Overview
Crossing Minds spent eight years turning preference prediction into infrastructure other businesses could use. Its path ran from entertainment discovery to merchant recommendations, then to research on selecting useful context for language models. In June 2025, the team announced it was joining OpenAI. [1] [9]
The technical work asked which item helps a person or task, then learned from the result. The commercial setting changed repeatedly. That distinction explains why a recommendation company could become relevant to a frontier model lab. Public sources establish a team transaction. They do not disclose the price, transferred assets, or investor return.
Founding Story
Alexandre Robicquet and Emile Contal traced their collaboration to a Paris machine-learning lab in 2014. They shared music recommendations and noticed that suggestions improved as they learned each other's tastes. They began testing algorithms in 2016, working around doctoral research in Stanford and Paris. Their founding account describes Sebastian Thrun encouraging the move into business. [2]
The company's profile gives 2017 as its founding year. Research experiments therefore predate the company. [3] Its first consumer product, Hai, appeared at Disrupt in September 2018 with entertainment recommendations. [12]
Index Ventures lists Robicquet, Contal, and Thrun as founders. Robicquet's acquisition post also calls Contal and Thrun his co-founders. [13] Addressing Contal, he wrote: “your brilliance and grit built the backbone of everything we created.” Addressing Thrun, he added: “your mentorship permitted us to be bold from day one.” These remarks distinguish the technical and mentoring contributions. [8]
Timeline
- 2014–2016: Robicquet and Contal collaborated on research and tested recommendation algorithms. [2]
- 2017: Founding year listed by the company. [3]
- September 2018: Hai appeared at TechCrunch Disrupt. [12]
- October 2021: Crossing Minds announced a $10 million Series A and described its pivot to business customers. [6]
- April 2022: Shopify invested; its merchants gained access to the platform. [4]
- November 2024: The team published its ICLERB retrieval benchmark. [7]
- June 2025: The company announced the team would join OpenAI and end independent operation. [9]
What They Built
The merchant product helped shoppers choose from large catalogs. Inkbox's case shows the workflow. A welcome quiz collected taste signals. Recommendations then appeared on a results page, in a personalized homepage section, alongside product pages, and in email. The system combined explicit answers with on-site actions. [5]
This addressed the cold-start problem: a store needs useful suggestions before it has a long purchase history for a visitor. Asking a few relevant questions produces information immediately. It also requires merchants to understand which questions predict a purchase. The case describes collaboration on the quiz and integration, rather than an algorithm operating without human setup.
The later research applied a related idea to language models. Retrieval adds outside material to a model's prompt. ICLERB evaluated retrieved examples by their contribution to task performance, rather than their resemblance to the query. The paper's authors were Marie Al Ghossein, Contal, and Robicquet. Its findings support a specific benchmark result, not a claim that the method wins on every application. [7]
In February 2025, the RAGSys explanation extended that direction. It described recording interactions and outcomes, retrieving useful examples, and generating corrective instructions after poor results. This was a proposed feedback system for adapting model context without changing the model's weights. [11]
Market Position
Target Customers
The company moved from individual consumers to businesses selling products or helping people discover content. Its 2021 founder account named Danone, Dutchie, Reelgood, and Inkbox. Shopify later offered access to its merchant ecosystem. [2] [4]
Market Size
Public materials establish a broad merchant distribution channel, but do not disclose Crossing Minds' paying customer count or obtainable market. Shopify's merchant reach is not the startup's customer base. No revenue-based market estimate is justified by these sources.
Competition
A merchant recommendation supplier competes with its host platform as well as other vendors. Shopify currently generates related products using purchase history, descriptions, and collections. Merchants can also select complementary products. [10]
That sets a practical threshold for a new supplier. Basic product suggestions already exist. A separate service must show that its extra personalization earns more than its integration and operating costs. This is a current competitive constraint, not evidence that Shopify caused Crossing Minds' exit.
Business Model
Crossing Minds sold recommendation infrastructure to businesses. Its Series A announcement described a B2B pivot and named Radical Ventures as lead investor, with Index Ventures, Partech, and Lerer Hippeau participating. The announced round was $10 million; Shopify's later investment amount was undisclosed in the company release. [6] [4]
The sources reviewed do not establish current or historical contract prices, recurring revenue, gross margin, or customer acquisition cost. Inkbox's account also describes integration support. Service effort therefore belongs in any analysis of the business, but its cost cannot be calculated from the disclosed evidence.
Traction
The Inkbox case reported increases of 250% in email click-through rate and 69% in new-user conversion. It also displayed a larger website-conversion increase. These are vendor-published case metrics. The page does not provide a controlled experiment design, confidence intervals, or a common measurement period. They document a customer relationship and claimed improvement; they do not establish average results across merchants. [5]
Post-Mortem
The durable asset crossed product boundaries
The most useful interpretation is that Crossing Minds carried a technical problem across different buyers. Merchant ranking selected products that would help a shopper. Its later research selected examples that would help a model complete a task. Both required feedback about useful results. The published research and Robicquet's move into agents and retrieval at OpenAI support that continuity. [7] [8]
This is an inference about the team's transferable work. Neither OpenAI nor the company disclosed enough transaction detail to establish which code or patents changed hands.
Customer proof did not explain the exit
The Shopify investment and Inkbox relationship show that the team reached commercial users. They do not reveal whether the independent business could support its research and delivery costs. A convincing financial explanation would need retention, contract value, support costs, and the consideration paid in the transaction. Those figures remain unavailable.
A competing explanation is that OpenAI offered a better setting for the research regardless of merchant economics. Robicquet's stated focus on agents and information retrieval is consistent with that account. Public evidence cannot rank that motive against liquidity, financing, or operating constraints. [8]
The company's June announcement supplied the clearest outcome: “This marks the end of Crossing Minds as an independent company”. It did not supply a shutdown date for every customer deployment or a price. The defensible conclusion is a team transition with an end to independent operation. [9]
Key Lessons
- Follow the reusable problem. Crossing Minds changed buyers while keeping preference learning and retrieval central. A pivot can preserve technical depth when the underlying job remains coherent.
- Separate reach from adoption. Shopify's investment opened a channel. It did not disclose how many merchants bought or retained Crossing Minds.
- Measure the result the buyer needs. The ICLERB work measured downstream task usefulness. The same discipline requires merchant tests to measure incremental commercial results.
- Price the delivery work. Inkbox's account includes help with integration and quiz design. A recommendation business must account for that effort alongside model quality.
Sources
- Crossing Minds: A new chapter.
- Recommendation is the next frontier of online experiences, October 18, 2021.
- Crossing Minds company profile.
- Recommendation Startup Crossing Minds Receives Strategic Investment from Shopify, April 21, 2022.
- How Inkbox helps customers find the perfect tattoo from its vast catalog.
- Crossing Minds Secures $10 Million Series A Financing to Help Businesses Make Smarter, Safer Recommendations, October 18, 2021.
- ICLERB: In-Context Learning Embedding and Reranker Benchmark, November 28, 2024.
- Alexandre Robicquet: Crossing Minds team joining OpenAI.
- Crossing Minds: team joining OpenAI, June 2025.
- Shopify: Customize product recommendations with Search & Discovery.
- Closing the Loop: Real-Time Self-Improvement for LLMs with RAGSys, February 23, 2025.
- Startup Battlefield Competition – Hai by Crossing Minds | Disrupt SF 2018, September 6, 2018.
- Crossing Minds: Index Ventures portfolio.