Cross Selling Analysis

wirtcal

Hey everbody!

this is my first post here in Rapid Miner Community and i must to say how much this tool is powerful.

I used it to make some scores for a risk operation (money recovery by costumer debt and particularity) and my results was great...

now i want to make one thing beyond this score.. really i want to make one thing completly diferent:

I have one project in University to do and i want to Use the Miner. There is a Truck dealer as partner in this project that will provide its analitic sails database with their Parts and Services detais. I'm thinking to make one model to read the cross selling opportunities but i dont know how to get start.

The data has many fields {
- date,
- item code/service code
- number of invoice --> many itens (rows) has the same invoice number **** i think this is the "key", right?
- price of item/service
- Costumer name
- Costumer Document (ID)
- Costumer City
- Salesman/Techinal name
And others filds witch i wont to use at this time.
}

So...

- With this "mission".. what modeling tools i wiil need to use?...
Association an Item Set mining? Similariry Compuuataion?, Clustering? Regression? ...

I know my english isnt very good, but i hope you understand me

Thanks!

wirtcal

Wow!!

In the post before, the Salesman/Techninal is an important field to consider in model... cause maybe some guys has feelling to cros seeling and maybe other doesnt.

There are other 2 interesting fields:
- Truck Fabrication Year
- Veichle Name
- Vehicle Size (light, mediun, heavy, extra heavy)

Opportunity to another model:
- what are the parts or services that particular profile truck needs (probably)
Need help with this too!!

Thanks!

steffen

Hello wirtcal

Since this a general questions, here a collection of the thoughts pouring out of my head.

The first approach which came to my mind is of course classical cart-analysis. See for example the process rapidminer-Samples->Learner->12_AssociationRules and the operators under Modelling -> Association and Item-Set Mining

In order to do this you need to prepare your data in such a way, that ...
a) every row represents a cart, as far as I understand your data the id of this cart (and hence of the row) is the invoice-number, I would go even further, and use the "Customer name" as id and hence treat all the services the customer has ever purchased as if he/has bought them at once.
b) every attribute represents an item in the cart, hence you have an attribute for every item_code or service_code available

I have a feeling that there is an/some operator(s) for this kind of transformation, but frankly, I can't remember and/or I can't find them (came back to rapidminer only two weeks ago).

Regarding the other fields:
- date
- city
- salesperson
- truck descriptions

I'd give this fields a special treatment. E.g. you could check whether the type of items bought change due to the type of salesperson (by creating different association rules for different salesperson, if you have enough data). Or you could calculate the best selling-time for each item/service by transforming the date into weekday/month/season and calculate how often the items / services have been bought in this time.

Another idea: Instead of using the item_code for truck, you could generate a item_code for every truck feature combination, eg. t1 = "light_TRUCKNAME-A" etc.. I suggest to play around with the data to see what makes sense/improves performance/is possible due to the sparseness of the data.

This seems to be a combination of knowledge discovery and creating a cross-selling-model. Some insights you will find maybe trivial, but others may even create moments of "Oh, I did'nt know that" when presenting to your partner. An interesting project.

hope this was helpful,

greetings

Steffen

wirtcal

Hey Steffen! Thanks!! (very very much)

It was helpful!!

Today i will do some experiments, watch some videos, reexplore the samples and read doccumentation...

i'm anxious and excited with this project!

best regards!