I'd try the following approach, and see where it leads you. Make a join from this table to itself, linked by customer. Require that the id for the product in the second copy of the table is higher than the one in the first copy. That way you'll eliminate links of products to themselves, as well as removing duplicates a->b and b->a. See what products are still very popular, and what combinations come out a lot.

From there, it's not to hard to identify those customers that bought both (or all three...) products.

Roughly tested SQL:

SELECT a.product as product_a, a.customer, b.product as product_b 
FROM Sales a INNER JOIN Sales b ON a.Customer = b.Customer
WHERE a.product < b.product
[download]

SELECT a.product as product_a, b.product as product_b, count(*) as pop
+ularity 
FROM Sales a INNER JOIN Sales b ON a.Customer = b.Customer WHERE a.pro
+duct < b.product
GROUP BY a.product, b.product
[download]

I would probably approach this with Graph::Base, more specifically, Graph::Undirected. Create a vertex for each customer and product. Add Edges between customer and products that they use. Then you could determine the favorite products by looking at the degree for each product node.

On the other hand, you could probably get what you want in the database world by using a 'group by' clause.

Just some thoughts.