ORDER BY 別の列を使用した SQL Server 2012 の group_concat

これがより速くなるかどうかはわかりませんが、purchasesで参加しない別のバージョンを次に示します。 STUFF() で 2 回 :

select customer_id,
  min(purchased_at) as first_purchased_at,
  stuff ((select ',' +  p2.product 
          from
          (
            select product, customer_id,
                ROW_NUMBER() over(partition by customer_id, product order by purchased_at) rn,
                ROW_NUMBER() over(partition by customer_id order by purchased_at) rnk   
            from purchases
          ) p2 
          where p2.customer_id = p1.customer_id
            and p2.rn = 1
          group by p2.product, rn, rnk
          order by rnk
          for XML PATH('') ), 1,1,'') AS all_purchased_products  
from purchases p1
group by customer_id;
 

デモによる SQL Fiddle を参照してください。

結果: