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

Using T test to understand shelf placement decisions

Lets consider a simple scenario.

Say ABC retail is setting up a new store in Frazer town area

Pre store launch catchment area/competitive store analysis helped in deciding store format and product mix to display

The store in Frazer town, the following categories are relevant for the consumer ( Mangloreans, Muslims,Goans )
Non veg food items
Hair care
Basmati Rice
Atta
Kerala Parottas

The store manager of Frazer town branch wants to optimize the 12000 sq ft of shelf space he has

He wants to know which “Hot” shelves and he has a few hypothesis based on experience which he wants to test

A/B Multivariate testing can come handy to discern which shelf combination maximises revenue and minimize inventory carrying cost Or if the product is moving slow ( it becomes a slow moving item and a promotion /advertising may be required to stimulate sales )

HOW DOES A STORE MANAGER DECIDE SHELF PLACEMENTS TO MAXIMIZE STORE REVENUE AND MINIMIZE SHELF SPACE INVENTORY ?

Shelf space optimisation framework

Product inventory and shelf space are a retailers most precious resource
If there is too much inventory it affects store profitability
If there is too less inventory it affects store sales
If the product is not kept in the right shelf it could affect consumers ability to find it
It is estimated that assortment and shelf optimization can lead to 7-15 % improvement in store sales and gross margin
Non aligned shelves could result in
Sub optimal customer experience ( as he could not find the product he/she is looking for )
Increased inventory carrying cost ( as the yield per square feet of retail space is low )

There needs to be a balance between breadth of a category kept and depth of a product category kept
Ex: In retail outlets do they stock breadth : Rice, Soap,Shampoo,Electronics,Stationary or breadth ( Anapoorna rice, taaza rice et )
In a jewellery store do they stock breadth : Diamonds, Gold,Silver or depth : Diamond ring, Gold chains etc

The questions defining the shelf space optimisation problem are as follows

1.Which are the top 5 products which need an increase in share of shelf ?
2. Which are the top 5 products whose shelf space has to shrink ?
3. How much additional shelf should I need for products which need to expand ?
4. How much shelf space should we shrink the space of the products occupying too much shelf space ?
5. Is their a number which quantifies the degree of misalignment ? How does one interpret this KPI ?

Using structural equations modeling to understand store performance drivers

Structural equations modeling is an extremely good technique to model multiple cause and effect simultaneously. Take for example we want to trace the causal chain from foot fall to conversion to spend dispersion to monthly sales turnover to store profitability. How do we get to see the complete cause and effect chain. Techniques like regression etc cannot handle one outcome variable recursively being a causal variable

Co-relating shopper sentiments to footfall and basket size

Statistics collected by Media agencies suggest that Teenagers are spending more time on the Web than watching TV. This is a huge inflection point as web has replaced TV as a more engaging channel. And within Online channel , Blogging and Online videos ( youtube etc ) seem to be most engaging activity. What that means is that it is important for retailers to track if shoppers express sentiment about the instore experience or product attributes online ? There are 2 kinds of scenarios which can be envisioned here.

Scenario-1 : When shoppers are expressing about their instore experience on http://www.yelp.com/ or http://www.mouthshut.com/ or http://www.eopinions.com/. But the sentiment volume has not reached a threshold where it has started influencing footfall, basket size and revenue per shopper.

Scenario-2 : The volume of sentiment expressed on online platform has reached a critical stage where more shoppers are coming to the store or the number of shoppers / basket size has decreased.

What this means is to that the retailer needs to have a framework which can keep track of the buzz velocity online and track in real time the effect of buzz velocity on instore footfall and basket size.