Use NSSO consumption data. First select sample regions. Then collect the consumption data for those regions. You are ideally required to collect long term panel data. Then use these data to forecast the consumption pattern. 

Hi there,

You should also consider logistic regression if you are trying to estimate probabilities...

Regards!

Jitesh - Here are some inputs.

1. Define the category into broad groups - Shopping, Commodities, Entertainment, Travel, Medical etc.

2. Now, define the time period where you want to track. This is bit tricky. It can be in days, weeks, months, quarters, half-yearly, yearly. You can select the period based on categories in (1). Try to keep the categories and period under one umbrella. For example, if it travel expense - it is likely to be once in 6 months vs shopping at groceries which may be weekly or bi-weekly. Hence you need to design the category and period smartly.

3. Track general events (Holidays, seasons) with another table along with the period in (2).

4. The best analysis for this would be Regression analysis. 

5. You can prediction using LR. Weka tool has many ready made prediction models that can be used. You can also find Coursera "Machine Learning" where first few classes are on predicting house prices for a given market.

All the best !!!

Thanks all for your responses .

Kiran, this are really good inputs. Can you share any material for reference?

The data level that  you need depends on the type of campaigns and offers that you want to do. For a one-to-one marketing, I like to go into a deeper analysis of the purchase patterns. Initially you should define the broad groups as it was suggested. Then review the way you receive the transactions and proceed to add different hierarchies. For example, assume 2 transactions from Starbucks:

card 12345,  date-of-transaction1,  amount1, Starbucks 34, Main Street 200, Zip 10000

card 12345, date-of-transaction2, amount2, Starbucks 89, Broad Street 56, Zip 34000

check if your credit card processor has the relationship that Starbucks 34 and Starbucks 89 both belong to Starbucks. Otherwise it would be helpful if you normalize those transactions indicating this fact. Then do the analysis on the Starbucks level.

The number of credit and debit card transactions can be very large, but remember 1 year of data means only 1 Christmas, 1 Easter, 1 Long Holiday, etc. You may want to have at least 3 to 4 occurrences of events. 

One technique that you can use if market basket analysis and clustering. You may want to segment your customers and then perform the predictive analytics for the next best offer or product. Also a gap analysis.

Hi Jitesh

I would identify which segment this customer belongs to based on purchasing channel. 

Then I would identify how long a customer takes on average to buy product A, product B, product C, etc . That is, I would compute the Latency (average time) between purchases.

I would also look at the product life cycle buying behaviour for this customer channel segment based on past data.

I will identify which products this customer has bought and which other customers are similar to this customer. Based on what similar customers have bought in the past and this customer has not yet bought, and based on what similar customers have already bought with the highest frequency, I will recommend the next best product for this customer to buy.

Kind Regards

Carol

Concentrate first on the transaction level of the debit and credit cards. You should have a detailed list of transactions indicating the place where the charge was made. You wil need to do data preprocessing since you may find something like: ChainX location 2 timestamp amount, ChainX location5 timestamp. Depending on the analysis that you want to perform, you should transform ChainX, for example: case 1- ChainX, case 2 - ChainX Nothern States, ChainX Central States, etc. You may append season, weather, holidays, etc., p.e., ChainX-SuperBowl, ChainX-NewYear. As suggested above, try market basket/associations, clustering, you may also try to use decision trees. There are several types of analysis like Customer Y is similar to customers in cluster Z, so offer the products in cluster Z that Customer Y is missing. Other types of analysis could be McDon is the number 1 restaurant in the central part of the country, but in the nothern part is is CarlsJ. In the case of credit cards, you can expand your analysis based on the type of card, the credit limit, the number of cards that the customer has with other institutions (using Credit Bureau data), etc. This way of focusing customer behavior will enable you to send more personalized offers. Be careful when you have businesses that centralize transactions, for example all the transactions for a movie theatre chain are located at location X. In this case you may have to negotiate with the move theatre chain so they can give the purchase detail. Hope this helps.