Analytical Engine

One of the most common problems of retail sector is pricing a product. There are numerous similar products lying in the shelves that are at different life stages and have complex interrelationship with other products & customers. Providing an optimal price of an SKU in order to extract maximum customer surplus is a challenge in itself.

In our recent stint with a Europe based retail chain for electronics goods, we helped our client use analytics to better price cell-phone accessories. A lot of advanced sophisticated pricing models can be built and deployed; however, when efforts and incremental value of such models are weighed against the approach of using elasticity and bundling, the latter option seems to be an optimal choice to get some quick hits for the business. For the uninitiated, here is a quick review of elasticity and bundling:

— Price elasticity of demand: The measure of responsivenesses in the quantity demanded for a commodity as a result of change in price of the same commodity.
— Product bundling: A marketing strategy that involves offering several products for sale as one combined product

Price elasticity approach: We calculated the price elasticity, and accessories that are having extremely high or extremely low elasticity along with reasonable business impact (sales volume or margins or life-stage) are picked for analysis. With the help of gross margins, we came up with the best price change for such accessories. In order to evaluate the financial impact, we built a business case to find out the potential revenue impact of such price changes.

Product bundling approach: A basic market basket analysis can help find products customers like to buy together. However, bundle needs to be evaluated for elasticity as well as likelihood of success before taking it to the customer. In order to increase the sale of some of the underperforming accessories, such accessories can be strategically bundled with highly elastic products. Gross margin of the bundle can be used for driving the price of the bundle.

Both these methods fetch actionable recommendations which can be tested in a very short time-frame in a test environment. One of our recommendations led to an additional £100K revenue impact per week for selected SKUs. A similar approach can be used for other items to better understand the nature and the monetary potential of the products lying in a store.

In addition to a very technical perception as a method of predicting a continuous variable, Linear Regression is also, and equally importantly, an analytical method to understand the business drivers of dependent (which is being predicted) variable. Because of its inherent nature, regression has been used to understand how a variable is influenced by the interaction of multiple variables. In this post, we will talk about a couple of such scenarios where we used linear regression in the context of strategy formulation.

Casino – Slot Machine Optimization

In our previous post we talked about how a simple LFM (Latency, Frequency, & Monetary) segmentation can be used for customer management across casino industry. In this post, we shall talk about how linear regression can be used for floor layout optimization of casino.

Recently, we worked on an engagement for one of the largest casinos in the world, where the casino wasn’t able to capitalize on the abundance of slot machines; moreover, the machines were spread across the floor without much thought or analysis behind why a machine should be placed at a certain location – and hence, were not tapping the right potential of either the machines or the location. However, even from a gut instinct, a lot of us who have been to a casino, know that most of the players have their own preferences when it comes to picking a machine – be it the corner vs. central isles, the red machines vs. the blue machines, the spinning wheels vs. the talking genies, and so on.

A simple profiling of slot machine performance across various variables like denomination of machine, jackpot, type of game, location of machine, etc. can fetch interesting results. However, all these variables need to be evaluated simultaneously to create a holistic picture.

Even though a lot of optimization concepts can be applied, a linear regression can be used to determine the drivers of performance and better understand how these drivers interact with each other. The drivers of performance help to determine an optimal floor layout, so that maximum returns can be obtained from the given slot machines on floor. We found that certain themes with certain colors drive the performance while for some types of machines it was progressive nature of machine payouts that drives money into them. Certain machines if placed around restaurants seem to do well, while others if placed around some attraction tend to perform well.

Sales Territory Prioritization - Distribution Performance:

For companies with retail distribution, it is critical that revenue is maximized while expanding business in hitherto untargeted geographies, and therein lies a need to prioritize geographies based on their expected worth. Such a strategy requires analysis for not just one factor but for multiple factors and an area where linear regressions is useful.

While working for a Financial Services firm, we helped our client prioritize territories where they did not have any presence. Based on the current product offering and existing market attributes, a regression model was built and untargeted territories were scored (prioritized). This not only helped to reduce the focus but also helped in designing a quick and efficient marketing strategy that targets the ‘right customers’ of such territories.

We found that, for some territories the major driver of revenue was the abundance of a certain segment of population, while for other territories, the key driver was the strong presence of a particular type of business in that territory. These drivers helped us tailor the marketing strategy in a customized manner by understanding the ‘needs’ of each territory.

This is not all by any means. At Diamond, we have effectively used regression in launching new products, managing churn, managing portfolios, etc. to not only come with a target(scored) population but also come up with precise recommendations based on the detailed study of drivers.

(There is no dictionary definition of P2P process, but often P2P is defined as a generic term used to encompass the payment for goods and services: The basic process of raising a purchase order, receiving the goods and paying the invoice is often put under P2P process .)

In a recently finished engagement with a manufacturing major, we leveraged Diamond’s procurement analytics solution to help the client identify dollar leakages and potential saving opportunities across their P2P process. This engagement had all the challenges associated with a large manufacturing setup viz. inbound and outbound from plants located across various locations, multiple vendors supplying the same raw material, a complex distribution and lack of unified IT systems. The list below can give you a feel of some the issues we identified –
1) Price variation/anomalies:
a. Procurement organization paying different price for the same raw material within the same time frame across different vendors
b. Price changes for certain non-contract items being anomalously higher (even in an increased demand scenario, where the organization might have had a chance to negotiate better prices)
2) Contract Adherence:
a. Shippers invoicing higher than the negotiated rates in purchase orders
b. Vendors charging prices which are not consistent with contract-negotiated prices
3) Freight Charge Variations:a. Different freight charges being charged for the same kind of delivery across supplies
b. Freight charges being consistently higher than industry standards
c. Variation in the freight charges for equal shipments from the same vendor Payment
4) Payment & Order Schedule:a. Payments made significantly prior to the negotiated deadlines leading to loss of revenue
b. Sub optimal volume discounts because of fragmented orders

Organizations are investing a lot of money and effort in IT solutions, designing data warehouse(s), and data collection methodology. These efforts, though, aimed at streamlining decision making processes, continue to exist in silos and are not creating the desired impact for the business. To help our client tackle this challenge, some of the things we tried to focus on, as part of the engagement were –

• Master Data Management: Integrating different data sources (General Ledger, Purchase Orders, Invoice, Vendor information etc.) and incorporating process information to come up with the final form of an analysis data mart
• Basic Data Profiling: Performing basic data validation, enriching/ cleansing and classifying required data, instead of building highly complex spend analyzers around existing solutions; The essence of all analytics done by a team should be the business benefit that can be derived, and not intellectual gratification
• Variance Reports: Leveraging the analysis data mart to create variance reports that help identify the variation in price/freights for a certain procured material
• Dashboards/Segmentation of the invoices : Developing a segmentation tool that helps identify where the money is invested/spent; Having a sense of where the dollars are going always helps businesses prioritize their spend rightly
• Payment segmentation: Identifying anomalies between the POs and Invoices, and keeping track of vendor issues such as reject rate/ quality, etc. can help renegotiate contracts, create appropriate procurement and penalty mechanisms for ongoing cost reduction
As can be seen, a significant part of our approach was about getting the basics right, and instituting basic analytics processes in place. Before moving on to advanced statistical concepts/techniques, we demonstrated the value of analytics through simplified dashboard and analytics reports,and showcased a potential saving of multimillion dollar to the client

Ian Ayres, in his book Super Crunchers, questions decisions made through intuition and advocates decisions made by optimal utilization of data. In the book, Ayres gives examples of how data driven decision making is impacting businesses, education, sports, government, etc. One such industry where a lot of data is collected but not necessarily used in decision making is the gaming industry. Despite Harrah’s being an analytics poster child, not a lot of gaming companies leverage analytics to its potential.

Casinos collect revenue and customer data from a variety of sources, such as player cards, slot machines, and gaming tables. They can also gather information from non-gaming sources: through call centers, surveys, hotels, restaurants, and events. Harrah’s is a pioneer in collecting detailed data on its customers’ activities. Harrah’s team sliced that data into finer segments ”identifying unique customer groups and targeting each group with pitch-perfect marketing strategies.” The results -  tremendous increase in customer loyalty that has turned Harrah’s from a relatively unremarkable player into an industry giant.

Recently, we worked on an engagement for one of the large gaming companies in the world, where the casino wasn’t able to capitalize the rich data it had on its players and slot machines. Direct marketing campaigns and slot machine floor layout optimization, were two big initiatives that were performed but without leveraging the rich information in the data. An analytics driven approach was needed to identify high potential players and optimize slot machine layout. In this post, we will talk about the approach we used to segment players and identify changing behavior patterns of players.

Though a lot of advanced concepts of customer analytics can be applied here, we used a basic RFM (Recency, Frequency, and Monetary) segmentation to effectively target players. RFM was tweaked to LFM (Latency, Frequency and Monetary) for a simple reason- unlike a website or a grocery store, where a person’s recent visit shows that he/she is likely to visit again in near future, a casino trip happens after some interval. So it makes more sense to calculate whether days between the consecutive trips are changing significantly. In an LFM cube the cutoff of each cell of cube can be determined using basic business rules based on frequency distributions.

One of the KPI that was tracked was average spend per trip, and due to this, some players with higher number of trips and medium spend are undervalued and under-targeted. The LFM cube gives us the visibility into such customers and helps us evaluate the true potential of such players. Such a cube can be run at regular intervals to capture the movement of players between the cells, and such movement is a good predictor of changing behaviors and can be an effective lever for targeting marketing campaigns.

In next post we shall talk about how analytics can be used around slot machines and how such inputs can be used to design the floor layout in a better manner.

A few weeks ago, Shantanu wrote on recommendation engines and how user feedback and ratings can be a part of recommendations you provide to your customers.  But if you have ever looked through user recommendations while shopping online for a product, stock, or movie, you know that they aren’t all helpful.  Ideally, user ratings would accurately represent the population, but not all feedback is created equal, and there are some inherent challenges in these systems:

1. Not everyone will rate.  People may read the ratings when shopping for an item, but they won’t always come back to rate.  Unless a site offers an incentive for rating a product, a customer’s only real incentive for doing so is to talk about how much they love or hate it; moderates may be under-represented.
2. Ratings will be biased.  People’s individual biases produce variances in ratings, even if strict guidelines (think about employee performance reviews) are presented.  In addition, new raters tend to rate high.  Their average ratings decrease over time as they rate more items, presumably because they are exposed to more items and have a better sense of an item’s value relative to alternatives.
3. Ratings are averaged, masking the underlying data. Because people often only rate items they feel strongly about (love it or hate it), and an average of those extreme ratings may not truly represent the actual sentiment surrounding a product.  For example, a review of rankings on Amazon.com revealed that the reviews for the majority of the products have an asymmetric bimodal distribution. For these products, the mean of the online product reviews does not necessarily reveal the product’s true quality, resulting in misleading conclusions about the product’s future success.  In addition, established products are at a disadvantage against new ones.  Consider a product that has received five rankings, four and one giving it an average rating of 4.8. If a new product enters the space and receives one rating, it will be ranked higher than the other simply because it has fewer ratings. 
4. Ratings may be false.  Take the case of the Whole Foods CEO who posted disparaging comments about a competitor on a message board while talking up his own company, later stating “Sometimes I simply played ˜devil’s advocate” for the sheer fun of arguing.  Other visitors may post false comments to artificially affect the rating, and these are not always removed from the calculation.

As more products are marketed and sold online, feedback-based ranking systems are increasingly common components.  Amazon and eBay were among the first to use visitor ratings to rank products and sellers, and in February, Yahoo launched its Buzz service, which asks readers to click on their favorite stories, then uses those ratings to determine the most popular articles on the web.

Why is this important?  Because accurate product ratings help predict that product’s success, and higher product ratings lead to more sales - or, as Allen & Appelcline state, the value of individual items (most frequently goods) rise or fall based upon the largely subjective judgment of individual users. So what can you do to ensure the rating system on your own website accurately reflects your customers’ views and the value of your product?

We’ve seen some thought around using analytical techniques and Bayesian mathematics to create better product rankings.  Some of the solutions explored include:

1. Adjusting ratings based on known biases.  Since some people rate higher or lower than others, one approach is to assign users a “User Optimism” value based on their rating history, and adjust the product’s overall rating based on the raters’ optimism value.  Another approach is to remove all ratings from people who have only rated 1 or 2 items, helping to eliminate new-rater optimism or “drive-by” fraud.
2. Weighting a product’s ratings based on the number of ratings received.  When a product has a small number of ratings, these ratings should count less than those for a product rated many times.  To achieve this, you can add a “magic value” into the algorithm that calculates a product’s average ratings.  This “magic value” brings products with few ratings closer to the average ratings of all products, then reduces its effect as more ratings are received, allowing established products’ ratings to float freely and more closely reflect the average of its ratings.
3. Adjusting the algorithm to account for bimodal distributions.  To account for products that only have ratings at the extremes, one approach is to use a dual-point estimation model to more accurately reflect customers’ views of the product and predict the product’s success.
4. Analyzing a customer’s rating history for fraud.  Someone who rates several products on your site should have a predictable rating pattern.  You can analyze individual customers’ accounts to identify and remove anomalies that might skew product ratings.
5. Encouraging customers to leave text-based feedback.  When a person takes the time to write out a product review, and knows that her name will be attached to it, she generally does a better job in her rating.  These reviews tend to be closer to the average than those without.

Although your customers’ product ratings may be imperfect, they can still yield insights and value with the application of a good set of analytical techniques.

We were working for a large Fortune 500 company to assess the performance of its sales team’s incentive program. A two tier incentive structure was in place for the sales force.
·        The incentive was directly proportional to the revenue generated by the salesperson
·        The salesperson generating revenue beyond a certain cutoff was given extra perks and monetary rewards
Prima facie this incentive structure made sense but when we analyzed it carefully, an array of systemic problems was observed. We mention few of the issues with potential ideas to tackle them
1)      Setting and maintaining goals – The incentive revenue cutoffs were predefined, static and the senior stakeholders had the power to override them. The cutoffs and tier allocations were not reviewed on a regular basis, making the tiers passive for the candidates who are performing well over a period of time. 
The key is to better understand the performance drivers of the sales force by  defining and measuring appropriate metrics  (using correlations, principle components, and factor analysis). Appropriately segmenting the sales force, defining the segment cutoffs using historical data and then reviewing them on a periodic basis is critical to achieve a better return from an incentive program.
 

2)      Quality Vs. Quantity of Money - Sales force were rewarded based on the volume (revenue) of the business and not the quality of the business (margins). Targeting the wrong prospects by the sales representatives sometimes led to value negative customers.
Using lifetime value calculation (LTV) of customers instead of absolute number of customers or amount of revenue, as the performance metric will solve this problem. Most organizations do understand the theory but do not use the metric, because calculating customer lifetime value is tricky and the data is not always available to do so accurately. In our opinion, it is better to have an incomplete understanding of the lifetime value (using averages and proxies for certain aggregated costs) rather than not making an attempt to calculate and use the metric. A flawed metric is better than a wrong metric.
 

3)      Poor and inadequate data - Data to track performance and understand the drivers of success for sales representatives was not readily available. The data was collected was mostly in standalone excel sheets. They were manually reported and coded making it difficult to link the independent tables.
A culture of measurement fosters investment in better data capturing systems. As analytics is used to calibrate the incentive structures on periodic basis, investment in centralized systems to standardize the data collection process will reduce analysis cycle time and increase accuracy of the analysis.

An integrated view of the incentive program helps in incentivizing the right people for the right reasons with the right benefits. Analytics is part of the answer. However, as emphasized by most of our projects, we believe that a sustainable solution is a combination of a robust understanding of business processes, technology infrastructure and analytical techniques.
�

A recent article in WSJ talks about how offline segmentation schemes and data is being used for online targeting of advertisements. The sidebar of the article mentions how Acxiom executes on this strategy

1. Acxiom has accumulated a database of about 133 million households and divided it into 70 demographic and lifestyle clusters based on information available from public sources.

2. A person gives one of Acxiom’s Web partners his address by buying something, filling out a survey or completing a contest form on one of the sites.

3. In an eyeblink, Acxiom checks the address against its database and places a “cookie,” or small piece of tracking software, embedded with a code for that person’s demographic and behavioral cluster on his computer hard drive.

4. When the person visits an Acxiom partner site in the future, Acxiom can use that code to determine which ads to show.

5. Through another cookie, Acxiom tracks what consumers do on partner Web sites.

At a client engagement we also have utilized a similar strategy (however without using any personally identifiable information like email address) to understand how segments of customers consume content on a website (see image below). This information is used to develop and personalize content on the website and can also inform site layout and design.

What we are witnessing is that companies are trying to bridge the worlds of marketing and web analytics. As we have previously mentioned, third party segmentation schemes which are so pervasive in the marketing world will increasingly become the translation layer between the two worlds.

What once was a good ole’ boy sport that relied on tribal knowledge buried deep inside a crew chiefs’ head is becoming a hi-tech, data rich, and analysis driven sport. More and more teams, crew chiefs, engineers, and even pit crews rely heavily on granular data elements that are used to feed complex models to extract the slightest bit of knowledge that can be used to save thousandths of a second on the track or in the pit. The sport of NASCAR is no longer defined by car lengths that won the race rather it’s becoming a sport defined by hundredths and in some cases thousandths of a second that decides between the winner and the first loser, a.k.a. “second place”.

After having an interesting and thought provoking conversation with Ken Brown, former engineering director at Dale Earnhardt Inc. and lead engineer for GM’s Corvette Racing program, I found myself drawing parallelisms between what I do in marketing science and statistics at Diamond for our clients and what NASCAR teams are doing to extract value from information. Not in the ROI sense but in the thousandths of a second sense. However, this in the end does translate into ROI for the thousands of sponsors that make the sport go round and round, no pun intended!

According to Ken, granular data elements such as wedge, nose weight, spring rates, tire pressures, camber, and gear ratios (just to name a few), all play a critical role in determining the how good a car will perform at a particular race. However, the key decisions are often made long before the car arrives at the race track. Essentially through track testing, wind tunnels, shaker rigs, and simulation exercises, the engineer and crew chief can decide which and how much to change each variable to make their car the best on the track. In marketing science I call these things covariates!

So how are teams doing this? Although the details of the process are unique to each team, the general process that goes into setting building and setting up a stock car is as follows:

1. Know the track (flat track, road course, banked)

2. Define/identify the significant set up parameters (or covariates) relevant to that track

2. Gather test data (prior race data, testing, wind tunnel, etc) and validate its quality

3. Analyze preliminary data via data visualization techniques (histograms, pareto charts, and cross-tabs)

4. Build models (e.g., DOE, Pareto Analysis)

5. Analyze results of simulation techniques

6. Set-up car for the race

7. Test and refine!

This process that is followed by the top teams is not unlike the process we follow in our information and analytics practice at Diamond when trying to identify the key variables that will attract and retain the customers or segments a business is going after. Moreover, the quality of the testing data is one of the most critical components to the entire process.

So the next time you see a professional stock car race, it’s important to know that before there was the driver in victory lane, there was a data geek in the background making it all possible.

If you want to develop an analytics capability within your organization – should you be talking with business intelligence software vendors or data analytics service providers?

Jeff Kaplan in a recent article in DM Review makes a case for the latter.

Delivering predictive analytics is not a trivial exercise. It requires the skills of being able to map the marketing goals to the appropriate predictive algorithms, perform data hygiene and transformations, build models and test the results.

Moreover, implementing predictive analytics requires the combination of three distinct skill sets:

1. database technology
2. data mining and
3. marketing domain knowledge

He goes on to identify executive support and expertise of the partner (in understanding data, statistical modeling expertise and database skills) as the two success factors of such initiatives.

My take: The principle of service before software in analytics makes sense because of the complexities that Jeff identifies (see related posts on the topic here and here). A service first approach also helps to rapidly identify the specific areas within the organization where analytics will be a quick hit due to data availability, quality, and acceptance of analytics within the group.

Another aspect which we have found to be critical in the success of analytics in the organization is being able to generate the right set of hypotheses for the analysis. The hypothesis driven approach helps an organization to focus on the problems that matter. A service provider, from prior experience, can help to generate a long list of hypotheses. However, in our experience it is critical to identify and recruit savvy individuals from within the organization, who have deep understanding of the day-to-day business operations and are passionate about improving the status quo with better analysis of data. They are able to pick out the specific hypothesis and analysis that can help the group to make better decisions. We call them ‘insight managers’. Normally, they become the analytics champion within the group and are vital to its broader acceptance.

Diamond has developed a model to deploy analytics within an organization and ‘insight managers’ play a major role in it.

Heinz recently reported a very successful year driven in large parts by its pricing strategy. Increasing prices and simultaneously gaining market share is the holy grail of Consumer Packaged Goods (CPG) companies. A 2005 report from GMA, Nielson and McKinsey identifies some of the salient points around sales, pricing activities of ‘winning’ companies (which they define as companies who not only increased its pricing compared to its category peers, but grew their segment dollar share at the same time)

Setting Price:

• Manage pricing regionally (and some at the market area, micro-market, or even store level) to reflect local variations in price sensitivity that arise due to differences in consumer preferences, competitive intensity and retailer dynamics.
• Use more price elasticity analysis to set the price and integrate consumer insights into pricing more aggressively
• Tailor brand-pack assortment to the needs of a particular region, channel or store format

Realizing the Pricing Increase:

• Use a profit-based rationale to secure price increases and also combine this tactic with a strong understanding of consumer preferences, shopping behavior, and price elasticity.
• Appoint a single head of pricing who collaborates with top leadership and critical functions on both the pricing strategy and its execution. An essential part of that person’s task is then translating the strategy into sales force guidelines that help avoid conflicts between profit and volume goals.

In short, the winning companies are using analytics to drive their pricing decisions and have gone beyond the simple volume based pricing strategies that is the norm in the industry. Profitability, consumer insight, channel management and price elasticity are all areas which require data analytics capabilities. The end of the ERP implementation cycle in the industry means that companies will start developing predictive analytics capabilities and try to use information and analytics as a sustainable competitive advantage.