Author:Aurora / iDIYuan

Executive Editor:Louis Lu / iDIYuan

At 4 p.m. on a typical weekend, a retailer can see shoppers moving through the aisles, asking employees questions and picking up products. Each of those interactions generates a signal, but most retailers still struggle to connect those signals into a clear picture of what is happening and what to do next. Enterprises today have more data than they can readily turn into action: CRM and ERP records, transactions, camera feeds, IoT and sensor data, frontline conversations and years of accumulated know-how, often scattered across separate systems. The challenge is no longer simply collecting data but connecting what an enterprise sees and hears to decisions, and turning those decisions into action.

Founded in 2017, Whale began with computer vision, IoT and retail digitization. It now casts itself in much broader terms: as an AI-native operating system for enterprise operations. “AI is no longer just a single application,” said Smile Huang, Whale’s vice president of sales for Asia-Pacific, during an interview in Singapore. “It can be used across enterprise operations.” The distinction Whale draws is between AI and the digitalisation projects that have absorbed corporate IT budgets for the past two decades. Digitalisation was largely about recording the business: capturing what happened and making it visible on a screen. An AI-native operation is meant to do something harder: continuously interpret what is happening and act on it, without waiting for a person to read a dashboard first.

Physical-world data is Whale’s entry point, not its end goal. Its broader ambition is to make AI part of the operating fabric of a company, rather than simply another retail tool or standalone enterprise application.

From sensing to action

Whale explains its product architecture in deliberately simple terms: AI can see, hear, express, think, remember and connect, while an infrastructure layer manages the models and systems underneath.

“We have always insisted that AI should be easy for everyone to understand,” Huang said. “It should not be something so complicated that only technical people can understand it.”

The products form a loop rather than a catalogue. SpaceSight is the visual layer, analyzing foot traffic, customer flows, dwell time and other activity in physical spaces. Echo analyzes frontline conversations, while Stardust connects the IoT hardware that feeds those systems. Harbor handles the memory, holding enterprise knowledge and data. Alivia is the reasoning layer, where sensing and memory can be turned into recommendations or tasks. Lume translates the output into content employees can use.

Beneath it all sits Novus, which manages which models run, where they run and what data they can access. Whale calls the shared architecture its World Model. Where language models teach machines to talk about the world, a World Model teaches machines to understand it: not just to perceive what is happening now, but to reason about what comes next and act on it. Whale approaches this from a spatial angle: equipping physical spaces with vision, hearing and perception, so that a store, a restaurant or a warehouse can understand what is happening inside it and respond.

Whale’s enterprise knowledge and AI-assisted content management architecture.Source: Whale.

The 4 p.m. store example shows how Whale says those pieces can work together. Huang described combining CCTV-derived data with frontline conversations, then asking Alivia what the women entering the store at that hour appeared to be paying attention to, and what the business should do next. “It is a bridge,” Huang said. “It brings data and action together, summarises them and then gives you a recommendation.”

For the loop to close, Alivia’s recommendation would need to become an action: a campaign brief assigned to a team, for example, approved by a manager and then measured against subsequent sales. Those results could then feed into the next recommendation. Huang’s example shows how Whale gets from data to a recommendation, but not what happens next: who acts on it, or whether it actually improves performance.

Whale’s enterprise AI-native workflow architecture. Source: Whale.

Whale’s larger proposition is that defined, repeatable parts of any workflow, in any function, can increasingly be handled by AI agents, while people remain responsible for work requiring judgment, creativity and flexibility. “AI agents can do a lot of repetitive work - work that has already been defined and standardised,” Huang said. Human value, she said, lies in “innovation, flexibility, and responding to urgent situations.”

Whale is applying the same principle to its own transformation. Huang said employees who once spent substantial time building presentation decks can now use natural-language prompts to generate interactive HTML pages with AI.

An example of Whale’s AI-assisted content workflow interface. Source: Whale.

Enterprises Pay for Outcomes, Not Just AI

The foundation model, Huang argues, is becoming less important as a source of differentiation. What matters more is how AI sits in a company’s data, workflows and operating needs: what information it can access, where it fits into the workflow, and whether it delivers measurable results. Huang cited a line that Whale founder and CEO Jerry Ye often uses: “AI is not really selling AI. It is selling know-how.”

For smaller customers, that can mean standardised products and functions. Large enterprises are more complicated: their existing systems and processes may mean Whale has to start with consulting, mapping workflows, organising information and figuring out what can be handled with off-the-shelf products and what needs to be customised. “Some customers are very clear about what they want,” Huang said. Others simply know they want to become more AI-driven: “I know where it hurts, but I don’t know what I need.”

Huang said Whale spent about six months with one customer working through its existing processes before deciding what to build. “What we deliver is not simply software or a dashboard,” she said. “It really has to be integrated into their daily workflow.” A high-end retail group in Southeast Asia offers a glimpse of what Whale is trying to deliver. Huang said SpaceSight helped the customer separate meaningful customer traffic from general footfall, giving it better information for decisions such as site selection and rent negotiations. It also showed how customers moved from browsing and trying products to making a purchase.

The commercial test is simple: does the technology cut costs, improve efficiency, solve a real business problem or help drive revenue  As Huang put it: “If I spend this money, how can I see the return on investment ”

That focus on measurable returns creates a deeper tension for enterprise AI. The closer AI gets to a customer’s actual business problems, the more customization it may require. Whale can tailor a system to a company’s data and workflows, but that work comes at a cost. The more consulting and customization involved, the more expensive each deployment becomes, and the harder it is to build a business that scales. Technology can scale; trust does not.

That tension becomes even more pronounced overseas. If serving one large customer requires extensive workflow mapping, the model is difficult to replicate across markets with different languages, business practices and regulations. Whale says it now serves more than 1,600 enterprises in over 45 countries. At that scale, the challenge is not simply exporting a product but building the local relationships, service capacity and trust needed to sell and deploy it.

Whale has responded by splitting responsibilities across regions. Huang said product development and core R&D remain concentrated in China, while overseas teams focus on sales, presales, project management and customer service. Asia-Pacific requires “a diversified playbook”, Huang said, because markets differ widely in language, business practices and buying behaviour.

In markets such as Singapore and Australia, Huang said, sales tend to be driven more by the business case and measurable value. In Indonesia, the Philippines, Vietnam and Thailand, relationships and local support can matter more. “You can introduce yourself online first,” she said, “but whether you can close the deal still depends heavily on whether you have good local service and whether the customer has met you.”

Whale is therefore building what Huang called a “partner ecosystem” of local business and channel partners, including local representatives and system integrators with established customer relationships and business networks. The aim is to use those networks to reach more companies and give customers local support before and after a sale.

Whale team members at an overseas event. Source: Whale.

That ecosystem extends to Whale’s investors as well. Many are banks, telecom operators and industrial groups with regional networks, giving Whale access not only to capital but also to potential customers and local business relationships.

In July 2026, Whale announced a $40 million Series C3 extension, bringing its total Series C financing to $100 million. According to the company, the round was led by a fund managed by CMB International and SMBC Asia Rising Fund, the corporate venture capital arm of Japan’s Sumitomo Mitsui Banking Corporation, with participation from Krungsri Finnovate, Singtel Innov8, Hyundai Motor Group and Charisma Partners.

Whale announced a $40 million Series C3 financing extension in 2026. Source: Whale.

The Singtel relationship illustrates how those investments can be more than financial. Beyond its venture arm’s stake, Singtel has become both a customer and market partner, deploying Whale’s retail analytics in its stores and working with the company on joint market development. For AI companies expanding overseas, the harder part is often not the technology itself but building the local relationships and trust needed to sell it. Strategic investors cannot replace that work, but they can open doors to customers and networks that might otherwise take years to build.

Three Hurdles on the Road to Enterprise AI Infrastructure

Whale is not alone in moving enterprise AI beyond chat interfaces and standalone copilots. AWS and ServiceNow already connect AI agents with enterprise data, workflows and governance. The connection itself is no longer distinctive. Whale’s point of differentiation is where it starts: with cameras, sensors and audio from the physical world, combined with enterprise systems.

The harder question is whether that foundation can become something closer to an operating system for enterprise work. Its next phase highlights three challenges that much of the enterprise AI industry is still working through.

The first is moving AI from recommendation to execution. Generating advice is becoming easier. Turning that advice into assigned tasks, confirming the tasks were completed and feeding the results into the next decision is much harder. Huang’s store example centres on analysis and recommendations, while Whale’s public materials point to automated audits, issue routing and corrective action prompts, an indication of where the company wants to go next.

The second challenge is turning custom projects into repeatable products, a commercial test facing nearly every B2B AI company. Whale’s consulting-led approach gives it deep knowledge of customer workflows. The opportunity is to turn that knowledge into templates and reusable modules, so each new customer needs less bespoke work than the one before it. Implementation time and the proportion of functionality that can be used across customers will be important measures of whether that model is scaling.

The third is data governance once AI enters the physical world, an issue no vision, voice or IoT-based AI provider can avoid. Cameras in stores and microphones worn by frontline staff raise questions around privacy, consent and data handling that vary by market. Whale says it respects employee and customer privacy, ensures proper notification and authorization before any data collection, and does not collect sensitive biometric identifiers such as facial recognition data or voiceprints. Customers can deploy locally or on premises and process data in line with local requirements.

The shoppers entering the store at 4 p.m. are only the starting point. The real test is whether the insight they generate can lead to an action, and whether that action can feed back into the next decision, at scale, across markets. That is a harder problem than building a product that works in one store. If Whale can solve it, the result would be something more significant than another enterprise software suite: an AI layer woven into how a business actually runs.

来源：数智猿