Start with the inventory, because most businesses cannot produce one. Customer records, sales history, website analytics, advertising data, financial records, support conversations, and whatever sits in spreadsheets. For each, note where it lives, who can reach it, whether it can be exported, and whether it agrees with the other systems. That last column is where the surprises are, since the same customer count frequently differs across three tools and nobody knows which is right.
Then decide what you are actually trying to answer, because collecting data without a question produces dashboards nobody reads. For a business entering year two the useful questions are consistent: where do customers come from, what does one cost to acquire, what are they worth, which work is profitable, and why do people leave. If your current data cannot answer those, the gap tells you what to start capturing now, since none of it can be reconstructed later.
Establish one source of truth per category rather than reconciling several. Customer records in one place, financial records in another, and a rule about which wins when they disagree. Businesses that maintain the same information in two systems end up trusting neither, and the reconciliation work grows with volume.
Consider retention and obligation while you are looking. Holding personal data indefinitely increases what is exposed if anything goes wrong, some records have required retention periods, and you may have obligations to delete on request. Deciding what you keep and for how long is part of the same exercise and is rarely done otherwise.
Keep it to two pages and revisit it annually. The value is in having looked rather than in the document, and the most common outcome of doing this exercise is discovering two systems holding contradictory versions of something important, which is worth the afternoon on its own.
Assign ownership for each category even when the owner is always you, because the exercise reveals the systems nobody is responsible for maintaining. Data quality degrades without somebody accountable for it, and the categories that drift furthest are usually the ones every system touches and no system owns, such as customer contact details appearing in four places with three versions of the same phone number.