The operations managers were delighted. They used the AI tool every day. They praised it in interviews. By the measures product teams tend to celebrate, IceCreamLabs seemed to have made something people wanted. Then came the awkward part: according to the company’s account, it could not close a single deal.
- IceCreamLabs co-builds AI companies with founders.
- Its method tests the problem, buyer, price, and competition before a full build.
- The partnership includes engineering and operations in return for negotiated equity.
The missing person was the finance executive. Operations managers experienced the inconvenience; someone else controlled the money. That person had never been brought into the conversation. In its March 2026 essay recounting the episode, the studio put the mistake neatly: “We had nailed the user. We had ignored the buyer.”
This is a useful place to begin understanding IceCreamLabs. The San Francisco business calls itself an AI venture studio. It identifies opportunities, recruits or partners with founders, and helps turn ideas into operating companies. Its most interesting proposition concerns the work surrounding the software: discovering whether a capable machine has a customer with a budget.
A catalog was the classroom
Before the studio’s current collection of ventures, there was the product catalog. In a July 2018 interview, founder and CEO Madhu Konety described CatalogIQ, a machine-learning product for retailers, brands, and marketplace sellers. Its assignment was wonderfully unglamorous: check whether a product’s images and attributes agreed, improve incomplete records, and help prepare the catalog for publication.
Picture the job from a merchandising manager’s desk. New supplier data arrives. Images need checking. Descriptions need attention. The long tail of merchandise competes with the best sellers for limited human attention. Konety described a situation in which reviewing every record manually could become impractical. CatalogIQ offered automated checking, categorization, and content generation.
That work helps explain the studio’s present interests. Retail AI demanded attention to messy inputs and an existing workflow. Today, IceCreamLabs names Walmart, Pepsi, and Siemens among the enterprises in its AI history. Its current LinkedIn description connects its earlier retail and consumer-goods work to its move into building startups. The studio is selling the judgment acquired while putting AI into businesses.
Four weeks to earn the build
The company’s April 2026 validation framework makes that judgment unusually concrete. It sets aside four weeks for testing an opportunity before substantial engineering. The sequence moves from the problem to the budget holder, then to payment commitments and competitive positioning. Each step gives a founder an opportunity to stop.
- 01Find the painInterview users and buyers about current work.
- 02Find the budgetMap who pays, influences, and approves.
- 03Ask for commitmentDiscuss price; seek a pilot, deposit, or letter.
- 04Find your openingCompare alternatives and define a position.
Consider the first week. The instructions call for interviews about how people work today, without introducing the proposed solution. That restriction matters. A founder describing a clever invention invites opinions about the invention. A founder asking about last Tuesday’s frustrating task may discover whether the problem exists at all.
The next question concerns authority. Who can approve a purchase? Which budget would cover it? After that comes a request for something tangible: a pilot agreement, a deposit, or another commitment. These signals differ in strength, but each asks the prospect to contribute more than encouragement. The process turns a pleasant conversation into a commercial test.
“We spend more time talking to customers and less time writing code”
IceCreamLabs, April 2026
In a separate April essay, the studio says it has slowed validation even while accelerating its build phase. This is the change in emphasis its writing makes explicit. As software becomes faster to produce, choosing the right customer problem deserves more attention. Shared models and deployment infrastructure can shorten engineering work; they cannot decide whose purchasing process a startup must survive.
The product is also a company
The portfolio now reaches beyond retail. CommerceFlow is positioned around B2B commerce, including procurement and catalog enrichment. Nina helps people search for property, with its own site focused on North Bangalore and conversations through WhatsApp. Modo is currently described by the studio as conversational data analysis. OpenTide addresses discovery and transactions between agents. NuWorks builds AI systems for enterprises and startups.
These businesses have different users, sales processes, and requirements. Their connection is the studio behind them: engineering, product work, shared infrastructure, and operating support. IceCreamLabs’ stated model pairs those resources with founders who bring domain knowledge and customer relationships. Someone who understands an industry can approach without already possessing an AI engineering team.
The studio’s stated typical range for a founder arriving with a solid idea and domain expertise. A planning estimate, not a guaranteed delivery date.
The arrangement has an ownership cost. IceCreamLabs says it takes an equity position based on the opportunity and its operational contribution. That makes choosing a studio a co-founder decision. Founders should weigh the value of supplied engineering and execution against the ownership they exchange, then examine who controls product decisions and what each side will contribute.
The question to borrow
The method fits founders who can reach a defined market and test a recurring problem. It gets harder when access to buyers is poor, purchasing cycles are long, or an offering has no clear advantage over existing tools. The studio’s framework treats those conditions as reasons to reconsider the entry point. A functioning prototype cannot settle them.
There is a small action any reader can copy. In the next discovery call, ask where the money would come from if the customer chose to proceed. Then listen for a named department, an existing expense, and a person who can authorize the change. IceCreamLabs’ early operations users already knew the product was useful. The company still needed to meet the person who could say yes.