Coveo Acquires Anglera to Enrich Product Attributes from Spec Sheets for AI Search Discovery: How to Read the 270% AI-Referred Traffic Claim
AI search company Coveo has acquired product data enrichment startup Anglera. How it extracts attributes from spec sheets with source citations, what terms were not disclosed, the caveats behind its A/B test, and what ecommerce businesses should check to be found by AI agents.
Key Takeaways
- On October 29, 2026, AI search and product discovery company Coveo acquired Anglera, which extracts and standardizes product attributes from spec sheets, PDFs, and images. Terms of the deal, including the price, were not disclosed
- A company that has competed on search quality is now buying the layer underneath search: the product data itself. AI assistants recommend only what they can understand, so missing attributes translate directly into lost opportunities
- In an A/B test reported by Anglera, AI-referred visits rose 270%. But the test was self-reported and small, covering about 300 SKU variants over 30 days. The design worth noting is that every attribute carries a source and is written back to the system of record
Coveo acquires Anglera, terms undisclosed
Coveo (TSX: CVO), the Experience AI leader, today acquired Anglera, a San Francisco-based, AI-native product data enrichment company.
www.prnewswire.comMontreal-based Coveo (Toronto Stock Exchange: CVO) announced on October 29, 2026 that it had acquired San Francisco-based Anglera. Coveo provides enterprise site search, recommendations, and conversational product discovery, and is also used in B2B ecommerce, including by the Australian wholesaler Metcash.
Anglera is a startup specializing in product data enrichment. Enrichment means filling in attributes missing from product pages, such as dimensions, materials, and compatibility, and making their formatting consistent. According to its Y Combinator company page, Anglera came out of the Summer 2024 batch. Its founders, Amay Aggarwal and Ray Iyer, are Stanford graduates who built AI systems for retail product catalogs at Uber Eats.
The announcement states plainly that the terms of the transaction were not disclosed. None of the price, the form of payment, or Anglera's headcount and revenue was disclosed. Coveo's fiscal 2027 second-quarter results, released the same day, also say nothing about the deal value.
Two points were made about customers. Existing Anglera customers will continue on the Anglera platform. Customers of Coveo's commerce product, Coveo Commerce, will be offered Anglera's enrichment as an add-on. Pricing for the add-on was not disclosed. Anglera's customers include multiple Fortune 500 companies, and it serves customers in the US, Canada, and Europe.
What Anglera does: extracting attributes from spec sheets and citing the source
Its target is complex catalogs where attributes can run to hundreds of fields. For products such as tools, industrial parts, and appliances, where accurate information exists only in manufacturer spec sheets or PDF manuals, ecommerce product data tends to be full of blanks. Anglera's agentic AI reads those documents in place of people and pulls out the values that are needed.
Combining the announcement with Coveo's blog post, the process can be summarized as follows.
- Find values in manufacturer spec sheets, manuals, PDFs, images, manufacturer web pages, and supplier spreadsheets
- Standardize units and formats to each customer's catalog rules
- Validate against those rules, attaching a source and a confidence score to every value
- Send low-confidence values to a person for review, and write confirmed values back to the source system
What stands out is that every value is tied to a source, and a person decides on low-confidence values. When generative AI writes product descriptions, plausible errors creep in. In B2B, a wrong spec leads straight to a wrong part being ordered. Coveo's blog cites typical problems such as a washer tagged as a refrigerator after a PIM (product information management) migration, and one feed reading "Height: 34in" while another reads "H: 34'".
Coveo Chief Product Officer Peter Curran said the company chose Anglera after evaluating it against an internal build and a broad range of enrichment technologies.
Unlike solutions that rely primarily on images or web scraping, Anglera grounds AI-generated product information in multiple authoritative sources.
Co-founder Ray Iyer described the motivation for starting the company: much valuable product knowledge sits in documents that no team can process by hand at scale. The engine reads those sources the way an expert would and ties every attribute back to where it came from. On its YC page, the company says it has cut the time per product from 15 minutes to 5 seconds.
Another design choice is that enriched values are written back to the source system. Rather than fixing data only inside Coveo's search index, it fixes the original data in systems such as the PIM. As a result, the improvements reach not only site search but also product pages, advertising and shopping feeds, and external AI agents.
Why a search company is buying the layer before search
Until now, Coveo has solved problems on the search side, with semantic search that matches on meaning and multimodal search that also matches on images. Coveo's blog explains the decision to acquire anyway as a response to three shifts.
The first is that queries have become longer and more complex. Buyers used to AI assistants describe their conditions in sentences, such as "lightweight running shoes that can handle heavy rain" or "a replacement pump that can fit an xb-2598 press". The second is that buyers now expect search to compare products and answer questions. The third is that product research increasingly starts outside the retailer's own site, in ChatGPT or Gemini.
All three shifts depend on rich product data. Comparing two products requires both records to carry the same attributes. Answering "will this part fit?" requires compatibility information in the record. Coveo writes that external AI shopping agents recommend what they can understand. In other words, no amount of search performance can answer with attributes that are not in the underlying data.
The two companies envision an improvement loop connecting search and enrichment. Coveo's search logs reveal searches that returned no results and attributes that buyers frequently ask about but that are missing from records. Anglera fills in those attributes, and Coveo measures the effect on discovery and conversion. The announcement calls this a continuous improvement cycle between what buyers want and what product catalogs know. This integration is still "planned", however. What is available at the time of the acquisition is writing enriched values back to the source system, which Coveo then indexes.
Coveo's own business is also leaning toward agentic features. In its results the same day, SaaS subscription revenue was $39.1 million, up 9% year over year. For its Merchandising Agent, which tunes storefront merchandising, Coveo disclosed that more than 20% of storefront optimizations are now completed through automated agentic actions. It also has a product called Discovery for Agentic Channels for getting products found inside ChatGPT. Coveo already had an outlet for delivering products to external AI, and it has now acquired a piece that raises the quality of the data flowing through it.
How to read the 120% and 270% figures
The most prominent numbers in the announcement come from an A/B test with an apparel retailer. Product pages with Anglera-enriched data saw a 120% increase in organic landing-page traffic and a 270% increase in AI-referred visits relative to a control group.
These figures rest on assumptions stated in the announcement's footnote.
| Figure | Source | Assumptions and caveats |
|---|---|---|
| Organic landing-page traffic +120% | Reported by Anglera (footnote to the announcement) | 30-day A/B test at an apparel retailer, comparing about 300 enriched SKU variants with a control group |
| AI-referred visits +270% | Reported by Anglera (footnote to the announcement) | Same test. Baseline visits, significance, and the retailer's name not disclosed |
| Missing attributes reduce discoverability by up to 30% | NielsenIQ (August 2025 article) | The supporting link in the article points to a 2021 article on search terms |
| Merchants with product data structured for agents: 15% | PYMNTS Intelligence with Visa Acceptance Solutions | Merchant survey, published June 2026 |
| AI readiness below 60: 77%; 80 or above: under 1% | Channable (analysis of 9,689 companies on its platform) | Channable's own scoring based on Google guidance; analysis by a feed management vendor |
According to the footnote, the test was reported by Anglera, ran for 30 days, and covered about 300 SKU variants. The footnote also says results may vary. The absolute visit counts, statistical significance, and the retailer's identity were not disclosed. If the product group had few visits to begin with, even a small increase produces a large percentage. It is more reasonable to read this as an example of direction than as a figure verified by a third party.
NielsenIQ's estimate that missing attributes can reduce product discoverability by up to 30% also needs care. The announcement cites a NielsenIQ article from August 2025. But within that article, the link given as the basis for the 30% figure points to a 2021 article on search terms. In 2021, generative AI assistants had not yet spread, so the figure does not measure product recommendations by AI. The same range will not necessarily apply to discovery through AI agents.
That product data is not yet ready for AI agents, on the other hand, has been shown repeatedly in other research. In a PYMNTS Intelligence study with Visa Acceptance Solutions, only 15% of merchants had structured product data for agent access. Meanwhile, 68% expect agents to account for at least 5% of digital sales within two years. When feed management company Channable analyzed 9,689 companies on its platform, 77% scored below 60 out of 100 on AI readiness, and less than 1% scored 80 or above. The two measure different things, but they agree that the gap between expectations and readiness is large.
What ecommerce businesses should check
The acquisition itself is about overseas enterprise vendors, but the problem it addresses applies just as well to ecommerce in Japan. There are four things worth checking.
First, where the correct information for your products actually lives. For appliances, tools, building materials, and industrial supplies, specs are often written only in a manufacturer's PDF or instruction manual. AI assistants do not treat a table inside an image or a number deep in a PDF the same way as an attribute on a product page. Information that is not in the structured fields of a product page tends to be as good as nonexistent to AI.
Second, whether you keep the source of enriched values. Filling attributes in bulk with generative AI is now something any company can do. The difference lies in whether, when a wrong value slips in, you can trace which statement in which document it came from. Source records also help when investigating the cause of returns or wrong orders.
Third, where you write corrected values back. If you fix only the search engine settings or a feed for one channel, the improvement does not reach other outlets. Fixing the PIM or the core product master may look like a detour, but it is the efficient approach.
Finally, whether you use your site search logs. Zero-result searches and filter conditions used often are clues to which attributes are missing. Even at a much smaller scale, the loop Coveo and Anglera are planning can be tried with your own logs and a spreadsheet.
The announcement does not say whether Anglera's enrichment will be offered for Japanese-language product data or to customers in Japan.
Summary
Coveo's acquisition of Anglera signals that the decisive factor in product discovery in the AI era is shifting from how smart search is to how complete the underlying data is. External AI assistants recommend only the products they can understand.
At the same time, the numbers showing an effect come from a small, self-reported test, and neither the deal price nor the pricing for the enrichment add-on has been disclosed. What to watch next is how far the link between enrichment and search becomes a product, and which customers' results are cited, at the November 12 webinar and in upcoming earnings.
