Basically, you'd need to segment (clusterize) your customers in similarity groups (which you assume have similar buying behavior) and for each cluster, you then run the basket analysis (apriori) algorithm.
HANA has both the apriori and clustering algorithms (such as K-Means) natively in PAL (Predictive Analysis Library).
You can build a data model on top of your sales data to run the k-means and apriori algos (an apriori run for each different cluster) and store the results (confidence, lift etc.) in a results table. Then, based on that result, you can create a calc view with a logic like: for each customer, based on its top 5 bought products (or more recent purchases, depends on the business characteristics), go to the apriori results table, see which products are more lifted for each of the top 5 products the customer already buys and suggest them as an additional offer).
That calc view would be something you don't need to persist, you can have it as a logical model that runs whenever a customer accesses your site, for example, and provide the calc view results through an OData service. The apriori algo itself can be run to update the confidence/lift values. How often you need to run it, depends again on the nature of your business: groceries, probably once a month would be enough; for fashion stores, maybe every week or every few days; for online stores such as eBay or Amazon, maybe you need to have it everyday or even more than once per day.
We developed a similar scenario for a CP company (provides the goods to groceries stores) in Brazil.
Best,
Henrique.