The Search Box Is Becoming the Company Brain

The most revealing thing about Coveo is that its oldest idea has become fashionable again. In 2005, three search engineers spun a project out of Copernic with a plain mission: make all of a company's information searchable in less than a second, wherever it lived. Twenty-one years later, enterprise software is crowded with agents promising to act, answer and decide. Each still runs into the same dull obstacle. Before intelligence can do anything useful, it has to find the right fact.

Coveo lives in that moment before the answer. Its software connects to product catalogs, help centers, websites, customer records and workplace repositories. It builds a unified index, respects the source system's permissions, watches what people click or buy, and ranks what should appear next. The output might be ten search results, one recommended replacement part, a useful article beside a support case or a generated paragraph with citations. The company calls the common machinery underneath all of them an AI-Relevance platform.

The founding trio knew search from an earlier technological age. Laurent Simoneau had led product and technology at Copernic; Richard Tessier and Marc Sanfaçon came out of the same orbit. Simoneau, Tessier and Sanfaçon separated the enterprise project into Coveo, initially selling software that ran inside a customer's own environment. Louis Têtu and Jean Lavigueur joined early and helped turn a technically credible product into a commercial company. Têtu, who had co-founded talent-software maker Taleo, later served as Coveo's chief executive from 2011 to 2025. Simoneau returned to the CEO role in 2025, with Têtu moving to executive chairman and Sanfaçon eventually becoming CTO.

The essential strategic move arrived around 2012, when Coveo shifted from on-premises software toward multi-tenant cloud SaaS. That changed more than deployment. A cloud service could improve models continuously, absorb interactions across large deployments and ship new relevance capabilities without waiting for each customer to install an upgrade. Coveo later bought the small commerce-search company Tooso in 2019 and Qubit, a London personalization business, in 2021. The acquisitions pulled the company toward digital commerce, where search quality has an unusually visible price tag.

A search engine with four front doors

Coveo sells into four overlapping jobs. In commerce, it powers product search, category listings, recommendations and merchandising. In customer service, it helps people solve problems themselves and puts knowledge beside agents inside their workflow. On websites, it connects visitors with pages, documents and generated answers. In the workplace, it searches across repositories while showing each employee only what that employee is allowed to see.

The underlying customer problem is wonderfully unglamorous. Enterprises have too much information in too many systems, written in the vocabulary of whoever stored it. A shopper describes a symptom while a catalog expects a part number. A customer types a question that the knowledge base phrases differently. A new employee remembers the policy but not which portal owns it. Basic keyword search treats these as mismatches. Coveo combines lexical and semantic retrieval with behavioral models that learn from queries, clicks, carts and resolutions.

That loop is also the product's practical value. A retailer can promote inventory without manually arranging every result. A manufacturer can let a distributor search a vast catalog using ordinary language. A support team can surface the right procedure before an agent starts asking colleagues. Developers get APIs, a Headless library and Atomic components rather than a fixed storefront. Business users get tuning tools and analytics instead of a permanent ticket in the engineering queue.

Its customer list reflects the range. Nespresso, Linde, FleetPride and Caleres represent the commerce side. Salesforce and Tableau helped establish the service and self-service case. Recent wins or expansions named by the company include Palo Alto Networks, Intuit, Deloitte, Nestlé, Dow, Arm, Cummins, Docusign, Okta and athenahealth. These names do not mean every customer buys every product. They show how Coveo enters through a specific problem, then tries to reuse the same index and models elsewhere. A knowledge deployment can grow into commerce; a website-search installation can move into service.

AI without context simply does not work.Louis Têtu · Executive Chairman

Generative AI made retrieval strategic again

A language model can write a polished wrong answer faster than an old search engine can return a disappointing page. Coveo's Relevance Generative Answering product is built around that risk. It retrieves from content selected by the enterprise, enforces permissions, finds the most relevant passages, gives those passages to a model and returns a cited answer beside conventional results. If a user changes a filter, the answer can regenerate from the narrowed material.

This is less magical than a chatbot demo, and much more useful. The model does not need to memorize a company's changing warranty rules, product specifications or support articles. Coveo keeps the index fresh and supplies the context at query time. Semantic encoders help recover material that shares meaning without sharing exact words. Ranking models and analytics help decide what deserves attention. Citations let the user inspect the source material rather than admire the machine's confidence.

The same logic now extends to agents. In 2026 Coveo announced a hosted Model Context Protocol server, Search Agents, Conversational Product Discovery and a Merchandising Copilot. The pattern is consistent: Coveo does not have to own the whole assistant. It can be the permission-aware retrieval service behind an assistant, or the decision layer that helps one choose products and content. By July, Search Agents were live with five customers, with dozens more evaluating them.

Subscriptions, integrations and the commerce pull

Coveo is enterprise SaaS. Customers buy annual or multi-year subscriptions to a cloud platform and may add paid capabilities such as generative answering. Professional services contribute a smaller amount. The sale is consultative because usage, data volume, security, integrations and business case differ sharply between a public website and a global B2B catalog. Partners matter: SAP, Salesforce, Shopify, Adobe, Sitecore and Optimizely put Coveo close to systems enterprises already run.

Commerce has become the sharp end of the strategy. It produced nearly 60 percent of new business bookings in fiscal 2026, with management pointing to complex B2B deployments in manufacturing and distribution. This is a good market for relevance software because the failure is measurable. Bad product discovery abandons revenue; better ranking can lift conversion, average order value and catalog coverage. Unlike a general workplace query, a product query often ends in a cart or it does not.

B2B commerce is especially awkward and therefore attractive. A catalog may contain millions of products, near-identical variants and region-specific inventory. A buyer may be entitled to see one assortment and one negotiated price while another buyer sees something else. Queries include partial SKUs, obsolete part numbers and descriptions of jobs rather than products. Coveo's pitch is that intent-aware ranking and semantic retrieval can handle that complexity while its Merchandising Hub still lets a human promote a product, pin a result or respond to commercial priorities. Automation gets the first move; the merchant keeps a hand on the wheel.

The financial picture is sturdy but not frictionless. Fiscal 2026 revenue rose 11 percent to US$148.3 million, while SaaS subscription revenue rose 13 percent to US$142.5 million. Gross margin was 78 percent. The company still posted a US$28.9 million net loss, influenced by operating costs and other items, while generating US$10.5 million in operating cash flow. In the following quarter, total revenue reached US$38.5 million and adjusted EBITDA was slightly positive.

There is a telling cleanup inside those numbers. Coveo bought personalization company Qubit in 2021, then chose to retire the legacy Qubit platform and focus investment on its core. By fiscal 2026 that deprecation was complete. The narrower story is easier to understand: one relevance platform, extended across more experiences. After the June 2026 quarter, Coveo said a Fortune Global 500 technology customer had expanded to eight figures in annualized subscription spending across internal and customer-facing AI applications.

Not every search problem is the same market

Coveo overlaps with Algolia, Constructor and Bloomreach in product discovery; with Elastic and Lucidworks in configurable search; and with Glean, Microsoft and Sinequa in enterprise knowledge. Google and major application platforms can supply their own retrieval. A company that wants a fast developer search API may not need Coveo's breadth. A company that wants to build every ranking model and interface itself may prefer open infrastructure.

Coveo's differentiation is the combination. It offers unified indexing, permissions, semantic and behavioral ranking, recommendations, analytics, generative answers and business-user controls across several experiences. A retailer can begin with product search; a software company can begin with support knowledge. If the platform works, both can expand. The disadvantage is the mirror image: broad enterprise platforms require integration, governance, sufficient traffic and an economic case. Relevance improves with good content and useful behavioral signals. Software cannot rescue a catalog nobody maintains.

The build-versus-buy decision is not philosophical. An internal team can assemble OpenSearch or Elasticsearch, a vector database, a reranker, analytics, access controls and a model endpoint. It can also spend years reconciling permissions, tuning relevance and giving merchandisers tools they can use without code. Coveo packages those chores and charges for avoiding the assembly. That is compelling where search touches large revenue or support budgets. It is less obvious for a small catalog, a lightly used intranet or a team whose needs stop at autocomplete. Coveo's natural territory is the complicated middle of a large enterprise, where several repositories and business rules must behave like one system.

A quiet layer beneath visible intelligence

Coveo sits between systems of record and systems of experience. It is not the CMS holding an article, the commerce platform holding a price, the CRM holding a case or the language model drafting an answer. It connects those worlds and decides what information should cross the gap for this person, in this moment, with these permissions.

That position explains the partnership with Bell on sovereign Canadian AI, where domestic infrastructure and controlled data matter. It explains the long Salesforce and SAP relationships, where Coveo adds a relevance layer without asking customers to replace their operational software. It also explains why the company keeps talking about outcomes. Retrieval is invisible when it works. Its value appears somewhere else: a completed purchase, a deflected case, a quicker repair, a new employee who does not have to message five people.

The partnership network is distribution as much as integration. An SAP account can encounter Coveo while rebuilding commerce; a Salesforce customer can add knowledge retrieval inside the agent console; a Shopify merchant can install product discovery from an app marketplace. Bell's AI Fabric offers a route into Canadian government and regulated organizations that care about where information is processed. These alliances also create dependency. Large platforms continue to improve their own search and AI features, so Coveo must remain useful enough to justify another strategic layer in the stack.

The amusing twist is that the future arrived looking like a better search box. Coveo's founders were trying to tame folders and enterprise networks. Today's interfaces talk in full sentences and promise autonomy, but they remain hungry for current, trusted context. The company has spent two decades building the plumbing for that context. Whether Coveo becomes a major agent-era layer will depend on execution, competitive pricing and how widely customers expand beyond a first use case. The architectural bet, at least, has aged well.