The problem with reconstructed intent.
If a signal is assembled from traces left after the fact, its apparent precision can hide a timing problem.
A large amount of business intelligence is reconstruction. A new job appears, a website changes, a technology is detected, a review is written, an executive mentions a project, or a contract eventually becomes public. A model connects the traces and infers what probably happened.
Reconstruction can be extremely useful. It is also downstream of the thing being reconstructed.
That timing distinction is easy to lose because retrospective data often looks clean. Once an outcome is known, the path toward it appears obvious. Before the outcome, the same path was uncertain.
Observed work activity has a different property. It does not need to know the eventual outcome in order to exist. The relationship can move while the result is still unresolved.
This is why we are careful with the word prediction. Evericade does not need to claim that every acceleration predicts a contract or every decline predicts churn. The primary claim is simpler: economic coordination can be observed before many conventional records of that coordination appear.
Interpretation can then happen on top of observation, with whatever model, analyst or workflow is appropriate.
The distinction sounds semantic. In practice, it determines whether your dataset is looking through the windscreen or reconstructing the journey from tyre marks.