Wednesday, May 1, 2013

Cross-Selling Part 3, the Mileage Chart


In this blog, Cross-Selling Part 3, we look at a simple report called “the mileage chart” that helps uncover cross-selling opportunities and simple cross-selling combinations.  We then describe how to prioritize and plan to make cross-selling work.

In Part 1, we looked at the human challenges of cross-selling successfully.  In Part 2, we looked at common data challenges. Later on in part 4, we will look at understanding and valuing customer behavior based on the number of categories they buy. 

To demonstrate the mileage chart, look at this example based on food, in this case a restaurant.  The example will be displayed in two ways, by counts and by percentage.  Here first is the example in percentage form:

Appetizers
Desserts
Drink
Entrée
Salad
Sandwich
Appetizers
100%
13%
45%
84%
56%
21%
Desserts
10%
100%
41%
88%
50%
20%
Drink
11%
13%
100%
89%
54%
20%
Entrée
8%
11%
35%
100%
47%
14%
Salad
9%
11%
39%
85%
100%
15%
Sandwich
11%
13%
44%
78%
45%
100%

The report reads across, and is generated one row at a time. It is similar to the “mileage charts” in old road maps that showed distance between cities.  The highlighted numbers represent all the product buyers of one category.  The other numbers in the row represent the percentage of buyers of that column’s product and also the highlighted product in that same row.

For example, of all the customers who bought appetizers, 100% bought appetizers.  Of that 100% that bought appetizers, 13% bought appetizers AND desserts, 45% bought appetizers AND drinks, 84% bought appetizers AND entrée, 56% bought appetizers AND salad, and 21% bought appetizers AND sandwich. We start over again with desserts where 10% of dessert buyers bought appetizers, and so on.

You can quickly tell from looking at the chart which categories most (or least) commonly are bought with each other category.  Similar in thought to the “people who bought this also liked that” you often see when shopping online.

You may have noticed that 13% of appetizer buyers bought desserts and only 10% of dessert buyers bought appetizers.  That is because the total number of appetizer buyers is different that the total number of dessert buyers.  To demonstrate that, here is the same data that produced the percentage chart in counts form: 
Appetizers
Desserts
Drink
Entrée
Salad
Sandwich
Appetizers
6,438
839
2,878
5,435
3,587
1,353
Desserts
839
8,388
3,435
7,358
4,195
1,637
Drink
2,878
3,435
26,861
23,959
14,606
5,373
Entrée
5,435
7,358
23,959
68,479
32,241
9,552
Salad
3,587
4,195
14,606
32,241
37,799
5,535
Sandwich
1,353
1,637
5,373
9,552
5,535
12,240

Now the highlighted numbers reveal the total number of customers buying in a category and the other boxes reveal the number of customers buying in the various 2-category combinations.  The most popular category is Entrée, with 68,479 customers.  The least popular is Appetizers with only 6,438 customers.

If we compare the two charts, we can see some differences in how we might develop offers by using both.  For example, in terms of total counts, more people buy entrée AND salad than any other combination.  But that is because entrée and salad are the two largest categories by customer count. 

While 85% of salad buyers purchased an entrée, on 47% of entrée buyers purchased a salad.  In terms of percentages, buyers of desserts and buyers of drinks were more likely to buy an entrée than salad buyers. 

Another way to look at the numbers is to find a cross-sell opportunity that has low cost in terms of cannibalization.  

For example, of the appetizer buyers, only 13% buy dessert.  If we can demonstrate that convincing a customer to add dessert will improve their Lifetime Value (we’ll discuss how we determine that in part 4) we have very little to lose in offering a dessert.  It will have little cannibalization of sales, and we can offer a product rather than a cash discount.

We can add to the mileage chart’s usefulness by looking at in in additional ways, for example:

1) Spending in one category versus the others.  I.e., of the people who spent a total of (let’s say) $10,000 in appetizers, they spent $500 in desserts, $3,250 in drinks, and so on. 

2) Compare 1st visit cross-over by itself to determine which combinations appeal to new buyers, and then create a separate repeat-buyer only chart.

3) Compare 1st visit to 2nd visit – in this case determine based on what people bought in each category during their first visit what they bought during their second visit.  That helps us to optimize new-customer follow-up offers.

4) Compare Lifetime Value by cross-over, to optimize simple packages that attract the best customers. We can look at retention rates, average order, and longer-term value.

While the mileage chart is limited to simple cross-tab relationships, it represents a great deal of information on one page.  With a minimum of explanation, it provides useful insights to product and product managers or salespeople, and is a useful tool in management meetings and team goal-setting situations.  The same format of report can be created in several different views in order to explain different opportunities, which saves time for managers by having reports that are all read and interpreted the same way.

In the next blog, Cross-Selling Part 4, we’ll look at how to value cross-selling across categories in terms of retention rates; average sales, and lifetime value.

Thursday, April 11, 2013

Cross-Selling Part 2, the Data Side


In Cross-Selling Part 1, we looked at the human side that impacts the challenges of cross-selling successfully.  In Cross-Selling Part 2, we will look specifically at the most common initial data challenges.

The most common initial challenges are that the two major pieces of data necessary to successfully cross-sell are not defined in the data:
1) Categories from a consumer’s perspective.
2) Complementary categories, which are things normally used together.

To define anything in the data in a meaningful way the definition must follow some criteria or set of rules.  Because categories relate to share of wallet, and share of wallet relates to what customers buy in total, we want to define the categories to fit the way customers use them.

It is the way customers use or think about items that determines how they should be categorized for cross-selling purposes. Not the way engineers, plant managers, or purchasing managers think about them.

For example, a component manufacturer making products used in fluid/hydraulic systems had well over 100,000 unique item numbers.  They produce components that go in everything from trucks to HVAC systems to home appliances.  But in the database, every item was considered unique, because it was engineered and built for a specific application.  There was no category shown in the data.

However, the product literature and company’s customer knowledge made it clear; they had five categories of items, which are:
1) Filters (things that cleaned the systems’ fluid)
2) Fittings (things that connected things together)
3) Sensors (things measuring pressure or flow)
4) Actuators (things that pushed fluid in the systems)
5) Valves (things that directed fluid in the systems)

By starting with key word searches in the product description fields, we managed to identify and apply the category to all but 500 or so of the over 100,000 items.  The rest we looked up by hand and applied manually.

While we got lucky that most products had a useful category word in their description, and the company already understood what the categories should be, it usually is not so easy.  For example, a grocer had categorized every item in the store with beef in it as beef.  That includes beef soup, beef bouillon, frozen steak dinners, all sorts of things. 

There was no quick way to categorize steak and hamburger – like the beef at the meat counter – in a convenient way.  Fixing a problem like that is not as easy as it sounds.  A grocery store can have 100,000 or more item numbers active in the system at any time, and many hundreds are added or replaced each week.  Without rules in place, a manual fix becomes unwieldy in a hurry.

Another company categorized products by the type of equipment they went into (regardless of what the component was) another categorized by the assembly plant products came from.  They all made sense at the time the data was entered, but they did not make sense when using the data to cross-sell customers.

Regardless, until the category is put in each item in the database, category in cross-selling is not useful to a database marketer.

The cross-selling approach using complementary categories is different than offering people something like what they have already bought before.  Cross-selling with complementary items is focused on selling additional products that are typically used together or go together. 

In Part 1, we used tires and wheels as an example.  Understanding and then building a database of which sorts of items goes together is crucial to building a cross-selling program.

For some companies, every category they sell fits together.  For the fluid/hydraulic component manufacturer’s products, every system required every one of their categories.  It was easy to tell what was left to cross-sell by determining what was missing from a customer’s orders or what categories were bought in small quantities relative to other categories.

Other industries have a challenge in determining what goes with what.  Hardware, for example, caters to different people with different needs.  Some customers buy mainly plumbing-related items, others buy mainly paint-related, others primarily electrical, and so on.  But within each group there is still a need for variety.  An electrician needs wires, boxes, and switches.  A paint buyer will also need brushes, rollers, tape, and cleaning supplies.

By using complementary categories, we can spot missing categories in what customers buy, and offer them things we know they must be buying elsewhere.

In the next blog, Cross-Selling Part 3, we’ll look at a simple report called “the mileage chart” that helps uncover cross-selling opportunities and discuss several different ways to use it.  We’ll describe how to not only uncover simple cross-selling combinations, but to prioritize and plan to make cross-selling work.

Tuesday, April 9, 2013

Cross-Selling Part 1, the Human and Organizational Challenges


One of the most consistent themes that run through customer data, regardless of industry, is that the loyal customers buy across multiple product lines. 

To flip that around, one of the strongest correlates for retention is how widely a customer buys across the product range.

There seem to be two major impediments to successful cross-selling.  It appears both human nature and technology conspire against it.

This blog discusses the human nature side, the next blog discusses challenges specific to data and technology.
Human nature hands us two challenges, the first human and the second organizational. 

The human challenge: Marketers tend to think of “loyalty” as retention.  Do customers return, how often, and how long between visits?  But should marketers think of loyalty that way?

Think of a grocery customer who buys fresh produce twice a week and nothing else.  They buy meat, dairy, canned and frozen goods elsewhere. They are “regular”, but are they “loyal”? 

If marketers instead defined “loyalty” as “Share-of-Wallet”, our example produce-only shopper no longer seems loyal.  On the other hand, the customer that buys everything they can from you is loyal, regardless of visit frequency or even overall spending.  Almost by definition, the most effective way to boost loyalty is often to cross-sell. 

A grocery shopper is a good example in another way, because a typical household shops 2, 3, or more stores regularly.  You might be the same, thinking of one store as a place to buy meat and fresh produce, but too expensive for canned goods and stock-up items.  A grocer can tell your share-of-wallet by looking at the quantity and frequency of items you buy, and the items you do not buy.

An additive factor also comes into play.  As a grocer cross-sells another item – thus increasing the number and range of items per sale – customer frequency goes UP not down.  Selling the extra item may cancel the need for a visit to a competitor and shift sales over to the successful cross-seller. In other words, higher average sales per visit mean more visits, not fewer.

Changing our ideas about loyalty so it is both retention and share-of-wallet pays dividends. 

The organizational challenge: The management structure of many organizations is designed in a way that prevents cross-selling. 

One reason is many product, brand, or department managers view each other as competition.  As a result, they will not develop packages, bundles, and offers across product lines even if that would most appeal to the consumer.  This is especially true when management structure and reporting is designed for managers to be judged based on comparison to one another. 

In other words, managers are judged based on sales compared against other departments, and not judged on sales in conjunction with other departments.

Another reason is products are often classified by source, department, or some engineering criteria that may be irrelevant to the consumer. A steel wheel and an aluminum wheel are very different to an engineer or in a manufacturing process, but they may be interchangeable to a consumer.  By the same token, a tire and a wheel are completely different to the manufacturer, but the consumer cannot use one without the other.  Tires and wheels are an example of complementary items – items that are used together – that make a natural package or bundle for consumers, but they may not be viewed as a package by the company.

The road to success runs through Taco Seasoning: Taco seasoning mix is a perfect example of an item that is part of cross-selling success.  A low-price, low-margin item, it tends to be purchased by more valuable customers, in larger average sales, than most any item in the grocery store.  Why is this so?

Because if someone buys taco seasoning mix, they will also buy multiple items across the store.  In produce, tomatoes, lettuce and onions.  Ground beef in the meat department.  Taco sauce in canned goods.  Taco shells in the bread area.  Taco seasoning mix is so inexpensive; people don’t think much about where they buy it.  But only loyal customers – customer who buy across multiple departments - are likely to pick it up, because they will also be buying all the other taco ingredients.

The idea of competition is a fallacy: Part of the thinking behind an organizational structure that sets managers to compete with one another is that managers have to fight to keep the other managers from taking sales from them.  But with cross-selling this is often a false threat.

As customers buy across additional categories, they tend to be more loyal and spend more not only overall but also in their original category.  A veterinarian that buys vaccines and then starts buying antibiotics is likely to spend more on vaccines than before.  A manufacturer that buys bolts and then starts buying washers is likely to buy more bolts than before.

In most cases, product managers couldn't steal from each other if they tried. If a product manager encourages customers to buy in other departments, they will see their own sales go up from those very same people. Not to mention sales sent to them by other departments, if other managers do the same.

Which takes us back to where we started.  Loyal customers -the best customers - buy across the product range.  

In the next blog, we’ll discuss how to look at data in cross-selling.  We’ll describe how to not only uncover hidden cross-selling opportunities, but to provide the database support management needs to make cross-selling work.


Building a New “Big Data” Set

We recently completed gathering a new (and very large) data set that will be beneficial for Agricultural marketers.  The data already gathered includes:

Soil type (and soil composition)
Slope
Erosion
Precise location
Watershed boundaries
Land usage

The data covers the entire US and all US territories.  Over 3,300 counties, with 10’s of thousands of observations defined by type, shape, and location per county.  It will become tremendously beneficial in improving targeting, cross-selling, and new-product growth in agriculture.

What separates this data set from previous tools, other than sheer size and completeness, is that it is BOTH Spatial AND Tabular. 

This means not only can you look at it on a map, we can analyze it with statistical tools.  That solves a major issue of complexity, since you can only see two or three dimensions on a map, which becomes the analytical limit for spatial data. But with tabular data and statistical software, we can utilize many variables and increase analytical power by hundreds or thousands of times.

New Data Means New Techniques
Data of this scope demands Community Analysis.  Community Analysis is a set of techniques designed to understand interactions, causations, and correlations in complex systems.  Through the process of Community Analysis segments are created that are homogeneous, substantial, actionable, differentiable, and accessible.  Different approaches have been developed in public health, genetic and molecular biology and database marketing that is being now be applied to agricultural markets. 

Represent the Environment
Community Analysis combined with “Big Data” allows a fuller representation of the environment.  Rather than comparing pre-selected similar fields, we can incorporate harvest monitor data compare similar sections of similar fields, areas of different types seeing similar conditions, and areas of the same type seeing different conditions.  Allowing each environmental variable to be available individually allows specific comparisons around environmental factors that could otherwise be ignored in field-to-field testing.

Find and Manage Cohorts
Cohorts are similar pieces of land or similar operations that share a common need or have a common benefit potential.  Cohorts can be defined by their environmental variables, operational characteristics, local agricultural practices, and many other factors.  Using “Big Data” and Community Analysis techniques similarities and difference can be defined that fit specific needs, goals, risks, or characteristics.

Refining Cohorts
Useful cohorts are segments that benefit from a specific message or action.  As a result, cohorts can be created on several levels.  Overall cohorts that segment entire farm operations or areas, product-driven cohorts that are prospect or re-sell targets for a specific product or service, and message-driven cohorts that are likely to respond to a specific type of appeal to suit their psychographics.  The process of refining cohorts occurs continuously as situations, offers, and response mechanisms change.

What This Means Operationally
Macro-level organizations create and target communications, offers, and support to constituents in a closed-loop fashion where the response (or lack or response) dictates the on-going cohort relationship.  Segmented communication is designed to fit individual needs, albeit the individuals are in specific groups, and those groups are large enough in terms of numbers and dollars to merit specific macro-level attention.  This delivers a one-to-one relationship feel while also allowing a segmented approach to communication that can be tracked back to ROI.

More to Come!
Now that gathering soils data is complete, we are moving on to gathering more environment-related data including variables such as stream flow and hourly weather.  We will also incorporate data from the three major censuses (US, AG, and Economic) and multiple meta-data points.  It is a big data set, and a big job, but it is a major step in progress to now have all soils!

For questions about use in US Agribusiness, contact Alan Weber - Alan@D2SG.net.  For questions about environmental or public policy use, contact Doug Ballou - doug@blue-window.org  If you are in Asia, contact HawZan Chong, HawZan@D2SG.net

Tuesday, March 26, 2013

RFA or RFM, Which is Better?


Monetary or Average Order – is RFA Better than RFM?

For decades direct marketers have used RFM as their primary segmentation tool, and the M (Monetary) traditionally stood for “Lifetime Spending to Date.”

Personal experience provided a couple of indications that Monetary wasn’t what it was cracked up to be.
  
Back in the 90’s, a catalog client asked for a list of best catalog customers, so he could send each one a thank you note.  Upon reviewing the list, it became obvious that about a third of the higher-dollar, recent, frequent customers had many, many orders.  They ordered small items (in this case, a paperback book) one-at-a-time, and spent hundreds of dollars.  But each individual sale was so small, given the cost of fulfillment, the catalog actually lost money on each order.  

The list of “best customers” was really about 2/3rds best and 1/3rd worst customers.

Then by sheer chance two different jewelry retailers brought us results where they had tested three segments; high-dollar coupons, low-dollar coupons, and a control group. Both had the same strange results. 

Upon reviewing the results, we found the high dollar offer worked best with high average order customers, followed by the control group (that got nothing), and then the low dollar offer.  For high average order customers, the low dollar offer actually suppressed response below the un-mailed control group.

The opposite – or maybe we should call it the same – happened on the low-dollar end.  The low dollar offer won, followed by the un-mailed control group, followed by the high-dollar offer that actually suppressed response.

For those in fundraising, this is hardly a shock.  The best ask is in the range, maybe a little higher, than the typical or most recent gift.  But for direct marketers targeting segments with common offers, does it really matter?

To try to answer the question, we went used a similar 20,000 name list like what we used for checking the 40-40-20 rule that’s described in another blog.  This time we used only the key-coded segment as the groups to create the analysis, and looked at both likelihood of response and dollars-per-name.  Dollars-per-name is defined as (response rate * average order) and it is a good indicator of the profitability of a segment.

RFM versus Response Rate gave us an R-Squared of 0.73, so it explained 73% of the difference in response among segments.

RFA versus Response Rate gave us an R-Squared of 0.80, so it explained 80% of the difference in response among segments. A clear winner!

If we look at the individual impact of Monetary versus the individual impact of Average Order inside a 95% statistical confidence window, we see Average Order is always positive.  That means the higher the average order, the higher the expected response rate.

Monetary, on the other hand, could have either a positive or slight negative impact on response rate, if we look at it in the same 95% confidence window. In other words, statistically speaking, Monetary did NOT improve our response prediction.  To confirm this we left Monetary out of a regression equation and just used Recency and Frequency.  R-Squared went UP to 0.76!

So far, that is just looking at response alone.  If we look at Dollars-per-Name, we tip the scales even more in favor of RFA.  RFM did well predicting Dollars per Name with an R-Squared of 0.79, but RFA did extremely well with an R-Squared of 0.91.  91% of segment to segment variation explained – that is very, very powerful.

Looking at slope in the regression, there was a wide range within 95% confidence for Monetary, varying by a factor of seven from low to high.  But the slope for Average Order varied only from 0.65 to 1.00, so it fell in a fairly narrow range. That is due largely to the high correlation in expected order amount with previous average order.

While the advantage of tailoring offers based on average order is obvious, even with the same offer to all segments Average Order statistically has a substantial advantage over Monetary in predicting both response rate and dollars per name. 

Which one do you use?