The segmentation that pays first is by value. Sort your customers by what they have actually spent, and by how much profit rather than revenue. Nearly every business discovers that a small proportion produces most of the margin, and that some customers are unprofitable once the time they consume is counted. That single sort tells you who to keep, who to seek more of, and occasionally who to let go.
The second useful cut is by how they found you. Customers arriving through referral, through search, and through advertising typically behave differently in ways that show up in conversion rate, price sensitivity, and how much support they need. If one source produces customers worth twice as much as another, that is a budgeting decision rather than an insight.
What makes this actionable rather than academic is doing something specific with each group. More of what worked for the profitable segment. Different messaging for the source that converts poorly. A price adjustment or a polite exit for the group costing more than it returns. Segmentation with no consequent decision is a spreadsheet exercise.
The prerequisite is recording the data as it happens, because none of it can be reconstructed. Source on every customer, revenue on every job, and a rough sense of time spent. Six months of that produces segments worth acting on. No amount of analysis produces them from records that were never kept.
Be careful about acting on segments that are too small to mean anything. Three customers sharing a characteristic is a coincidence often enough that building a strategy on it wastes effort. Wait until a pattern holds across enough customers that you would bet on it, which for most first year businesses is somewhere above ten, and in the meantime record the data rather than interpreting it. Premature segmentation produces confident decisions from noise, which is worse than no segmentation at all.
Record the data required for the segments you actually intend to use rather than everything you could capture, since a field nobody fills in reliably is worse than no field. Source, value, and what they bought covers most useful segmentation for a small business, and adding six more attributes generally produces a record with gaps rather than a richer picture.