Tag Archives: minethatdata

Studying a Twitter ecosystem one user at a time

If you’ve been following my (roughly) monthly posts on Jason Falls’ blog you know that I’ve taken this tack: On his blog I cover the key concepts of a particular web analytics approach, then provide additional support for that idea here.

A recent example is from two months ago. I posted about the use of Brownie Charts as a way to report Content Interest Index. I posted a parallel piece here on another use of the technique (Using Brownie Charts to Measure Bounce Rates). You could say this blog has become my laboratory: Results of preliminary experiments are described here, while the “real” story is broken on Jason’s blog. Tomorrow will be a little different.

Tomorrow, on Jason’s blog, I’ll be posting on someone else’s innovation. It is a review of an extraordinary book: Hashtag Analytics. I’m a huge fan of its author, Kevin Hillstrom, and over the years I’ve spent way too many hours creating Excel-driven models in order to replicate and fully understand his findings.

I’ll be doing that again, this time in support of Kevin’s approach to monitoring Twitter communities. Check back at this tag (hashtag-analytics) to read updates on my “lab work.” Ill be reporting over the next several weeks.

When A Hashtag Community Member Is “Removed”

You may want to check Kevin’s blog as well — especially later this week, when Kevin reports on the future vitality of the hashtag community #measureHe posted about it last week. Now he plans to theoretically whack an active member. Here’s an excerpt from his post, where he invites readers to suggest whom to “remove”:

In every e-commerce company, somebody is responsible for forecasting sales for the next twelve months, by day. So it makes logical sense that any community manager would want to know what the future of his/her community is, right? This is something you don’t find in any of the popular Twitter-based analytics tools. This is my focus. This is what I love doing, it’s completely actionable, and it’s an area of analysis not being explored!

Next week [the week starting January 24 — that’s today!], we’ll do something neat — we’ll remove one important user from the community, and we’ll see if the absence of the individual harms or helps the future trajectory of the community. If you are an active participant in the #measure community, please send me a user_id that you’d like to see removed in the forecast … I’ll run an example for the individual who gets the most votes.

And in two weeks, we’ll compare the #measure community to the #analytics community … competing communities doing similar work … which community is forecast to have a stronger future?

It’s a fun stunt / modeling experiment that has real world implications. It should serve as a proof of sorts of the predictive power of his Hashtag Digital Profiles and the statistical work behind them. More relevant to online community managers, it should illustrate why showing your participants “love,” lest they never return, is of tremendous importance.

What To Expect Here

I will be applying my own Hashtag Analytics to a different online group — one that has the advantage of weekly meetings. It’s a fairly new group, so the rules may not fully apply (Does an acorn sprout follow the same natural laws of growth as a full-grown tree?). To ensure I don’t jinx my test or influence the community — in a far more direct way than Heisenberg was referring to — its identity will remain unknown until I’ve gathered and analyzed a critical mass of data.

Do stop back.

January 25, 2011 Update:

Here are two related links I didn’t have yesterday. The first is Kevin’s post where he removes that member to the #measure Hashtag community. The second is my review of his book today on Jason Falls’ blog:

  1. Hashtag Analytics: Removing a Member of the Community
  2. Lessons from the Twitter Lover Guru

Estimating the true value of a web visitor

Marketers are still grappling with finding the real value of digital marketing efforts. At the end of a promotional campaign, marketers find even the most “trackable” of web visits difficult to value. If we didn’t know this already, a survey of chief marketing officers, conducted by Atlanta’s Heidrick & Struggles, clarified the problem.

Kevin Hillstrom, author of Multichannel Forensics, suggests a reasonable way to approach web campaign valuation. It works well for e-commerce site visits, and even sheds light on valuing other types of visitors.

Placing value on non-buying visitors

Start the valuation process with the obvious: Visitors who immediately convert to a sale. But then keep going. Apply common sense numbers to those visitors who do not immediately buy. And why not? Visitors may come back two, three or more times before making a purchase. Like a fish that nibbles before biting, these “lost” visits aren’t so lost after all. They should be fairly depreciated, not totally ignored.

Here’s what Kevin says about ignoring all but the “one-visit” purchasers:

We try our hardest to allocate orders to the advertising vehicle that caused the order, seldom considering a series of events.

Take paid search as an example. Assume that a paid search campaign results in a 3% conversion rate and a $100 AOV [Average Order Value]. We run a profit and loss statement on the 0.03 * 100 = $3.00 demand generated by the campaign, factoring in the cost of the campaign.

What about the 97% of visitors who did not purchase?

Hillstrom asks, “What if you had this data?”:

  • Of those who are left [i.e., 97% of the base], 50% will visit the website again within one week, with 3% converting, spending $100 each.
  • Of those who are left, 50% will not visit again. Those who are left will visit again within three weeks, with 3% converting, spending $100 each.
  • Of those who are left, 50% will not visit again. Those who are left will visit again within one month, with 3% converting, spending $100 each.
  • Of those who are left, 50% will not visit again. Those who are left will visit again within four months, with 3% converting, spending $100 each.
  • Of those who are left, 50% will not visit again. Those who are left will visit again within six months, with 3% converting, spending $100 each.

He goes on to explain:

There is value in each case, value that most of us choose not to measure.
When I iterate through the five cases above, I calculate an additional $2.75 of future visitor value. [I get $2.68 in my number-crunching, as the number in the lower right of the graphic shows. Here’s my math in a Google Spreadsheet.]

Value of a Site Visitor Assuming $100 AOVIn other words, we measure the $3.00 generated by short-term conversion. We don’t always measure the $2.75 of future conversions.

Now there may be additional expenses associated with the $2.75 number — that customer might require additional paid search expense or might use a shopping comparison site, whatever. So we need to run a true profit and loss statement on the additional $2.75 generated by future visits.

If each first-time visitor (one that doesn’t convert immediately) is worth $0.30 profit over the next twelve months, you think differently about attracting visitors, don’t you?

I agree. The BrandWeek article I linked to in the first paragraph said that the CMOs surveyed, “Expressed an awareness of digital’s potential, along with a recognition that they weren’t close to tapping it.”

Building sales models that take into account the messy realities of online behavior is one way we can start.

Thrive in this down market by finding and catering to social customers

Database marketing consultant Kevin Hillstrom has done impressive work in helping retailers trace their customers across sales channels, using Multichannel Forensics (Note: this link is to a PDF file). Now he’s helping clients — and blog readers like me — to find creative ways to re-segment customer files based on responsiveness in a Web 2.0 world.

NOTE: Click for larger image

In his post, Kevin lists five segments of customers. Three are familiar: Organic (those customers who are your without a traceable stimulus, such as advertising), Advertising (they’ve purchased because of a non-discount-related ad) and Begging (you’ve given them discounts and other strong incentives).

Two are new to the Web 2.0 world, and thinking about them is a valuable exercise. They are Algorithmic and Social customers. Here is his description:

Then we have customers who use algorithms to purchase. Yup, these are the customers who use tools like paid search to purchase merchandise. These customers are different. They don’t always respond to future advertising, and when they do respond, they combine advertising and algorithms to make decisions. This is where your Net Google Score comes into play. Catalog brands really struggle with algorithm customers, and online marketers struggle with e-mail marketing programs for algorithm customers.

Increasingly, we find ourselves managing social customers. If you’re Crutchfield, you have customers who buy merchandise, customers who write reviews, and customers who are referred from blogs to your site. The latter two groups represent “social customers”. Social customers are different than are typical catalog customers, and are different than typical e-commerce customers.

These two segments describe a type of purchasing behavior that is brand new. Especially the Social customers.

Hillstrom goes on to say this about Social customers:

Catalogers are way behind the curve when it comes to managing social customers. In fact, almost everybody is behind the curve regarding social customers. Hint: Social customers don’t necessarily embrace catalogs, and sometimes get really angry when [you fill their] mailbox.

Smart catalog marketers are hyper-sensitive to the nature of their customer conversations. Even if you’re not in the catalog retail business, you should be too.

Here are two examples:

  1. When people arrive at your site from an organic search, greet them with the phrase they searched for (when it is a relevant phrase) and offer several links that can help them better find what they’re seeking. Then trace them to a conversion and compare to a control that receives no greeting.
  2. When people arrive from a social site, watch where they go within your site. For statistically significant instances (lots of page views, subscriptions to e-newsletters, etc.), consider making a friendly, overt presence on that social network (remember these 11 magic words when posting any social media comments).

Other ideas will come to you when you realize the obvious: The source of a customer changes that customer’s future buying behavior.