Constructor: product discovery before agentic checkout

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A retailer or distributor can have a large catalogue and still make the buyer work too hard to find the right item. Search may miss a synonym, rank an accessory above the main product, or answer from stale data. Constructor addresses that discovery layer; it does not become the merchant’s product-information, enterprise-resource-planning, configure-price-quote or payment authority.

What Constructor provides

Constructor, founded in 2015 by Eli Finkelshteyn and Dan McCormick, is a U.S.-based proprietary SaaS platform for search, browse, recommendations, collections, merchandising and retail media. It is API-first and headless: the customer retains the interface, catalog truth, order systems, payment configuration and approvals. The platform combines a regularly updated catalogue with searches, clicks, conversions and anonymous visitor IDs, using natural-language processing, transformers, large language models and reinforcement learning to rank products. Merchandisers can boost, bury, pin or override results. 1 2 4 5 6

Customers provide product and inventory data through connectors, JSON, JSONL, CSV or REST. Documentation recommends a daily full sync plus intraday deltas; REST integrations need batching and retry logic and support up to 10,000 products per request. Variant links, IDs, freshness, event attribution, consent and permissions are implementation requirements. 4 5 6

From search to agents and checkout

The earlier proposition was behavior-informed enterprise search. TechCrunch’s 2021 report described billions of monthly queries, 233% revenue growth and customers including Sephora, Bonobos and Backcountry. Constructor later expanded into browse, recommendations, collections, merchandising and retail media. Financing included a reported $55 million Series A in 2021 and $25 million Series B in 2024; the latter release put valuation at $550 million and total funds above $85 million, while earlier financing remains incompletely itemised. 11 12 13

By 2026, AI Shopping Agent (ASA) handled open-ended needs using catalog data, manuals or articles and behavior; Product Insights Agent (PIA) answered product-page questions from specifications, manuals, reviews and related data. A Shopify PIA widget was announced in August 2026. On 5 October, Constructor announced global Stripe Agentic Checkout for discovery, questions and payment within ASA or PIA. Production requires integration work and Stripe credentials. Answer Engine Assistants remain beta; these announcements do not prove authoritative compatibility, price, stock, permissions or purchasing decisions. 1 7 9 10 15

What the public outcomes show

The strongest evidence is named but mainly provider-published. Constructor’s Petco case reports a 13% increase in ecommerce conversions after replacing an Elasticsearch-based solution on AWS; it mentions A/B testing but not baseline, dates, denominator or confidence interval. home24 reports double-digit search-conversion growth from a 50/50 A/B comparison over three weeks after a 2–4-week Proof Schedule, across Germany, France, Switzerland, the Netherlands, Belgium, Italy and Austria and four languages. Absolute rates and significance are not published. 16 17

The customer index also reports Target Australia +A$13 million in search revenue, Saatchi Art +47.29% revenue per visitor, AG Jeans +16% conversion, flaschenpost +4.14% relative search conversion, Cromwell 21x return on investment and Belk +$35 million revenue. Periods, denominators, counterfactuals and independent replication are generally missing. Constructor’s FY26 announcement, covering 1 February 2025 to 31 January 2026, reports 322 billion platform interactions, 82% customer growth and 116% EMEA revenue growth, plus PIA examples of +52% add-to-cart and +56% purchase rate and ASA examples of +10% website revenue and +6% search conversion. These are company claims, not audited or guaranteed return on investment. 3 8

An illustrative controlled pilot

For an illustrative multilingual B2B distributor, start with one governed product family. Reconcile IDs, variants, revisions, units, materials, compatibility, inventory timestamps, entitlements and duplicate events. Feed approved descriptions and manuals for discovery, but keep price, stock, credit, quotation and order status in authoritative systems.

Run a read-only incumbent comparison with synonyms, ambiguous units, stale stock, near-duplicate variants, malicious content and unauthorised-purchase requests. Pre-register zero-result, wrong-item, search-conversion, add-to-cart, revenue-per-visitor, qualified-enquiry and human-exception rates. Require current, inspectable evidence and keep recommendation separate from an approved offer.

Decision and controls

Constructor merits a controlled discovery pilot for enterprise retail, distribution or B2B catalogues with reliable product, inventory, pricing and event data. It is less proven for non-SKU industrial work involving drawings, tolerances, capacity, co-design and binding RFQs. No public pricing, minimum contract, setup fee, retention arrangement or model-training opt-out was found. Confirm those terms, export and deletion mechanics, content rights and support obligations before committing. Keep human approval for compatibility, safety, customer-specific pricing, sponsored placement, stock exceptions and checkout authority. 13 14

Sources and next reading

Explore OfferSignal · Focused Business & AI Workshop · All research

References