Retail business platform vs Retail analytics projects

One can run an analytical process in the retail industry as a project which requires expensive capital expenditure or on a retail business platform which requires a monthly operating expenditure. Which are the right analytical process to run in project mode and which are the right retail analytical processes to run in platform mode. This could depend upon a multitude of factors like
1. What is the frequency of the retail analytical process ? Weekly ? Monthly ? quarterly ? or real time
For example a real time recommendation to be made to a customer thru a contact center may well be coupled/embedded to the core application instead of routing it thru an offline business platform to derive cross sell propensity scores
2. Does the analytical process lend itself to competitive differentiation ? For ex cross sell models can be a source of competitive differentiation whereas a market mix model can leverage a platform
3. Is the data to be transferred under the purview of regulations ?
For example customer data and risk scores may need to satisfy European data protection act which may introduce additional complexities in the retail business platforms

Business platforms for Retail Analytics

Business platforms are picking up a lot of interest during these troubled times because they offer customers a way of optimizing cost for running their analytical processes. So what exactly is a business platform ? A Analytics business platform necessarily consists of the following

1. An analytical process which guides the step by step execution . Examples of analytical processes which have been "Platformed" are
- Market mix models
- Behavorial segmenation workflow models
- Campaign response models
- Forecasting market share or sales etc

2. A set of Key performance indicators which can be used to measure the health of the business process. Example : Forecasting process efficiency could be measured by MAPE and Forecast variance.

In operating a business model the client typically gives a slice of the data to be executed in the business platform. For example if one has to run a forecasting model information regarding past sales, promotion calendar for Gift offs, temporary price reductions, coupons would be required in addition to media spends . Once this data is integrated in the business platform, it can be trained and run to project future store sales with a certain level of confidence factor.
There are multiple advantages with the business platform model for running analytical processes
1) The customer need not invest in expensive statisticians, software and hardware
2) The customer has to pay only on a 'per run' basis which can be structured as an operating expense as a part of store operations instead of an expensive capital expenditure
3) There are business related best practices which are readily encapsulated into the business platform which can be leveraged.

I

Sales forecasting problem

1. What is the expected sales of my products across stores ?
2. What is the expected market share of my products ?
3. Do I need to increase inventory of certain products expecting a spike in sales of certain categories ?

All these are formulations of the problem related to forecasting a particular store related metric which would help store managers optimize their operations especially inventory planning and negotiating stocks with vendors.

Lets take for example forecasting sales of a certain strategic product. Sales is driven by 7 key factors

1. Consumer promotions like Gift off, Temporary price reductions, 10% extra
2. Media spend by company owning that product
3. New launch products ( which could potentially cannibalize it )
4. Seasonality of the product ( Weather could influence sales or festive seasons could influence sales )
5. Competitive media spend
6. Competitive product pricing

What are some of the forecasting techniques which could be used to predict sales ?
1. Exponential smoothing
2. ARIMA
3. Holtwinters model
4. Regression

What packages are available in the market to forecast ?
1. Dmantra from Oracle
2. Forecast server from SAS
3. APO from SAP

We can step thru some of the best practices which are used for forecasting in the next blog