Coveo

What it is

Coveo is a permission-aware search and retrieval layer you point at your own content — a crawler and index, a relevance and ML layer on top, and generative answering above that. The company (TSX: CVO, Quebec City) sells it as the AI-Relevance Platform in four deployment shapes: Commerce, Service, Website and Workplace. The closest mental model is Algolia’s customer-facing search economics welded to Glean’s permission model, which is exactly why it comes up when Glean’s per-seat license stops making sense.

The permission mechanism is the part worth understanding before anything else. Coveo uses early binding: the crawler pulls each item’s access control list at the same moment it pulls the content, and the security identities allowed or denied are stored in the index alongside the document. Queries are filtered against those identities before results are ranked, rather than fetched and then checked. A security identity provider and cache keep group membership resolvable at query time.

On 10 February 2026 Coveo shipped a hosted MCP server, generally available, exposing Search, Fetch, Answer and Passage Retrieval as tools to ChatGPT Enterprise, Claude and other models, with a Coveo app listed in the ChatGPT Enterprise Apps and Connectors directory. Queries through it bill against the existing subscription entitlement rather than a new line item. Coveo was named a Leader in the 2026 Gartner Magic Quadrant for Search and Product Discovery on 23 June 2026.

Financially it is a real, boring public company: Q1 FY2027 (quarter ended 30 June 2026, reported 30 July 2026) revenue of $38.5M, up 8% year over year, adjusted EBITDA of positive $0.1M, operating cash flow of $9.2M, and $107.1M in cash with no debt. FY2027 guidance is $160M-$164M total revenue.

Why it shows up in ops stacks

Because it is the pricing-shape answer to the Glean question, not a feature answer. Glean bills roughly $40-50 per user per month with a 100-seat floor, which prices employees. It does not price the 200,000 anonymous visitors hitting your support portal, because they are not seats. Coveo’s Service and Websites plans are entitlement-based — the unit is 100,000 queries per month — so a logged-out visitor costs a query, not a license.

The second reason is that one index can serve three audiences. Public docs answer anonymously, entitled knowledge-base articles answer only for the customer whose contract covers them, and internal runbooks answer only for the agent in the console — all from the same crawl, because the ACL travelled with the item at index time. Building that with a per-seat internal assistant plus a separate public search tool means two indexes, two relevance models and two sets of content freshness bugs.

Coveo’s own numbers say buyers are moving this way: in Q3 FY2026 (quarter ended 31 December 2025, reported 29 January 2026), SaaS subscription revenue was $36.6M, up 13%, with the core Platform at $35.8M, up 15%; the company reported its highest new business bookings ever, with generative AI solutions driving more than 25% of new bookings and Commerce close to half.

A use case worth buying it for

A support org running a self-service portal with mixed anonymous and entitled content, where the deflection target is real and the content already lives in Salesforce, ServiceNow, Zendesk and SharePoint with permissions already modelled there. Crawl once, serve the public docs to anonymous traffic, gate the entitled articles on the customer’s contract identity, keep agent-only runbooks agent-only, and then point the MCP server at that same index so an internal Claude or ChatGPT Enterprise rollout answers from the same content without a second retrieval pipeline.

On deflection: Coveo’s early generative-answering customers reported case deflection improvements above 20% in A/B tests. That is vendor-reported from earnings commentary, so treat it as the number you will be quoted in the sales cycle, not the number you will get.

Pricing

Coveo publishes no list price. The pricing page describes entitlement-based plans for Service and Websites, seat-based plans for Workplace, and modular pricing for Commerce. The base unit is 100,000 queries per month. Relevance Generative Answering meters Generative Queries per Month on a separate counter, and the Passage Retrieval API meters Passage Queries per Month on a third. A free trial exists; a quote still requires sales.

Third-party buyer data (Vendr) puts the average annual contract around $34,700, with deployments spanning roughly $30K a year for a single-module rollout to well past $500K for multi-module enterprise estates with high query volume. Budget $35K-$60K a year for one module, six figures once Commerce and generative answering are both live.

The meter to model before signing is the split between counters. A portal doing 1M searches a month where 10% escalate to a generative answer is ten search units plus one generative unit — not eleven of the same thing. For contrast, Algolia publishes about $0.50 per 1,000 search requests on Grow, rising to roughly $0.75 per 1,000 with AI features, and around $50K a year for its top annual-commit tier. That is the floor Coveo has to beat on plain site search.

Best for

Support, self-service and digital-commerce leaders at companies where content is already ACL’d in a system of record and the audience is a mix of anonymous visitors, entitled customers and internal agents. The economics work above roughly $100M revenue, or anywhere a 20% deflection gain on ticket volume clears a $35K-plus annual floor.

Do not buy it as an internal employee assistant alone. If the only job is letting 300 employees chat with company knowledge, Glean or Microsoft 365 Copilot is a shorter path and Coveo’s Workplace seat pricing gives up the advantage that made you look at it. Do not buy it for a 40-page docs site either — the implementation cost dwarfs the problem.

Alternatives, and when to pick them instead

• Glean — the installed-base leader for internal enterprise search and the comparator in most Coveo evaluations. Pick it when the audience is entirely employees, the seat count clears 100, and out-of-box assistant UX matters more than unit economics. It loses the moment a meaningful share of your traffic is anonymous.
• Algolia — the market-share incumbent for customer-facing site and product search. Pick it when the job is search and discovery with no ACL complexity: published pricing, faster to ship, materially cheaper at low query volume. Pick Coveo instead the moment results have to differ by who is asking.
• Onyx — the fastest-growing entrant in the segment, open source and self-hostable. Pick it when data residency or cost control requires running the stack yourself and you have an engineer to own it. You trade the managed connector fleet and the relevance tuning for control.
• Inkeep or eesel AI — pick either when the scope is documentation and support answers only, the content is public, and you want it live in days. Neither is trying to solve the mixed-audience permission problem.

Watch-outs

• Early binding makes permission freshness an indexing problem, not a query problem. A revoked account stays effective until the security identity cache and the source refresh catch up. Guard: set the identity provider refresh cadence to match your offboarding SLA, and test with a genuinely revoked account before go-live rather than a test user you disabled in the portal.
• Two meters bill against the same page view. Teams size the search entitlement, ship generative answering to every query, and discover the generative counter is the expensive one. Guard: model both counters against real portal traffic during the pilot, and gate generative answering to intent-qualified queries instead of firing it on every search.
• Implementation is a project with a partner, not a switch you flip. Connector setup, permission mapping and relevance tuning run weeks, and much of the delivery goes through implementation partners. Guard: get the SOW, the named partner and the go-live date inside the same contract as the license, with the first renewal dated from go-live rather than signature.
• Growth is decelerating and margins sit at breakeven. SaaS subscription growth of 13% in Q3 FY2026 gave way to 8% total revenue growth in Q1 FY2027, with adjusted EBITDA at $0.1M. Guard: the balance sheet answers the solvency question — $107.1M cash, no debt — so treat this as roadmap-pace risk, not vendor risk, and negotiate a multi-year term with price protection rather than betting on a single-year quote.