Lesson 7 of 11

Customer segments: k-means clustering

Nakhla's marketing team sends the same offer to every customer, with poor results. Better to split them into similar segments and speak to each one differently. The catch: there are no ready answers (no “segment” column). That's where clustering comes in.

k-means is the best-known clustering algorithm. You choose the number of segments k, then it repeats two steps until nothing changes: (1) assign: each customer joins the nearest centre. (2) update: each centre moves to the average of its customers. Try it step by step:

The centres (✚) start in arbitrary places. Press “Assign customers”.

The model doesn't name the segments; you, the analyst, interpret them: “many visits, small basket” = daily grocery buyers; “few visits, big basket” = occasions and gifts. Then the decision: volume offers for the first, gift packaging for the second.

PYTHON

# customers: [visits per month, basket in tens of SAR]
customers = [[2, 4], [3, 5], [2, 5], [9, 6], [10, 4], [8, 7], [4, 21], [3, 19], [5, 23]]
centers = [[2, 5], [9, 5], [4, 20]]

def nearest(p):
    best, best_d = 0, None
    for i, c in enumerate(centers):
        d = ((p[0] - c[0]) ** 2 + (p[1] - c[1]) ** 2) ** 0.5
        if best_d is None or d < best_d:
            best, best_d = i, d
    return best

for p in customers:
    print(p, "-> segment", nearest(p) + 1)

Watch outIn marketing you'll often meet RFM: recency (last purchase), frequency and monetary value. These three make excellent k-means features, and you can build them with SQL from the orders table.

Customer segments: k-means clustering

Exercises