Montréal / Where to startShared data does not always mean shared meaning.
A Montréal business serving several markets may have product records, customer content and policies in more than one language. The challenge extends beyond producing a fluent answer. The workflow must identify the correct version, market, policy and authorised source. An assessment follows those distinctions through the proposed use case so that a shared AI capability does not silently flatten differences the business needs to preserve.
01Define the context the answer must carry
Identify where language, market, channel and transaction date change the information a user should receive. Map these attributes to the systems that hold them. A translated policy document and the applicable policy are separate questions; the assistant needs reliable selection rules before wording can be evaluated.
02Establish ownership of approved information
Review who maintains product facts, approves content versions and decides which record takes precedence. Examine how corrections reach search and retrieval sources. Determine which gaps block the selected use case rather than assuming every record must first move to a new data platform.
03Evaluate quality by business consequence
Test examples with incomplete market assignments, outdated documents and inconsistent product attributes. Include the appropriate language and business reviewers where needed. Measure whether the answer is grounded in the right evidence and how much correction it requires, alongside recurring costs and the complexity of maintaining several contexts.