Start with the narrowest supported claim.
Imagine a company adds a second AI provider to its public vendor disclosure. The first defensible statement is that the second provider now appears in that disclosure. That is narrower than “the company switched,” and much narrower than “the incumbent lost the customer.”
The company may be testing a new model, serving another geography, or supporting a product team with different requirements. Each explanation is compatible with the original observation.
What would support a replacement?
A replacement interpretation becomes more plausible when several independent observations align: a new integration, an old integration retired, and a dated statement about migration. Even then, the scope may be limited to one workflow.
A missing page is weaker evidence. It can reflect a redesigned site, a renamed subsidiary, an expired posting, or a failed collection attempt. Treat those alternatives as questions to resolve.
Keep three labels separate.
- New adoption: an additional vendor relationship is observed.
- Co-adoption: evidence suggests the company uses both vendors.
- Potential displacement: a sequence is consistent with replacement, with the supporting evidence and remaining limitations recorded.
An example of a useful research note.
Illustrative example: “Aster Labs added Model Vendor B to its disclosure on September 4. Model Vendor A remains listed. This supports a co-adoption interpretation; spending, production scope, and contract status are unknown.”
That note gives a sales researcher a discovery question and a data analyst a carefully bounded observation. Calling the same event “customer churn” would conceal the uncertainty.
Use a signal to ask a better question.
Before acting, ask which entity is represented, whether the observation has independent support, and what evidence would change your interpretation.
A precise claim can be more useful than a confident one. The aim is to preserve the path from evidence to conclusion.