Most customer data problems arrive described as a platform gap. The records are in a CRM, a point of sale, an ecommerce system, and an inbox export, and none of those systems agree on who a person is. The useful first question is narrower: which decision is stuck, and which identity does that decision require?
Start from the decision
I do not begin by connecting every system. I begin with the decision someone is trying to make. Who is worth a retention offer. Who has gone quiet. Which audience can actually be built from the history you already have. That decision names the grain of the customer, the history it needs, and the systems that have to participate. Everything else can wait.
If the decision is a campaign audience, I need a stable person, a way to see recency and value, and a path back into the tool the team already sends from. If the decision is operational — a service case that should know the order history — the grain and the systems change. The same "unify the customer" sentence covers both, and they are not the same build.
Then the identity, not the warehouse
A unified customer master is a resolved key plus the history you are willing to trust. It is not a copy of every table. I look for the keys each system already has: email, loyalty id, phone, account number, and the anonymous transaction that has none of those. Where two records are the same person, I connect them and keep the source history, so the result can be explained. Where they only might be the same person, that stays a question, not a merge.
Loyalty belongs here only when identification is the missing input. A loyalty program can give a customer a reason to say who they are, and that identified behavior can feed the same model. It is a means to a richer first-party history. It is not the point of the work.
What I leave alone on purpose
I do not replace the CRM or the point of sale to answer a segmentation question. Those systems keep running. The model sits beside them and refreshes, so the audience does not become a spreadsheet someone has to rebuild. If the underlying pipelines cannot be trusted, that is a modernization problem and it should be named as one, instead of being hidden inside a customer project.
If this is the shape of the problem, the customer intelligence page is the longer version. Describe the systems and the decision in your own words — on the intake or through the form — and I will tell you whether this is the right first move.