Prepared with AI assistance from the linked primary sources. Company and government claims are attributed to their issuers. How we use AI.

A stronger model can improve a product without making the company behind that product difficult to replace. Defensibility depends on what remains valuable when a customer can access similar underlying capabilities elsewhere. Two Asian examples highlight assets worth examining: data generated through physical operations and knowledge embedded in a specialised workflow.

Data has to connect to a useful task

TacnIQ.ai’s September 2026 financing release describes tactile AI that interprets physical-contact signals. The company says its sensor-based deployments across several industries generate data used in model development. It reported US$1.5 million secured from a lead investor within a larger pre-seed round. TacnIQ company-issued financing release, 23 September 2026.

The potential advantage is a relationship between collection and learning. If a system encounters situations that matter to customers, records them appropriately and uses them to improve future performance, operating experience may become a useful asset. But the existence of a dataset does not establish exclusivity, legal rights, quality or a competitive lead.

A meaningful assessment would ask which failures the data helps resolve and whether the improvement transfers to a new customer. Information from one setting may be highly useful there yet offer little benefit in another. Volume alone cannot answer that question.

Workflow knowledge can be harder to package

Sakana AI’s March 2026 account of its MUFG lending project describes using earlier lending documents, internal rules and manuals in a secure environment, with employees checking draft outputs. The work covers tasks beyond document generation, including organising information and supporting financial analysis. Sakana AI and MUFG lending project, 6 March 2026.

The valuable layer may therefore include how information is selected, how exceptions are handled and how a person can verify a recommendation. These are properties of the complete application. Replacing its model does not necessarily replace the accumulated understanding of the customer’s process.

Equally, customer-specific work can become a constraint. A product that needs extensive rebuilding for each buyer may have strong relationships but limited repeatability. Defensibility and efficient growth are separate questions; both matter when interpreting claims about a proprietary platform.

Trust needs an operational meaning

For a buyer, trust becomes concrete when the system exposes its sources, respects permissions, records actions and supports a workable recovery path. Consider a hypothetical purchasing assistant that produces a persuasive answer but cannot show which contract version it used. A more modest system with traceable evidence may be easier to approve for real work.

The same principle applies to service. Implementation knowledge, support response and predictable behaviour can make a supplier valuable even when its underlying model is available to competitors. These advantages need to show up in customer outcomes rather than remain claims in a pitch.

The test after the demonstration

Watch for renewals, expansion within existing customers, shorter implementation effort and documented performance on the actual task. Ask whether the company can retain those benefits as models, customer systems and operating conditions change. The cited releases do not independently establish that either business has a durable moat. They identify where such an advantage might develop—and what additional evidence would make that proposition more convincing.

Sources

See something we should correct? Contact the newsroom.