Howdy.

No, I didn’t know much of your effort is toward “teaching”. I’m happy to learn, though. This also helps me to understand your point of view. You favor “simplicity” and in teaching models I favor analogy and metaphor, often starting with very simple models and building upon them.

Done and done. Thanks.

Yes and Hear Hear! to your “forge direct links between the various behavioral sciences and successful marketing efforts.And it seems to me we’re doing that!” One of my goals for this blog is being realized. Thank you for helping me to realize a goal.

You write “I would not suggest anyone use the Recency metric without some kind of segmentation, because since we are talking about likelihoods here, you need a coherent population of some kind. The error rate when looking at a single individual would be high, but across a population, again speaking to “likelihood”, it’s a great yardstick for placing your bets.” Yes, yet another example of people making first order estimates based on inadequate data because they failed to figure out the basic parameters of their experimental systems.

Then you offer “…sometimes you find that the Recency relationship is not as simple as it is at a more macro level, not as linear as ‘the longer it has been, the less likely they are to repeat’.” and I’m chuckling a bit.

Your description of response and profitability curves, etc., leading to new segmentations makes me think if the way we (NSE) uses Bohmian and Mandelbrot methods could provide a worthwhile predictive solution to this.

(you were just itchin’ to get “dis-Engagement” in here, weren’cha?)

Good thoughts and reasoning. Variable clouds v segmenting? I’m not sure having two options makes things simple enough.

(laugh, darn it)

WA vendors supply the tools. They provide instructions on how to tag a site. Sometimes they even help with that tagging. However I don’t think you can hold the vendors to blame for shoddy tagging on the part of their customer.

There’s been several posts about integrating WA tagging into CMS. If that’s possible then go for it. It will certainly reduce human error if the CMS adds the tags it needs to (and I’m a big advocate for CMS and WA vendors partnering here). But its likely there are always going to be those exceptions that need to be “hand coded”. And errors definitely creep in.

I’ve been working with our businesses around the world for over 4 years now, purely trying to get the tagging implemented in a consistent, and clearly documented manner and we’re still not 100% there, and this is mainly due to tagging errors - whether inadvertent, or a local business thinking they know better and doing their own thing.

In fact recently we’ve taken the step Debbie suggests - working with a third party to design a test that automatically crawls any of our 150+ sites and checks the tagging. But its difficult to get an automated crawler to log on to a secure site, or complete a multi-step application (it can be done of course, its just not easy). So you’re often still left with the issue of having to manually check some pages.

You suggest that WA vendors who partner with people who have “calibration” (we call it auditing) as a core competency must in some part be assuming responsibility for proper calibration, but I disagree.

Partnering isn’t about accepting the responsibility for their customers tagging. Its about producing a tool that is pre-configured to check that particular WA vendor’s tags - to understand their specific JavaScript - rather than (as we have had to do) develop an auditing solution from scratch. Whether this is offered as part of the WA vendor’s solution, or as a value-add extra is up to each vendor. But the need for such a solution is definitely there.

3.15am your time - is early morning this side of the water. Probably about 9:15 in the morning. That’s not to say I don’t stop up all night thinking about this stuff. I am admittedly a proud geek.

::
Given the above, ET would have determined that….
(Long list of impressive knowledge about someones browsing behavior and motives)
::

I guess then that NextStage’s tools must use an opt-in approach? Or are you using “methods” IE; combinations of research with consulting like Future Now with Persuasion Architecture?

You currently can’t know which five sites they have visited unless;

a) you had some kind of technology which is reporting activity back to you.

b) you were watching the person (like in a usability test).

c) You’ve developed profiles based on how all people generally tend to behave and have a system to optimize it.

For instance on click Zeroeth;

“BeneVino, using ET and without using Cookies, has already gathered enough information from you to know if you’ve been to the site before, regardless of which computer you’re working on.”

At this point you need to have either asked or guessed based on potential personality types. Regardless of which computer you’re sitting on gives it away meaning that it has to be a method rather than a singular technology. If that is the case then yes, I might be getting it and the difference between NextStage and behavioral targeting.

What behavioral targeting does is simply target ads and offers based on your behavior. Depending on the behavioral network you will receive ads based on preferences you have pre-identified or clicks you have
made. In your scenario after the auditory stimulation they might be *thinking* about food and your offers are presumably designed around this potential situation.

In the behavioral targeting world you wouldn’t know that unless the behavior indicated it. (IE they went to a search engine and typed “burger” - in which case you could serve an advert)

Where your service appears to be different is that you’re building a psychological profile which reminds me of Future Now’s persuasion architecture, mapping needs according to logical or emotional decisions and depending on fast or slow personality types.

This in particular sounds very similar to Keirsey & Myers/Briggs modelling; “the way you think, the way you feel, the way you believe things should be.”

So is your method to design websites based on this kind of principle? If so I get it. NextStage sounds like a system that combines technology (maybe a panel of opt-in visitors) and scientific design based on personaility types as well as a lot of persona based research.

::
Visitor Analytics, Customer Analytics

My belief is that such forms of analytics are too narrow.
::

I agree. It’s why in my up-coming book I am dedicating a chapter to Persona based design and another process on how to measure it. Combined the three forms of analytics and scientific testing is very powerful.

::
“It got quite heated at times as it should. Passions were ignited and people were drawing lines in the sand.”

I can’t share enough how much I disagree with the above as a solution methodology
::

I agree again. I said so at the time to some of the individuals involved via email, face to face and called for the subject to be dropped at one point because much of the debate revolved around semantics. As far as finding a solution to the problem was concerned the debate only increased awareness of the problem.

My comment on the Omniture blog was an observation around the passion that was demonstrated by everyone. Passion about your subject is required as I’m sure you’ll agree and what was really quite outstanding about the whole debate was how many people got fired up around the world. Without passion and fire in some of the discussions this would never have happened.

Lines in the sand just demonstrated the strength of the passions and yes egos.

::
Why were people getting heated?
::

One camp didn’t believe Engagement was a valid metric while the other camp did.

::
What was causing passions to be ignited?
::

The valid arguments on both sides of the fence added fuel to the fire.

::
Were lines being drawn to keep something out or not let something in (these are demonstrations of two very different psychosomatic states)? Were they being drawn to keep something in rather than not let something out (ditto)?
::

You are right about the egos/emotions in play. There were bruises and counter punches being thrown around which didn’t serve to help and made the whole debate a bit of a flame war after a while. But it calmed down and people got back to business

In my view lines were being drawn to prove or disprove the value of engagement as a KPI. There was one camp that didn’t want what Matt calls the “uber” metric and another that saw the value of it. The argument revolved around the uber metrics complexity and the fact that engagement was often a proxy for sales or outcomes.

There were facets of both arguments that were valid. If you put the engagement formula Eric designed in front of non analytics folks in an organisation they would leave the room in droves. For those guys conversion rate is a stretch. Eric states it’s not for everyone.

However that doesn’t mean it’s not valuable.

You segment high visit activity and content. You’re then looking for the recency trend that suggests they will dis-engage.

Once you see that trend you may be able to affect the behavior (onsite) of a dis-engaging segment.

Your welcome for both.

My thanks to and appreciation of Jim Novo, as well.

Recency…since that element was added to this discussion I’ve been doing some research on “recency”, the “notion that a human in habit tends to stay in habit” a you eloquently put it. The more and more reading I do the more and more I liken it to NSE’s Visitor Return Report. It definitely isn’t Loyalty although I’m thinking recency is often confused with loyalty (and this is probably already addressed by others in this discussion. I’m behind in my readings. What can I tell you).

There is a phrase used in my studies, “As soon as you’re ready to not put up with your life as it is any longer, it’ll change.” This concept seems to be the domain in which recency concepts dwell and are not where loyalty concepts dwell. From what I’ve read, recency only deals with (what one would hope would be) an inverse exponential — the more often someone returns the more likely they are to return, the less time between visits the less time between their last visit and their next visit.

There is nothing in the recency model to account for several social science elements, many of which are elements of habituation (sequencing or chunking are two that come to mind immediately).

Stabbing at the Future

To see if it’s done, perhaps?

I appreciate your “…there’s fear that data driven insight is going to suck the creativity right out of most of our jobs — that we’re all going to become slaves to the Algorithm in the end.” and “I predict that the real challenge in the Future of the Web Analytics is going to be more around the social technology implementation and maintenance, and not so much the physical technology.”

You just know I’m going to agree with you, yes? Let me push the envelope just a bit further; I’ve learned that people who are comfortable with their creative skills and creativity have absolutely no problem incorporating data-driven insight into their practice.

That “comfort” lends itself to being open to new methods and new technologies, especially if it means granting them the ability to improve their own processes.

First, 3:15am? It’s good to know I’m not the only one who’s feverish mind keeps them thinking all night.

Second, you ask “Could you explain the differences between Evolution technology and Behavioral targeting?”. My first response is no, I couldn’t. I can’t because I’ve never encountered a definition of behavioral targeting that makes sense to me in the framework in which I study. I can offer some responses based on your offering of “I would define behavioral targeting as changing a websites marketing materials based on a visitors actions.”

The possible response I offer is from a demonstration we did in Feb 2000, before NextStage was officially NextStage; Evolution Technology Ontology on an eCommerce Site. I state openly that nothing has changed from the original presentation other than correcting a few typos, fixing some grammar, some formatting and using “Evolution Technology” or “ET” in place of what we originally thought to call it.

Let me also offer an alternative from not so long ago.

On the day I read this comment a real situation occurred which I’ve documented with a reader of these blog posts. (This is where I get to be educated) Could you explain to me what Behavioral Targeting could do with the following, please? I’m asking because when I read about Behavioral Targeting I often think to myself, “This is useful how?” and a) I appreciate people may ask the same about what NextStage does and b) that the question stems my lack of understanding, not necessarily a lack on the part of Behavioral Targeting.

An individual goes to five websites that they’ve never gone to before and this blog (which they frequent) in a single session (serially, not in parallel) starting with this blog. This blog is bookmarked, the other sites are accessed by typing their names into the location bar at the top of their browser. The sites and their order are this blog, Lowes, HomeDepot, WalMart, then two local (to them) recreational vehicle sales sites (let’s call them RV1 and RV2). The only time they go beyond the homepage on any of the sites is on RV2 where they follow a “Service” link.

Total time front to back is perhaps five minutes starting on or around 2:30pm local time, most of that time they’re scanning this blog. Once they follow the RV site “Service” link they close their browser session.

Given the above, ET would have determined that

the visitor was male
their age (±5 years or so)
during a pause in their browsing of this blog they received auditory stimulation
after that auditory stimulation they started thinking about food
the thoughts of food were focusing on 4-5 hours in the future
this individual wasn’t interested in traveling
this individual wanted to buy something right now, wasn’t doing research for a future purchase
was shopping by product first price second
and it would have started making conclusions about the price point. It would have known how much this individual was psychologically prepared to commit to the purchase and would not have been able to assign a dollar value at this point. At least I doubt it could have assigned a dollar value this individual was willing to commit at this point in the browsing session.

Connected to a sufficiently robust CMS system and with some business rules in place, ET would have determined that this individual wanted to purchase a price specific, outdoor type cooking supply not related to travel for use later that day and delivered appropriate content (it would have started showing grills, grilling supplies, propane tanks and refilling information. Our visitor’s goal was the last. Also, at this point ET could have made very accurate dollar value guestimates if it hadn’t done so already). We have pieces and parts of this scenario detailed in Reading Virtual Minds, Chapter 4 “Anecdotes of Learning” (I really need to finish writing that book one of these days).

Visitor Analytics, Customer Analytics

My belief is that such forms of analytics are too narrow. Knowing someone only when they’re a customer or a visitor robs the analytical process of understanding the individual as a person and is an objectification of the person that most people object to when asked. Simply put, the more you know about any individual the more you can plan, anticipate, provide, respond, prepare, …, for them when and should they return. To that end, you can engender them to return by knowing them well enough and properly exercising that knowledge.

Should social communications prevail, I suggest that this level of knowledge (hence demonstrations of trust, fair-exchange, respect, etc) will become mandatory. People want to be treated as individuals and special in their own right. Even people who claim to want to “only be one of the crowd” are stating that their singular (to them) identifier is their quest for (what they think of as) anonymity, thus the knowledge requirement is again mandatory. This is something I believe you point to with your “Learning to combine these data sources is the way the industry will move in my opinion.”

And note that what I’m offering is my opinion, nothing more. My opinions change rapidly when sufficiently convincing contradictory information is recognized.

“It got quite heated at times as it should. Passions were ignited and people were
drawing lines in the sand.”

I can’t share enough how much I disagree with the above as a solution methodology. I was not part of the above and don’t know much (if anything) about the discussions occuring at that time. What I can offer is that such demonstrations are indications that logic is not being used, instead emotion and ego are in play (see “We have too much data to analyze so we’ll just close our eyes, shoot and pray”). Most people demonstrating “heated”ness are doing so because of some form of boundary or territory violation (physical, psychological, … Your use of “drawing lines in the sand” is an obvious indication of this).

My suggestion to people with whom I’m having a discussion and when recognizing emotion and ego are coming into play (including my own ego and emotion) is to (basically) ask what’s going on, inviting them (or myself) to focus their (or my) rise in energy at its own cause. People able to perform that refocusing often discover quite a bit about themselves and also usually end up producing stronger, more salient arguments for their causes.

So, as a general question, Why were people getting heated? What was causing passions to be ignited? Were lines being drawn to keep something out or not let something in (these are demonstrations of two very different psychosomatic states)? Were they being drawn to keep something in rather than not let something out (ditto)?

As for the heat and sand drawings around Engagement, the reformulation of Eric’s original equation demonstrates that everybody’s definition (provided the three basic rules apply) can be used equally, that inclusion and exclusion of different elements only comes down to a question of increasing or decreasing accuracy. If the model currently being used is providing sufficient accuracy for a business purpose, excellent and why be troubled further? If it’s not providing sufficient accuracy, here’s a method for increasing that accuracy.

This methodology is no different than using a variety of astronomical tools to get an increasingly accurate understanding of the heavens. A backyard astronomer is excited investigating with a 10″ Schmidt-Cassegrain, someone wanting to research extrasolar planets needs much more to get the job done. The question are “How accurate do your measurements need to be?”, “What you want to look at?”, “What do you hope to find?” and so on. A more earthbound example of this is meteorology. Want to know where a hurricane is really going to go? Get as many weather modeling products as possible and do a least-squares analysis of their results.

You write “So I am interested to know how your system works.” So am I, sometimes.

Then “It seems to suggest that two of the three data sources I feel the industry needs to work with (Quantitative & qualitative) could be somehow combined in one solution. Or am I wrong?”

I’m much more comfortable answering if you are correct rather than if you are wrong (I’m not a moralist). My biased view is that any solution based on a limited parameter set has already declared itself a limited solution. Accuracy will depend on the degree to which the original parameter set can be duplicated. This is true of everything. NextStage’s models use over 80 “variables” right now and we’re constantly in contact with other researchers to validate new variable forms once they are recognized.

Can data sets be combined? I hope so and so long as conservation of units applies. We agree in that, I believe. Will they? Again, my belief is that they will. When it will occur and through what agencies I have no conjectures at present.

(and now, back to work…)

Visits, and likelihood to Visit again.

So we take some level of Activty - 50 visits - to define “Best Visitors” and then ask, How are we doing with best Visitors, what percent have Visited in the past 60 days? You can segment this by Campaigns, by Content, etc.

For example, a study like this is extremely effective when you launch new Product / Content Categories, or change Policies. What effect did the changes have on Best Visitors? If Best Visitors = 80% of monetization, then that’s the right question to ask. The question that doesn’t give you the right answer is to randomly survey all Visitors - the “majority”, who only contribute 20% of the monetization. They will overrule the Visitors who contribute 80% because there are fewer of these Best Visitors.

Make sense?

The tricky thing with Visitors - especially if you don’t have an e-mail address - is you have to do something *now* when you see certain patterns that historically lead to dis-engagement, for example, does not sign up for RSS feed or newsletter.

If you can hook dis-engagement to something tangible like lack of feed behavior - those who don’t subscribe tend not to come back - you might be able to do something in CMS to address them. The bare bones version of this approach is the old exit pop newsletter subscribe.

If you can’t hook it to something tangible and have to rely on visit patterns, then you’re into something like Joseph’s NextStage.

Either way, using Recency allows you to act more quickly against the testing - you don’t have to wait for a segment to PROVE they’re not coming back, you can PREDICT it.

The good thing about doing customer analysis as opposed to visitor analysis is that you don’t need a fancy-dancy web analytics set-up to do it.

With customer analysis it’s fine, it makes perfect sense because as you say it can be done with Excel. I like the cultural aspect as well and it’s something I’m currently writing about.

However what I’m trying to do and what the first iteration of Eric’s formula was primarily about was defining the best chance of business.

What I’m now considering is whether the same formula could be used to define dis-engagement of web visitors, therefore allow us to predict when to offer our unidentified visitors something different to keep them interested.

If we could predict that we could add incentives when a certain behavior is displayed.

Your post (and the one prior to it) went some way to explaining how you’ve done that but I’m still a little unclear. You defined an interesting segment which I don’t quite follow;

>>Recall that with visitors, we looked at a segmentation using under or over 50 visits for Current Value>Last Visit within 2 months or over 2 months to define Potential Value 60 days ago?

What is the behavior we’re segmenting here? I realize this has to be business dependent as well so how have you benchmarked it in the first place?

I think your approach using RF is predictive and should be included in the white paper Eric is producing with a clear case of how to measure visitor (not customer) dis-engagement. RF is already covered in the formula so i think it’s just a case of explaining how to reverse engineer it.

http://blog.jimnovo.com/2007/04/25/engagement-customers/

and see if it makes more sense / this model is something you can start with. Or, tell me why it’s not!

From a “culture” perspective, the great thing about using the LifeCycle Grid approach is it’s exactly the same format every time. Once someone (Marketing) understands it, then every time you apply it to a new segment there isn’t any downtime.

So, for example, let’s say you want to see what dis-engagement looks like for buyers who *became new customers 6 months ago*. This way they have spent some time on the books and have had a chance to let the LifeCycle play out a bit.

First you apply the Grid to *all* buyers

Then you apply the Grid to only *jewelry* buyers

Then you apply the Grid to only jewelry buyers who have purchased over $1000 past 6 months

Then you apply the Grid to only jewelry buyers who have purchased over $1000 past 6 months who buy only precious stones

Then you apply the Grid to only jewelry buyers who have purchased over $1000 past 6 months who buy only precious stones mounted in gold

The same model, again and again. Going through these iterations teaches you a lot about what dis-engagement looks like across different segments. You start to see the patterns, and create tests for cells in the Grid, and discover where the most profitable “re-engagement” timing and offer is for each segment.

Make any sense, or not really your question?

BTW, you probably *will not* see anything very useful using a demographic segmentation. This is a behavioral model and the segment variables need to be behavioral - spend, type of product, content visited, actions taken.