x average ROI, a 55% average reduction in first response time, and up to 98% resolution rate. All three are the company's own figures, and this page repeats them as such rather than as findings.

"Up to" is doing heavy lifting in the resolution number. A 98% resolution rate on a narrow band of password resets and order-status queries is entirely plausible and tells you nothing about your own ticket mix. The only number that matters is what it resolves on your tickets, which is what a pilot measures and a case study does not.

ROI multiples in this category are also usually computed against fully-loaded agent cost at current volume, which quietly assumes headcount actually falls. Many teams instead absorb growth without hiring — a real benefit, but a different one from the arithmetic implies.

The customer list is more informative than the metrics. Acorns, Grammarly, Upwork, UPS, WordPress, Cohere and Scale AI are named — a spread that suggests it holds up at genuine volume.

What The Zendesk Acquisition Means For A Buyer

This is the part a feature comparison will miss, and it cuts both ways.

On the reassuring side, an acquired product has a large company behind its uptime, security posture and compliance paperwork. For an enterprise buyer that removes a real category of risk — the vendor disappearing mid-contract.

On the cautious side, acquired point solutions tend to be absorbed. Capabilities get folded into the parent's own suite, standalone roadmaps slow, and integrations with the acquirer's competitors rarely improve. If you run a helpdesk that is not Zendesk, ask directly how long non-Zendesk integrations will be supported and get the answer in the contract rather than on a call.

None of that means avoid it. It means the diligence question has changed, and anyone comparing Forethought against independent vendors on features alone is comparing the wrong thing.

Running A Pilot That Tells You Something

Support AI demos beautifully and disappoints specifically. The gap is always in your own ticket distribution.

1. Pull your last thousand tickets and categorise them by hand before any demo, so you know what a realistic deflection ceiling looks like.
2. Ask what resolution rate looks like on your worst category, not your best. The average hides everything.
3. Start with Triage rather than Solve. Routing errors are cheap to be wrong about; customer answers are not.
4. Run Discover early and treat its output as a documentation backlog — that value arrives whether or not you deploy the customer-facing agent.
5. Turn QA on across everything before you change any process, so you have a baseline to compare against.
6. Ask explicitly about the Zendesk roadmap and non-Zendesk helpdesk support, and get it written down.
7. Measure your own resolution rate at 30 and 90 days rather than trusting the published figure.

Agents, Channels And Ownership