Thoughts From The Cake Scraps

Privacy is, and always has been, something that needs to be taken very seriously. People will only return to or do business with sites they trust. This means that while a website can track everything, it does not mean that it should. In fact, several large media sites have recently been sued over the use of “flash cookies”, which can be used to identify returning site visitors even after the user has deleted their standard cookies.

From a measurement perspective resetting cookies is excellent, as it creates a full view of the customer and their activity with your site. The allure of that level of information is obvious (lifecycle marketing, retargeting, etc.), but the user cleared their cookies for a reason and resetting it via a non-standard method can easily break trust. If you are thinking of trying out the use of “flash cookies” to reset browser cookies, you are in a very gray area and run a real risk of breaking the trust of your users–not to mention possible legal action.

In fact, because the speed of change in the web analytics industry, you should review your privacy policy to make sure that it is up to date and that the appropriate information is available to your users. To take it a step further, keep an eye out for the recently published Web Analytics Association Code of Ethics; it provides a starting point for discussion around best practices within the web analytics industry.

Use these best practices to preserve the trust of your users. Always keep privacy top of mind – even when it means reduced tracking capabilities. If data collection and privacy are not handled properly they can become a lightning rod for criticism and an unnecessary distraction from the real value that web analytics can provide.

Originally written for Catalyst, reposted here with permission.

This has been a Thought From The Cake Scraps.

Recently within the #measure community there has been much talk about Google launching https://www.google.com which is an SSL version of Google search.  The big issue is that when a user clicks through a search result the referring site is stripped off and there is no way for the destination site to tell what keyword was used.  What will happen next depends on who you talk to.

Perhaps this means that businesses don’t know what is driving traffic to their site (via natural/organic search) so they will not want to spend ad dollars on keywords that may or may not be used frequently by their visitors.  Maybe businesses switch spending to a different service, such as Bing, where they can feel more comfortable knowing what they are spending on.  For the web analysts, like myself, there is the issue about where the traffic shows up in marketing channel reporting.  We will no longer know how much traffic Google is driving to our sites.  Since Google is such a major player in the search game this could be a huge issue.

But there is one problem with all of this.  While I -  as a web analyst – care about this,   I -  as a customer, as a searcher – don’t give a crap what a company does or doesn’t get.  That’s their problem to work out.  In fact, with all the talk about privacy from all the Facebook changes, as a searcher I would be happy with any new security Google can provide for me.  So this means that the users of search are not going to drive Google away from defaulting to SSL searches and perhaps do the opposite and attract more pepole.  That only leaves the businesses paying for advertising, but they hardly have any strength at all.  Really, it just means businesses will not spend money as efficiently as they could be so they would have to buy more keywords or risk lost revenue from lack of paid traffic.

That inefficiency is one of the smaller ways Google could make some money.  There is an even bigger opportunity that is revealed in Google’s own statement about the service:

Searching over SSL doesn’t reduce the data sent to Google — it only hides that data from third parties who seek it.

You see where I’m going with this?  Google could start charging for access to the natural search.  This is sold to senior leadership at businesses by stressing that without the data the company 1) won’t know what is driving search traffic to the site and 2) won’t be able to spend their search advertising dollars efficiently.  It is a grand slam for Google.  Furthermore, before you get up in arms about the thought of paying for this data, have you ever used Acxiom data to gather targeting information on customers you otherwise know nothing about?  If you’re not in the direct mail space, then what about Hitwise or Comscore?  All of that is the same thing. Nielsen Ratings? Same thing.  Just a company collecting (or buying) massive amounts of data and packaging and selling that data to clients.

At the end of the day businesses simply cannot afford to say “screw Google, we’ll buy ads elsewhere”.  They will pay for this data, and probably line up to do it (after the mandatory complaining about how they used to get it free).  We all know that Rupert Murdoch is changing the face of news on the internet by charging for it.  There were even talks about making Google pay to index it.  Google needs to keep looking for new ways to generate revenue; charging for search data may be it.  At the very least it could come free with Google Analytics, giving you one more reason to switch.  I bet that wouldn’t make Omniture / Adobe very happy.

I think things could get interesting.  What do you think?

This has been a Thought From The Cake Scraps.

Businesses want to see the product that they have.  That is fairly obvious.  What may be slightly less obvious, though we probably all know it if we stop and think, is that businesses want to provide us with the tools we need to engage with them.  A basic offline example is carts at the grocery store.  This is provided to you simply to make shopping with that store easier and so that you can buy more stuff (as opposed to if you had to carry it all in your hands).

These same sorts of tools exist on almost all e-commerce websites in one form or another.  A simple example is a search box.  The option to search provides the visitor with a tool to find what they are looking for.  Another example is product categories.  A company doesn’t just put all of the products they have and place them into a single index page.  They try to group them by how they think their (potential) customers would group them.

All of these things make it easier to engage with the business.  The tools make it easier to shop or discover product.  This, hopefully, will get you to buy more product which is how the business can justify the investment in whatever tool they have developed.  Some, like the search box, are easy to see the value in while others may just be for fun.  Either way the end goal is the same.

Clearly it costs money to develop and implement these tools, therefore the initiative has to generate money.  This could be directly (sale now) or indirectly (brand goodwill for later purchase and/or consideration).  The hard part is measuring the value of one of these tools retrospectively.  If it was upfront, then test, test test.  But, in some cases, that just isn’t possible or wasn’t done for some other reason and yet the leadership will still want some number to hang their hat on even if it is just a very rough estimate.

The problem with reading tools after the fact is that it is very hard to prove direct value.  Sure you can see how much demand came through the tool.  Or how many visits interacted with it and how many times it was used in total.  You can look at participation metrics for the tool.  You can look at data point after data point but one very tough question will remain: how much of this is incremental?.  Even if you launch something and it is used by a significant portion of your visits, you will still have a hard time saying what is truly incremental.

For instance, mother’s day is coming up.  Let’s say you launch a page that lists many products for mom.  You may see that for those visits the % of time that a product is viewed and then purchased will decline.  But don’t worry, they are just browsing for gifts.  They are using the tool exactly as expected.  But if this change in behavior happened on the whole site you would have an issue because people are viewing product and simply not buying.  You might read that as a poorly designed product page or that your visitor can’t find the information they want.

On the flip side, you may launch some tool and find that anybody that touches the tool is like pure gold.  Conversion is up 50% for visits that use the tool.  Everything looks great, right?  Well, what if all of those people were your best customers anyway; what if they are your most engaged and most likely to buy?  Perhaps your tool is actually lowering conversion for these people, but the fact that they perform so well compared to the site as a whole causes you to miss this issue.  And, at worst, to declare the new tool a success!

The solution, of course, is to test everything.  That way you will have a true and clean read of the exact lift of a new tool.

What are your favorite tools that you wish all sites had?

This has been a Thought From The Cake Scraps.

Conversion, however that is defined for a site, is always the reason the site exists.  For me, a conversion is a visitor viewing at least 2 pages.  If I can do that then I know that I have engaged the person enough to look around the site a bit more.  You probably have some other definition of a successful conversion on your site.

A fairly common conversion point for a website is account creation or e-mail capture (e-mail sign-up).  The most common way to look at how successful the sign-up process is is by using a fallout report.  A nice funnel that shows how many you started with and how many fell out at each step.  The bottom of the funnel is the total number of people that made it through the process.  This is a great way to look at things but there are two very different ways of doing it.

The first way is based on pages viewed.  This is a very common way to look at fallout.  People that made it from Page A to Page B to Page C.  This works nicely in a very straightforward way.  It gives you a nice view of the total performance of that site path.  Unfortunately, at least in some WA tools, that is about all you can get at with the basic reporting capabilities.  The problem with this is that you might be missing some huge cake scraps, or golden nuggets, of information by looking at the data in aggregate.

Setting a success event on each of these pages will provide a much greater degree of flexibility.  For instance you could very easily look at campaign tracking codes and see how many of each event was set for each tracking code.  This might give you information that you simply didn’t have before.

Say, for example, that you had both display advertising and paid search campaigns pushing traffic to your site.  In all likelihood you know what the conversion is off of each of these tracking codes but you might not know how many of your email sign-ups are coming from each campaign.   It is very easy to start setting a success event on the sign-up confirmed page so that now you can get a count of that event by campaign tracking code.  Perhaps you find out that your paid search converts better but they don’t come and sign up for email.  This might cause you to change the messaging that you are doing in paid search (perhaps message email strong to drive sign-ups or message something else since e-mail sign-up just didn’t work).

Similarly, if you had a 2-step process, and set a success event on each page, you would be able to see if one type of campaign had huge sign-up issues.  Perhaps you would learn that you want to create a different on-site expirence for that type of campaign to drive up sign-ups.

Another thing that is great about using events is that they are easy to trend across time whereas fallout reports based on page views can be a bit more difficult or time consuming to generate.  The downside is that you probably have a limited number of events, so use them wisely.

What type of conversion goal do you have for your website?

This has been a Thought From The Cake Scraps.

I think I am going to try to refocus myself a little bit and talk about my life in the world a web analytics a bit more.  This was one of the things that I talked about at the onset of the blog and, while I have made a few posts on WA, I don’t think I did quite enough.  I didn’t publish what I was learning.  I didn’t take time to look back at what I learned in any given week.  Learning happens, often times, in small steps and so if we never take a moment to look back we never realize just how far we have moved.

Have you looked back on your acquired knowledge recently?

This has been a Thought From The Cake Scraps.

In my previous post on participation metrics I got a great question about how participation works across multiple visits.  And as long as I was going to write about that question, I thought it would be a great time to also state the single best reason to use participation metrics.

First, I will address the question about multiple visits.  It is a great question, and I would expect nothing less from a guy who is crunching Milwaukee Brewers stats dating back to pre-1900 over at Brewer Leaders.  Hell, Ted “Double Duty” Radcliffe wasn’t even playing yet (bonus points if you tell me why he was famous in the comments).

Getting back to the subject at hand.  Participation metrics are generally only good for a single visit and a visit is arbitrarily defined by ‘the industry’ as you leaving the page or being idle for 30 min.  A full dictionary (pops a PDF) can be found here.  Therefore if you added a bunch of stuff to your cart and ended your visit the next time you came only the pages that you then touched would be counted.  This is the same for the non-participation metrics as well.  Therefore it would be possible to have a visit that only has one page and has an order and revenue associated with it.  Just think of a person sitting on the order confirm page, going idle for 31 min, then checking out.

There are metrics that will track across multiple visits, but because this is all based on cookies, to say they are reliable would just be wrong.  They are only as right as cookie deletion rates allow them to be.  Anyway, that is a whole post in and of it self.  But now you know about participation metrics (or most other metrics) across visits.

The other things I wanted to touch on – and the title of the post – relates to when to use participation.  The single biggest reason, in my opinion, is ease of communication.  People like to see big round numbers.  If you can say that a promotion page was viewed by 50% of your visitors by participated in 60% of your revenue that is easy.  There are less moving levers.  With revenue distributed, when page views go up for some reason (say you are doing a lot of liquidations and people have to click around to find something is a size or color they want) it is going to give an odd looking number compared to what people might be used to seeing.

Giving a nice number or % of total revenue just makes communication much easier.  And sometimes, when you’re working on something that you don’t really want to work on, finding the answer and communicating it out and being done with it is all you want.

These metrics discussions sure are fun, no?

This has been a Thought From The Cake Scraps.

I have not had a post on some basic elements of Web Analytics in quite some time.  Previously I have talked about how a person is tracked on a web site both with internal campaigns and e-mails.  I think that stuff is great to know for anybody surfing the internet.  It gives you an idea of what all that stuff in the URL is.  Check it out if you haven’t.

Once question that comes up quite a bit centers around what a report or analysis means when it talks about Revenue Participation or Order Participation or other ‘Participation’ branded metrics.  The first thing you need to know is that it is not the same thing as non-participation branded terms (i.e. revenue <> revenue participation).  The second is that participation metrics are related to single pages within a website.

Simply put, when a metric has “participation” attached to it, the metric changes from being a distributed metric to a non-distributed metric.

Lets just concentrate on revenue, but know that the example is not specific to revenue.

Lets assume that I came to a site and purchased $100 worth of stuff.  Let us also assume that I saw 20 pages in that time, including checkout pages.  When an analyst is looking at reoprts on a page basis there are 2 ways to look at that $100 I spent.

The first way is to attribute (or distribute – however you want to think about it) that $100 across all 20 pages.  This means that each page gets $4 worth of demand.  This would include any page that I viewed, including the checkout pages.  This is nice because no matter how many pages I am looking at I am not double counting revenue.  It makes it simple.  You can just add up whatever pages you are interested in and you have your revenue number.

The second way is participation.  this would give each page that I saw $100 worth of revenue attributed to it.  You probaly don’t need me to tell you – but I will anyway – you cannot add up multiple pages with this method.  If you did that for my hypothetical purchase you would get $2,000 worth of revenue participation.

It seems a bit odd, but there are definite uses for each way of looking at revenue.  The first way – distributed – seems logical at first, but then you are giving revenue away from an index or homepage and giving it to a checkout page.  There is not $4 worth of demand on each checkout page.  With participation, each page gets full credit, but then you cannot add up multiple pages.  Each has its place.

I hope that that clears up the difference between participation metrics and non-participation metrics, at least as far as it relates to those metrics from a page within a website standpoint.

Do you have a preference between these metrics?

This has been a Thought From The Cake Scraps.

I was recently working on an interesting project where I was estimating the demand impact of a change that we had implemented to our site.  Without getting into the details, a change was made so that the customer would be less distracted during their shopping experience.  This then – hopefully – keeps the visitor more engaged with the site and if everything else goes well, they will then buy.

The tricky part is that the likelihood to convert changes at different points on the site, though it is a bit difficult to get it.

For example, if all a visitor sees is the homepage, they are going to convert at some percent.  Assume 10% for easy math.  If a visitor doesn’t bounce – meaning come to the site, see one page, and leave – say the percent to convert increases to 20%.  If the visitor then sees a product page they are now going to convert 30% of the time.  And finally if they enter the checkout process they will convert 40% of the time.  The point is that the level that the customer is at in the site changes the likelihood of conversion.

This seems like it would be a very obvious thing, and to a certain extent it is.  The key component here is not that these differences exist and you know about them.  The key is taking that knowledge into account when making an estimate for demand impact of a change.

If visitors have all of these different conversion points and a change is made that causes 1,000 visitors to not leave the site you need to take these conversion points into account.  Saying that the 100 more people will buy (using 1,000 visitors * 10 % conversion from homepage) is just as misleading as saying that 400 people will buy (using 1,000 visitors * 40% conversion rate from checkout pages).  When making a demand impact, make sure that you include a few inputs for these different areas.

For example: 100 visitors * 10% + 300 * 20% + etc.  As long as the percents add up to 100% and the visitors add up to your total you are in good shape.  You can then take this number times your average order value and you now have a demand estimate.  Note that you could even take this a step more and apply a different average order value to people who have been in different areas of the site.  For instance someone who is shopping for Outerwear or A laptop will probably have a different average order value then an individual looking at flip-flops or computer cables.

Ultimately you can segment this to any level that you are able to get.  Just make sure that the work that you put into arriving at the final number is worth it – especially if you are using the Omniture Excel Client.  Make a judgement call.  If it is just going to be small dollars or you really just need a ballpark then take the 1,000 visitors * 25% or something like that.  It is a guess, but it should be an educated guess.  Each different analysis will require varying levles of confidence.

Have you done anything like this before?  How did it go?

This has been a Thought From The Cake Scraps.

What is the market doing?  How is my site doing in relation to other sites?  How much time are people spending on my competitors site?  Hitwise can tell you all of this.  The real question is where do you go from there?

It has been my experience that people fall into two categories on this topic.  In the first group are the people who are absolutely convinced that you need to know how your competitors are doing and that Hitwise data is a must.  Residing in the second group are the people that say that the data may be interesting, but “it doesn’t impact what we as a company need to get done.”  It is important to note that the latter group isn’t saying the data is useless, rather it is just not going to impact what the company does.

That is what is said in the meeting.

In practice I have found that the data does impact the business.  People do make decisions on the data – and they should!  Don’t ever believe otherwise.  You have to know what your competitors are doing if you want to position yourself correctly.

Take the simple statistic of traffic to a site.  On one hand it is easy to say that you cannot change the traffic to another person’s site.  The data is not actionable.  Just focus on your own site.  I ask you to look deeper for a moment.

You know that your traffic is down 10% – perhaps you even have an alert set for such dips – but their traffic is down 15% to last year.  You could look at this and think that you are doing better than them.  That is good.  Keep doing what you’re are doing.  Then you look at your ‘competitive set’ a.k.a. a group of sites similar to yours.  Their traffic is down 15% as well.  Still looking pretty good for you.  The danger is thinking just that.

Yes, their traffic is down.  Your traffic is down less.  This is interesting, but you have to look for the real questions.  What are they doing or not doing compared to last year?  What are you doing or not doing compared to them?  This is where the power of the data is; not the data itself but adding that extra dimension.  If you can learn from them as well as yourself you can really help out yourself at a much reduced cost.

I think this brings about the way that the two groups mentioned at the beginning of this post really need to blend.  The first group – Hitwise junkies – are wrong if they are just looking at the data.  There needs to be an additional dimension added to turn the data into information.  The second group – the nay sayers – are also wrong if they see no value in the data.  The right questions have to be asked.  Comparisons can be made, but there needs to be an intent for action behind them.

If this blending can happen and form a third group then real information, not just data, is at your fingertips.  Do you agree?

This has been a Thought From The Cake Scraps.

More often than not if you are somewhere you know how you got there.  Hopefully you don’t have too many weeknights (weekends I will exclude) where you just wake up and have no idea how you got to where you are.  You may be smart enough to know how you arrived at a particular location, but your website – at least by default – is not.

This post covers the principle of having a Product Page Finding Method (PPFM) tag on your site.  If your site was successful in getting a visitor to a product page, you should really know how they got there.  And if you are a visitor you should know that this is one more way you are being tracked.  For more information on being tracked check out my posts on Internal Campaigns and E-Mail tracking.  I will point out now that this post is less about describing to a visitor how they are being tracked and more about how a website should track the visitor.  This is because a PPFM tag is less common and may not apply to many sites a visitor may go to.  Nevertheless, it is still something to keep an eye out for.

Back to tracking how a visitor got to a product page.  The easy solution is a ‘Next Page’ or ‘Previous Page’ report.  This will tell you what pages a visitor was going to or coming from, respectively.  It may seem like the answer to our question of how the visitor arrived at a product page, and it does at a simplistic level, but is of no use for aggregating data.  Consider an index page that lists all of a companies laptops.  How often does a customer click through to an individual laptop (a product page)?  There is no easy answer to this if you have more than a few laptops displayed.  A PPFM tag will solve this problem.

If you add a PPFM – that’s Product Page Finding Method – tag to each link on the index page then when the visitor clicks through to a product page you can tell Omniture to look for PPFM=INDEX_Laptops01 and it will store it to an e.var ( a commerce variable).   Then you can run a report in Omniture and look for instances of INDEX_Laptops01.  Compare that to the Page Views for your laptop index page and you have the rate at which a person is clicking form that index page to a product page.

Another trick is to make sure that all of your index pages are tagged and have INDEX in the PPFM tag.  That way you can actually do a search to pull back all instances of an index page click on any index page.  With any luck you have your pages named in a similar fashion – so you can get total index page views – and you can then get a site-wide rate that people are clicking though to your products from your index pages.

Now that we understand the concept of a PPFM, lets look at a few other uses for it.

Basically, you should not have an instance where a customer navigated to a product page and you do not know how they got there.  Other ways they could get to that product page include a ‘direct to product page’ search and a cross-sell placement from another product page.

The ‘direct to product page’ is useful if you have a search box that will allow a customer to go directly to a product page without going through an index page.  An additional way to tag this would be to have a search results tag – for instances when a search returns many products – and then any click from that index/search page to a product page would give credit to the search tag.

The cross-sell tag would be used on any product page where you are displaying some other products the customer might also like to buy.  Any click on these links will bring the customer to another product page and then the cross-sell tag would get credit.  You might also have a similar tag for items displayed in the cart.

The last thing to discuss is credit.  On a $100 order who gets the credit.  The simple way to do it is the last used tag.  The bad part is that with this method if a customer uses and index for the first 3 items and the last item they clicked a cross-sell item, the cross-sell tag will get all of the $100 attributed to it.  That isn’t really accurate.  The better way is to distribute the $100 via linear attribution.  That means that in the example above each of the index pages would get $25 and the cross-sell would get $25.  The tricky part here is that if a customer is browsing they may click to 10 different products from 10 different index pages and each of the index pages would get 1/10 a share of the revenue even though the customer only bought from one of the index pages.  Just something to keep in mind.

With this tagging in place on your site you should always be able to answer how a customer arrived at your products.  It does not quite answer the question on a page by page basis – i.e. for Product A the PPFM tags used to arrive there were cross-sell 24%, indes 53% etc. – but it will give you a much better idea, on the whole, how your visitor is getting to your product pages.  Just a little tip that can save a ton of work

This has been some Thoughts From The Cake Scraps.