March 05, 2009

Channel Migration Visualization

Many database marketers who practice Multichannel Forensics categorize customer activity on the basis of the channel the customer purchased from. Others take this a step further, combining the physical purchase channel with the advertising channel that drove the order.

  • A brand has two physical channels: Telephone and Website.
  • A brand has eight advertising channels: Catalog, E-Mail, Paid Search, Organic Search, Affiliates, Shopping Comparison Sites, Portal Advertising, Blogs, Twitter, Uncoded.
Each combination represents a micro-channel. Many combinations have little relevance, and are not included below:
  • Catalog / Phone.
  • Catalog / Website.
  • E-Mail / Website.
  • Paid Search / Website.
  • Organic Search / Website.
  • Affiliates / Website.
  • Shopping Comparison Sites / Website.
  • Portal Advertising / Website.
  • Blogs / Website.
  • Twitter / Website.
  • Uncoded / Phone.
  • Uncoded / Website.
At this point, you have twelve micro-channels, and now you have something. Take a look at the ecosystem revealed by a Multichannel Forensics analysis (click the image to enlarge it).

This is fun stuff!! You can clearly see the path a customer takes as she moves from newbie status to becoming a loyal customer.

E-Mail marketers --- pay close attention to the middle of this chart. I frequently find that e-mail marketing is the "glue" that links a customer as the customer moves from the past (direct marketing, catalog marketing) to the future (social media, organic demand).

Paid Search marketers --- pay close attention to the role you play in this chart. You are frequently a source of new customers, but the customers you bring into the business are often not responsive to traditional direct marketing. That's ok, as long as the brand doesn't try to convert that customer to traditional advertising via direct marketing.

Often, there's a transition the customer goes through.
  • Traditional Direct Marketing (Catalogs).
  • Traditional Digital Marketing (E-Mail Marketing).
  • Loyal Customer Generating Organic Demand.
And there's a digital transition that the customer goes through.
  • Paid Search
  • Organic Search
  • Loyal Customer Generating Organic Demand.
There's a ton of profitability to be had by marketing appropriately to customers walking down one of these two paths. And increasingly, we'll add in social media and micro-blogging and mobile marketing.


  1. Anonymous7:08 AM

    Hi Kevin,

    Very interesting post. Thank you.

    Viewing the graph made me think of attribution models. This seems like one or an extension of one.

    I don't think we've quite mastered multichannel/multisource attribution models in online marketing (at least from recent discussions with leading practitioners).

    The biggest challenge we face is sequencing the customer path through the different touch points (micro-channels as you referred to them). In other words, in which order did visitors go through the different micro-channels (before converting or taking an action on the site).

    Most web analytics tools, when configured correctly, will give you the total no. of visits/visitors for each micro-channel. However, they do not provide the sequences in which these visitors accessed the different micro channels. Did they first used paid search, followed by direct to site and finally by email? Or do they always access the site from one single micro-channel.

    Visual Sciences (now Discover on Premise) and Omniture’s Discover (to a certain degree) are two tools that could do this task. But most other tools do not have such capabilities.

    Your graph suggests you've managed to overcome this problem.
    Therefore, it would be very helpful to get your thoughts on how best to overcome this challenge. And any other idea/suggestions you might have on this really important topic.

    Thanks again,

  2. Hi Michael!

    The work I do is different, in that I do not attribute the path that results in a single order. I attribute the path that happens across multiple orders.

    In other words, if a new customer purchases via a catalog, that customer has a "x" percent chance of buying via paid search on their next order, then a "y" percent chance of buying via a blog on a third order.

    I connect these paths, then forecast sales for each micro-channel for each of the next five years, helping CEOs understand how micro-channels are growing or shrinking, so that the CEO knows where to spend marketing dollars.

  3. Anonymous2:35 AM

    Hi Kevin,

    Thanks for your response.

    Using your example I have three key questions.

    "In other words, if a new customer purchases via a catalog, that customer has a "x" percent chance of buying via paid search on their next order, then a "y" percent chance of buying via a blog on a third order."

    1) How do you know that this specific customer that first purchased via the catalogue came the next time via paid search? I'm assuming you are using a customer ID to uniquely identify the customer. Are you using a web analytics tool to match the customer ID with the referring source (in this case paid search)?

    2) This approach requires matching individual customers back to referring sources. How do you automate that process (given the different measurement applications)?

    3) It sounds like this model only applies to interactions where a customer purchased. In your experience, can it be applied to prospects as well?

    Kevin – I hope you don’t feel I’m asking for too much. Not seeking any trade secrets. I appreciate this is what you make your living from.
    Would be great if you could shed a little more light on how to use these important methods.

    Thanks again,

  4. My clients combine data from different sources.

    For purchases, they store that data in a data warehouse. They know the date the customer ordered, they use their web analytics package (typically Coremetrics or Omniture) to assign referring URL to the purchase, they know the channel the customer purchased in (phone, web).

    With purchases, my clients simply send me every single item the customer ever purchased on a DVD, and I grind through the information using SPSS (others would use SAS). With this software, I can write programs to analyze the information, calculate the probabilities, and create the map illustrated in the blog post.

    The vast majority of my clients do not take the extra step of linking website visits between orders back to the purchase database. A few of my clients do this --- when I worked at Nordstrom, we embedded e-mail id into the cookie, then exported Coremetrics visitation data into the main customer warehouse. For a third of our customers, we were able to track all web activity and purchase activity (including retail purchases) across channels.

    Life is fun when you know that the customer clicked-through an e-mail on a Thursday, visited the website again on a Friday, and then purchased in a store on Saturday --- and you can link all of those activities together.

    Life is still fun when you can analyze purchases across time, merging referring URL to each purchase.

    Finally, my clients build external routines to link the referring URLs to purchases. These routines are called "matchback" routines --- companies like Experian or Abacus(Epsilon) or Acxiom do this for my clients, then take the results and populate the customer warehouse with the combinations I listed in this post.

    The approach can be applied to prospects --- some folks export the web analytics data from Coremetrics or Omniture, at a cookie level, categorize landing pages and shopping cart pages and referring URLs, and then use SPSS or SAS to do this style of work.

  5. Anonymous5:19 PM

    Hi Kevin,

    Fascinating and helpful. Certainly got me thinking.

    Thank you,


Note: Only a member of this blog may post a comment.

Cost Differences

Do you remember Bernie Mac in Oceans Eleven ... negotiating van prices? Muttering nonsense about Aloe Vera while squeezing the sales dude...