Federating data is a fairly 'solved' issue. Ironically (or not) enterprise IT is ahead of the average organism in this space.
Faced with legacy systems communicating (or not communicating) via various channels, Middleware, ESBs, SOA, message queues, content-based routers, business process management, event processing, and data virtualization have all been created so that a development group can factor these systems into canonical representations of business events.
At that point, the input method ranges from the simple (existing web app to API) to complex (treating a batch-oriented system in a semi-synchronous manner or screen-scraping an ancient smart terminal as a queriable data source).
At that point, the real challenge isn't the input method but master data management and security/authority/privilege.
Storing, doing simple metrics, making assumptions and returning the data is not a problem. Standardizing input methods is.