The Marketing Automation Corner - Bringing Science & Technology to Marketing

As a former economist and data miner, I tend to always look at marketing as an art that is driven by science and technology. My name is Jim Stafford, and I have worked for over 13 years in the enterprise software world in product management, product marketing and sales engineering – all supporting Marketing Automation (MA). While many posts will be analytical in nature, my exposure to all facets of MA will drive a wide variety of posts. We'll cover topics that include: multichannel marketing, data mining, contact optimization, email optimization, and more.  I hope you enjoy them and join me in a dialog that will help all of us learn from one another and grow.

Like many other terms, Marketing Optimization (MO) can hold different meanings for different marketers.  For online marketers, it means developing marketing campaigns that do A/B testing on emails and microsite pages to see which one’s generate the most opens, click-thrus, conversions, etc.  Those emails and web pages that underperform are eliminated in favor of the best performers.  For other marketers, MO means optimizing your communication strategy across campaigns and marketing channels to improve response, customer loyalty and profit.  It is the later meaning that this article will focus on.

Initially, optimization was used as a way to mathematically determine the optimal allocation of scarce resources. The concept has been borrowed by business analysts to aid decision-making.  Optimization has been used in the areas of the manufacturing supply chain, airline revenue yields, and financial investment risk assessment. More recently, the concept is being adopted by marketing.

Every day, marketers face realities like competing business goals, campaigns, channels, budget constraints, and product managers with myopic views, to name a few.   Large companies are often faced with campaign calendars that may not represent an ideal communication plan with its customers.  The below diagram illustrates this phenomenon for an electronics retailer.



As you can see, campaigns and customers associated with these campaigns can easily overlap.  If you are a prospect in each of these campaigns, will you feel overwhelmed by the number of contacts?  If you are a marketer with limited budget, how should your prioritize your spend across campaigns to generate desired response rates or ROI?  With multi-LOB companies with many products and services, it makes great sense to employ some degree of intelligence into the marketing equation to ensure a win-win outcome for companies, LOB’s, and last of all but not least, customers.

MO across campaigns and channels typically relies on the development of business rules, the utilization of sophisticated mathematical algorithms, or both.  Most software applications that use mathematical algorithms typically use linear or non-linear algorithms that attempt to maximize an objective function (e.g., response rates, profit), while imposing constraints.  Constraints may include: budgets, minimum/maximum number of offers per customer and/or campaign, channel capacities, etc.  While very powerful, optimization algorithms are problematic to use.  They require statisticians that build customer response and valuation models, as well as profitability models.  This takes time and money.  Then there is the issue of ensuring the algorithms actually find the global minimum (cost) or maximum (response rate) as desired.  The image below helps to visualize this issue.



It’s possible for algorithms to find “local” minimum/maximums that lead to sub-optimal marketing outcomes.  That being said, in the hands of the right practitioners, mathematical optimization can create significant marketing ROI.  So, short of the required expertise and/or budget, what are marketers to do?

More recently, software vendors have tackled this issue via the development of business rules that marketers can build.  Business rules can work within and across campaigns to optimize your communication plan.  Examples of business rules include:

• No more than 2 weekly communications via any channel to a customer, to minimize fatigue
   
• Make the best of multiple potential offers based on profit, revenue, or likelihood to purchase, as examples.
   
• If a customer may be touched by multiple campaigns in the next month, only communicate with them about the two campaigns with the highest priority.

These types of rules can easily be developed using a point-and-click interface like that found in Aprimo’s Contact Optimization module. 

So, business rules are easy to create and use -- there must be a downside, right?  Yes, there are tradeoffs associated with simplicity and ease-of-use.  Some of those include:

• We are really not optimizing an outcome from a mathematical point of view.
   
• Business rules support “subtraction”, i.e., supporting the imposition of a maximum number of touches, offers, etc.  Linear and non-linear algorithms can do that, but they can also impose minimums like, the number of offers or contacts per campaign.

Keep a look out for my next article that will continue this discussion and provide some real-life case studies.

 

• Transaction data - For every sale, you at least need to know the product and the amount and date of the purchase.
   
• Past campaign response data - For every campaign you've run, you need to identify who responded and who didn't.  You may need to use direct and indirect response attribution.
   
• Geo-demographic data - This is optional, but you may want to append your customer file/database with consumer overlay data from companies like Acxiom.
   
• Lifestyle data - This is also an optional append of indicators of socio-economic lifestyle that are developed by companies like Claritas.

• Missing data analysis. What fields have missing values? Should you fill in the missing values? If so, what values do you use? Should the field be used at all?
   
• Outlier detection. Is “33 children in a household” extreme? Probably — and consequently this value should be adjusted to perhaps the average or maximum number of children in your customer’s households.
   
• Transformations and standardizations. When various fields have vastly different ranges (e.g., number of children per household and income), it’s often helpful to standardize or normalize your data to get better results. It’s also useful to transform data to get better predictive relationships. For instance, it’s common to transform monetary variables by using their natural logs.
   
• Binning Data. Binning continuous variables is an approach that can help with noisy data. It is also required by some data mining algorithms.

I'd love to hear your questions or comments if you have a few moments over the Holidays. 

My best to you and your family!
Jim

Forrester's recently published study on Interactive Marketing (email, social, dialog, banner, etc.) reveals 68% of survey respondents expect to achieve increased email marketing effectiveness over the next three years.  Furthermore, survey respondents also indicated they would increase interactive marketing budgets by 60% by shifting funding away from traditional channels: direct mail (40%), Newspapers (35%) and Magazines (28%).  The picture that is emerging here is one where marketers have high expectations on interactive marketing and expect to focus less on traditional channels.  A lot will be riding on this reallocation of marketing budget -- so what will marketers have to do right to fulfill their hopes and expectations?  This particular blog will address best practices that must be followed by email marketers.  Future blogs will address social and dialog marketing in detail.

I don't know about you, but I'm not sure I can handle more emails coming into my professional and personal inboxes.  I get so many from the same companies that I don't even open them -- not even when they come from companies I opted into.  Companies that email too frequently create so much "white-noise" that it affects their open rates as well as the open rates for other companies.   In addition to white-noise emails, I also get many others that made me think -- "why did I even get this?...I don't smoke, so why am i offered a smart smoker trial?"..."I have only rented mystery and adventure movies from you, so why are you telling me about The Lion King release?"  You experience the same things and feel the same way too.  So, what can email marketers do to ensure success and rise above the noise and mediocrity we see everyday?  It takes only three things -- relevancy, segmentation and testing.  These three tactics are the key building blocks to optimizing your email marketing efforts.

Relevancy - A blog I posted a couple of weeks ago spoke to email relevancy -- that it's about personalizing the email, segmenting your audience and testing your content (copy, images, subject lines, etc).

Segmentation - Your audience will differ by demographics, personality, shopping habits, geography, etc.  Simple segmentations where different messages are sent to each segment can deliver huge marketing ROI.  A recent Marketing Experiments webinar offered a case study on American Greetings.com (AG).  AG's goal for their email campaign was to increase individual Ecard purchases as well as Annual Subscriptions.  They created two segments -- Segment A contained customers that purchased humorous Ecards in the past, while Segment B contained customers that purchased traditional Ecards.  Each segment got an email that spoke to their interests based on this past purchase behavior.  This simple use of segmentation resulted in a 70% improvement in conversion rates when compared to a control group -- that's HUGE!   Just imagine what more sophisticated segmentation schemes might produce!

Frequency - Ok, so I have a real issue with this particular topic.  I  can't begin to tell you how much junk I get in my inbox.  I don't even open emails from some marketers and yet I still get an email every day from them -- please do some analysis on open rates and realize, I'm just not into Chocolate Covered Strawberries -- OK?!  Oh yes, back to the informational part of my message...  The same webinar by Marketing Experiments (I suggest you Google them!) provided another case study on a very large anonymous Ecommerce company.  They segmented their customers into seven segments.  Each segment got a different number of emails over a 60 day period.  At the extremes, one segment got an email every other day, while the other got an email every 15 days.  During the webinar, the audience was polled to see what they thought the optimal number of emails would be.  They chose 3-4 per month based on their own experiences and readings.  Well, the actual results were quite surprising.  Their test showed that customers that received emails every two days produced 3X the revenue of the segments that got 2-4 emails per month.  In fact, there was a significant positive correlation across all segments based on the number of emails they received (see below graph).



You would think this is illogical.  Most email marketers believe we face the tradeoff shown below -- that there will be an increase in revenues at first, but then we'll experience more unsubscribes or non-opens as the frequency increases.



So, what is the disparity between the experience of the webinar audience and the results of this study?  Well, we are simply seeing that each company has a unique customer base and a unique relationship with them.  You can't just assume your optimal frequency should be what is best "on average" or for a specific company they read about.  It means that every company must do segmentation and testing to determine the right frequency for their unique audience.

Caveats? -- there is always one or more:

1) Tell your ESP that you'll be doing experiments and they may see greater volume than normal.  After all, you don't want to be blacklisted.

2) Also look at open rates and unsubscribes during your testing.  The anonymous email marketer in the 2nd case study saw no correlation between frequency, and open rates or unsubscribes per email sent.  But your experience may be different.  Remember, an unsubscribe doesn't just effect revenue from a given campaign, but it also erases expected/future customer lifetime value.

During a recession you not only have to compete against your regular competitors, you must also fight the most dreaded enemy of all—no decision! During tough economic times the only thing you can count on is that you’re going to have to work twice as hard to close the same amount of business as before. Therefore, you must ramp up your campaign management efforts accordingly.

RFM and data mining help you find a subset of your customers that are most likely to react/respond to your marketing campaigns (personalized emails/microsites, direct mails, call center, etc.).  By targeting ONLY those likely to respond, you achieve about the same response/sales at a fraction of the cost.


Aprimo's Multichannel Campaign Management and Data Miner solutions provide the tools you need to survive and even flourish in the current economy.  No data miners on staff? Simple RFM (recency, frequency and monetary value) can help you increase response rates and overall campaign performance.  The notion is those that bought most recently, purchase more frequently, and have spent the most, are your best prospects for future marketing campaigns.

If you're a B2B marketer, Aprimo Lead Manager (LM) can automate the nuturing and scoring of your prospects until they qualify for "Lead" status.  At that point, LM can automatically assign and route the leads to the appropriate sales person.  With Aprimo, no leads escape your sales funnel.  More to come on all of these topics soon!
 

The key to achieving your desired conversion rate is relevancy -- pure and simple.  It's more than using your microsite software to support specific campaigns.  It's about testing and delivering personalized emails with relevant content that drives customers to a personalized and relevant experience on your microsite. 

A few words stand out in the above paragraph that merit additional attention.

Personalize - This means many things to many people.  It can be as simple as embedding the customer's name in the email message.  A recent study by Aberdeen found that personalizing an email with a name increased conversion rates by 200-300% over non-personalized emails.

Relevant - The message/offer needs to resonate with the customer.  Relevancy can be driven by events,  prior purchases, and/or through segmentations.

• Events - A customer that downloads a whitepaper or article about a product or service could be ripe for a follow-up email or call.  A dramatic increase in bank account balance could signal a call-to-action from a bank about investment options.  A very personalized email could be triggered to drive customers to personalized microsites with a relevant message that speaks to the customer's need or interest.  Lead nurturing applications can play a key role in supporting your marketing efforts related to customer events.
   
• Prior purchases - Simple cross/up-sell campaigns can be driven by product purchases.  For instance, a customer that purchases a water filter could receive an email that drives them to a microsite that attempts to enroll them in scheduled deliveries (recurring sales!) of replacement filters.  Data mining can also use information about prior purchases (RFM type data) to predict the likelihood of a customer's interest in other products or services.  Then we simply communicate to customers about the products they are most likely to purchase (based on a statistical probability to respond).  We won't always be right, but more times than not, this type of personalized communication will increase conversions and improve our campaign results.
   
• Segmentations - There are many ways to create segmentations.  One is based on industry, product and customer knowledge that is accumulated over time.  For instance, "I've worked in this industry for 10 years and know that females, aged 21-25 are the best targets for my product."  Another interesting segmentation approach that improves campaign results is customer clustering.  Clustering is a data mining technique that creates customer segments where everyone in one segment is similar to each other based on customer attributes (e.g., gender, age, prior purchases, geographic location, income class, etc.).  While everyone in a given segment are similar to one another, each segment in general is quite different from any other.  Once we profile each segment, it is easy to develop a personalized message that goes beyond first name.  The actual copy/text of the email can be personalized to be perceived as even more relevant.  If just using first name for personalization leads to a 2-3 X conversion improvement versus mass email, just imagine what affect personalized copy will have.  Let's look at an example.
  
  A large print newspaper in the northeast was experiencing declining subscribers like many of it's counterparts nationally.  The newspaper appended Census data (number of residents, race, ethnicity, age, income, home value, average commute time and many other variables) at the zip+4 level to all of it's subscribers. It then used clustering to create five different clusters of customers based on the Census data.  Their idea was to profile each group and develop editorial zones based on these segments.  Each editorial zone would receive it's own unique newspaper content based on assumed interests as derived from the cluster profiles.  One cluster was comprised of the highest proportion of customers with high home values, 4-year degrees and the lowest proportion of people with blue-collar jobs.  This cluster also  enjoyed the highest penetration in terms of current subscribers.  You can see how the content this group would be interested in would differ from the cluster with lower education and income.  By personalizing the newspaper content, the newspaper reduced the rate of subscriber loss from all segments/zones. 
  
  This information was also used to promote customizable online versions of the newspaper as well.  Subscribers now opt-in/out to various content.  As such, they are directly professing their interests in a topic or issue.  This information is even more powerful from a marketing perspective than what we "deduce" via analytics, and can drive a circular process where we get to know the customer better and better over time.  This increases customer loyalty and ROMI.
  
  Many organizations have even taken this idea farther from a Social Marketing perspective.  Customers can form their own clusters by opting in/out of particular forums or discussions.  Creating customer segments based on the forums or discussion groups to which they belong is valuable low hanging fruit.  Some leading edge companies are also applying Text Mining to customer posts to take proactive steps for customer loyalty/retention, cross-sell and acquisition efforts.  More to come on Social Marketing and Text Mining in future posts.

Let’s face it -- we’ve all gone to website pages that take forever to load.  Sometimes, the load times are so long, I just close the window or hit return and hope a competitor’s webpage loads faster.  So, what’s the culprit here?  Typically it’s a large Flash file or just a page with one or more images that have not been optimized for the web.  Non-optimized images and long page load times adversly affect conversion rates and marketing campaign results.  Whether you’re building a simple webpage, a marketing oriented microsite or an email, images must be optimized for the web. 

Optimizing images for email and microsite pages is the act of finding the sweet spot between a great looking image that takes too long to download and a grainy image that downloads in a second at dial-up speeds of 56kps.  The idea is to modify the image to retain a nice rendering while decreasing the overall file size.  Optimization becomes more and more critical as we add more and more images to any microsite page.

There are a number of software packages out there that allow you to resize images – Photoshop, Fireworks and Paint Shop are a few of the most popular.  By resizing, I mean changing the absolute size in pixels, as well as the file size itself.  Changing file size refers to changing the amount of data compression used for an image.  The most common image files used for the web are JPEG, GIF and PNG.  The difference in JPEG, GIF and PNG is the way they compress data.  GIF and PNG compression work almost exactly the same, but PNG often produces slightly smaller files.  JPEG compression is designed to optimize images with fine gradiations of color, while GIF and PNG are better at compressing images with large areas of color, such as illustrations.  The more you compress JPEG files, the more artifacts you see.  This is because you are actually removing “data” from the file.  Here are a few JPEG examples I created using Fireworks.



Look at the captions below each of the images.  The original uncompressed image on the left is 96K and would take 15 seconds to download if you were using a dial-up connection.  The image on the right manitains almost the same overall quality but has been compressed to less than 25% of the original file size, resulting in only a three second download.  Now imagine that there are four images on this page that are these sizes.  A webpage using non-optimized images would take 60 seconds to fully render, while a webpage using optimized images will only take 12 seconds.  This is the difference between losing and gaining customers that visit your web and microsites.  As you can see, "size" really does matter!

While GIF and PNG compressions do not actually lose “data” like JPEG compression, they do lose color fidelity.  GIF and PNG files are limited to 256 colors or less.  When compressing these files, we typically move to 32, 24 or 8 colors.  Here are a couple of examples of compression using PNG files.



The above images are virtually identical in appearance.  However, by looking at the captions, you can see that the 16 color, 8-bit image on the right is only about a third of the size of the original.  

Many companies today are adopting marketing management technology that allows marketers to easily create their own marketing campaigns, emails and microsites.  This is great, but companies need to also put safeguards in place to protect their brand.  This is where marketing asset management comes into play. That is, the notion of creating assets like logos and other images that have been optimized and approved for use in marketing campaigns.  Please feel free to visit our blogs on Brand Asset Management to learn more.