Do I need a data analyst?

No. Three sorted columns in a spreadsheet is a legitimate implementation and will find the segment that matters.

How many segments should I use?

Fewer than you think. Splitting each axis into thirds gives you plenty, and most businesses only ever act on two or three of the resulting groups.

RFM analysis for a business with one location

Score each customer on how recently they bought, how often, and how much. Three numbers, all already in your records, no modelling. The segment worth finding is high frequency and high spend with poor recency: your best customers, who have stopped coming.

RFM predates almost everything in modern marketing and remains the highest return-per-hour analysis available to a small business, because every input is already in your records.

The three axes

Recency. How long since their last visit. Usually the strongest single predictor of whether somebody comes back.

Frequency. How often they come.

Monetary. How much they spend.

Doing it crudely is fine

Split each axis into thirds: high, middle, low. Every customer lands in one of 27 buckets. You will only ever act on a few of them.

The segment that pays for the exercise

High frequency, high spend, poor recency.

These are your best customers, and they have stopped coming. They are the most valuable group you have and the most recoverable, because they already liked you. Almost no small business looks for them, because a blended sales figure hides them completely.

The other two worth naming

High recency, low frequency. New or occasional customers who came recently. The group where a loyalty programme has the most room to change behaviour.

High everything. Your regulars. The action here is recognition, not discounting. Do not spend margin on people who are already coming.

What not to do

Do not build a campaign for every segment. Pick one, act on it, measure whether they came back against a group you left alone, and only then do another.

Refresh it monthly

Recency changes constantly, which is what makes the analysis useful and what makes a stale version misleading.