Sorry, we could not find the combination you entered »
Please enter your email and we will send you an email where you can pick a new password.
Reset password:


Plus Report - By Thomas Baekdal - October 2013

How to Really do Analytics For Web Shops

Shared by Plus subscriber
Maarten Brouwers
This is Baekdal Plus content. It is shared with you for free by a member. Please reshare it.

I recently wrote the article: "Five Key Elements of Analytics for Publishers". In it I explain why analytics for publishers is different from web shops, but I also ended it by saying:

A conversion doesn't end when people buy something. It ends when they no longer return for more.

This, of course, needs a bit of clarifying. So in this article I'm going to outline how I would set up my analytics if I was operating a web shop for a brand. Not technically (there are thousands of articles about that), but conceptually. And you will see why I have a somewhat unorthodox view on analytics compared to most people.

I wish you could all start your own business

If there is one thing you learn when you start your own business, it is that most analytics means almost nothing. Pageview, visitors, time on site, bounce rates, source, multi-funnel paths and so on are all very nice, but they are all secondary to you as a business owner.

Another thing you learn is that there is often little correlation between things like visitors and pageviews with actual sales. And bounce rates can go up and down completely independent of what is actually happening on a product level.

I see this all the time on this site. One example is free trials. You might think that if I could just get more traffic, then that would lead to more free-trials for Baekdal Plus, which in turn would lead to more subscribers. That's the usual conclusion marketers make. Boost one number and the rest will follow as a result.

But that is not at all what I see here on my site. In fact, I see the opposite. When I write something truly popular that brings in a lot of traffic, the number of free trails does indeed go up, but the number of subscribers per week goes down.

Yes, down!

On the other hand, if I write a high-quality Plus report that attracts little traffic, the number of free trials go down, but the number of new subscribers goes up.

There are exceptions, of course, but that's also the whole point. The reason something works while other things don't is rarely caused by having 'more traffic'. It's not the traffic that generates the sale. It's the reason why the traffic is there in the first place.

Another very common problem is when brands blame marketing for what is obviously a product problem. This is the reason why I am often telling you that your most important marketing channel is your product.

And, last month, Kevin Hillstrom wrote a good article illustrating just this problem:

At some point in a meeting, especially when business is below plan, the merchandising team turns on the marketing team. It's "their" fault. If it is a catalog business, then the wrong customers are being mailed. If it is an ecommerce business, then email and search programs are not targeting the "right" customer.

The marketer must have this data at their disposal. Must. Have. This. Data. The marketer has all the metrics in the world to prove that catalogs, or email campaigns, or paid search, or social, or mobile works. None of it, and I mean none of it, matters when the merchandising team is making glaring mistakes.

And then we have Seth Godin talking about creating products that are worth paying for:

You don't need a better way to talk about what you do, or a better gimmick, or a better social media strategy. In fact, you need to reinvent and rebuild what you make for a new reality, a reality where paying for something is an intentional act of buying something way better than the free alternative.

I'm sorry if this seems obvious. It's apparently not obvious to all the frustrated people I encounter who are still trying to sell the old thing in a new market.

In other words. It doesn't matter how many unique visitors you have if the problem is somewhere else (as it often is).

As a business owner, this is one of the first things that you learn. You learn that all the standard metrics in your analytics dashboard are practically meaningless. Things like bounce rates is worthless if your real problem is that you have failed to create enough new products, or if your diminishing profit is caused by you losing repeat customers. You can't fix that by using tactics that optimize your traffic stats.

So let's design the perfect analytics concept for a brand with a web shop, from a business perspective.

First step is money, not visitors

The very first thing an analytics system must be able to do is to give you a good summary of how you are doing. In other words. We need only two metrics - money as in revenue and money as in profit.

And no, I don't want to see this in real time, per day, per week, and not even per month (all meaningless at this point). I want to see it per quarter and/or by season.

So from a financial point of view, we see this:

However, looking at sales per quarter doesn't really help us understand our business from a strategic point of view. To do that we need to look at our sales per season. What those seasons are depends on what kind of product you make.

A tea company, for instance, might have a summer season, two rain seasons, a winter season, a Christmas season, and two short sales seasons.

Note: I deliberately categorized each season in relation to how people feel about it. We don't have 'fall seasons'. We have 'cold, windy and always raining seasons'. It's a little thing, but it helps people focus on the real message your brand needs to be about.

Another company selling barbecue grills might have a long winter season (no special impact during christmas), a party season (high-summer), and a cozy season (early fall).

And we are still only looking at the money part. So we take these season and analyse our sales for each one, compared to the ones that went before.

Like this:

Notice that I'm not comparing month by month because that makes no sense. Obviously, as a tea company, your sales are going to decline during the summer because of the weather, so comparing April with May is just silly.

Not only that, but your marketing and sales strategies are going to be wildly different from one season to the next. In the rain season it's all about keeping snug and cozy, in the summer it's all about pleasure.

Looking at seasons instead of months helps you understand your data in a far more useful way.

In this example here, we see a company that is failing. Since late 2012, it has lost sales for all its most important seasons. It's only improving when it's selling its products at a discount, and during the summer.

So something changed, and we can actually see exactly where it went wrong. It was just after the summer+sale season.

Things were going rather well with this company, but then some executive, who had only been looking at his analysis by month, got scared. He noticed that the sales in July+August were much lower than April+May... and he rushed to implement a new strategy. A strategy that is now failing badly.

I ask you, how many analytics systems do you know that can give you this level of clarity?

Now we come to the fun part

Now that we have the big picture, we need to start to understand why it happened. We need to learn what the failed strategy was caused by. Was it due to marketing or the products... or something else? And to determine this, we need to understand the overall product and customer patterns.

First we need to look at the products (still divided by seasons), and we categorize these into existing versus new products.

In this particular example we see that the overall sales dropped catastrophically. But it wasn't a uniform drop because the existing products did just fine. So there is nothing wrong with the old product line. The problem is solely caused by all the new products which clearly didn't appeal to their customers.

This also tells you a lot about their future, because the 'new' products today are going to be the existing products of next season. And if people don't like them today, they are definitely not going to like them later. If unchecked this problem will only get worse.

Note: This was what killed Kodak.

But we still don't know if the drop in sales was caused by marketing or the product. To learn that we need look at the behavior of the customers.

Here we see another very interesting pattern. We see that marketing has actually been doing a fabulous job at attracting attention and convincing people to convert... as the level of new customers has increased even though overall sales dropped.

However, *after* people bought their first product, something catastrophic happened and many of them decided to never return. The result also being that the positive growth of their long term loyal customers (repeat), has started dropping as well.

This company doesn't have a marketing problem. They were doing absolutely fine until they created their new strategy but then they started making products nobody wanted. But even at that, their marketing team has continued to successfully convince people to try it anyway.

Think of this in terms of normal analytics. You can't solve it by just optimizing your traffic stats. This company is very good at attracting traffic and getting it to a point of conversion. The problem they have is that *after* people buy one of their new products, they get so disappointed about the experience that they never return.

More pageviews, lower bounce rates, better landing pages, and more unique visitors wouldn't solve that. It might actually make it worse, because this company is currently alienating its market.

See how that works?

Of course, this is just one example. Another example could look like this:

It looks somewhat similar to what we had before. They are still not selling their new products, so the new strategy that was implemented in late 2012 is still failing badly.

But when we look at the customer patterns, we see a very different picture. Now, their loyal customers are buying even more than before, but their new customers have dropped catastrophically. And look at the ratio between new and returning customers. It's getting closer to each other, which means that more people decided to come back after their first conversion.

In this case, the product strategy is good. People can't get enough of what they sell, but their marketing campaign is a complete failure.

They are simply failing at attracting people's attention in any meaningful way. And in this case, because it's a marketing problem, many of the traditional analytics elements suddenly make much more sense.

The point is that if you don't have this macro-vision of what is actually going on with your company, how would you know how to fix it? You can't make that determination by just analyzing your traffic, especially not if your traffic isn't the problem.

How would you determine if sales one month were better or worse than the other if you don't take into account different customer behavior during different seasons?

In the example I used in this article, we see a tea company make a terrible mistake after being scared of a short-term drop in sales, which caused them to implement a new strategy that is nearly killing the company.

Moving into analytic patterns

After we have explored the bigger picture and now a have very clear idea of the bigger issues, we move into the more geeky world of analytics. This is the world that you and I know, where we play with the data in order to identify patterns and behavior.

For instance, if we have determined that our sales are down because people don't care for our new product strategy, we can then look at our analytics. This might tell us that during the same period we have managed to increase visitors and almost doubled pageviews per visit (much lower bounce rate).

But because we have already identified the problem on a macro level, a graph like this immediately stands out as being odd. How can we have more traffic when our sales are dropping?

We are flipping the entire analytics model upside down. Instead of starting our journey by looking at our traffic stats, we start with the result.

Then we take this result and divide it up into groups of consumer behavior. In this case by dividing it up into seasons for people who drink tea.

We then analyse each specific season in terms of customer behavior and product performance to identify the problem. And once done, we can turn to our more detailed analytics tools in order to figure out why it happened.

Think of it this way:

In the funny and excellent book, The Hitchhiker's Guide to the Galaxy, a race of people decide to ask a big computer, 'Deep Thought', to give them the "Answer to the Ultimate Question of Life, the Universe, and Everything".

The computer then turns to its analytics system and after seven and a half million years it finally found the answer, which it proudly informs the people is: "Forty-two".

"Forty-two!" yelled Loonquawl. "Is that all you've got to show for seven and a half million years' work?"

"I checked it very thoroughly," said the computer, "and that quite definitely is the answer. I think the problem, to be quite honest with you, is that you've never actually known what the question is."

"But it was the Great Question! The Ultimate Question of Life, the Universe and Everything!" howled Loonquawl.

"Yes," said Deep Thought with the air of one who suffers fools gladly, "but what actually is it?"

A slow stupefied silence crept over the men as they stared at the computer and then at each other.

"Well, you know, it's just Everything... Everything..." offered Phouchg weakly.

"Exactly!" said Deep Thought. "So once you do know what the question actually is, you'll know what the answer means."

And this is exactly the problem we have with analytics today. We start by looking at all the traffic and hope that maybe somehow we will be able to find some kind of answer to all of it. We put data next to each other, hoping that we might be able to see a correlation.

But through this whole process we forget to consider what question it is we are trying to answer. What is it that we are actually looking for?

"Forty-two" ... ahh! so that's it!

In other words, today's analytics is based on the unknown-unknown. We don't know what we are looking for, but we are still trying to find an answer for what we don't know.

It's just silly.

But by turning this whole model upside down, we move into the known-unknown. And that is about a million times more useful.

Consider the two graphs from before:

Because we started with the result and started analysing this at a macro level, we now have a known-unknown.

We know all that!

What we don't know is why. But unlike before, we now have a very specific question that we need to find an answer for. In this case that question is:

"Why do our customers not return after buying our new products?"

We have just turned an unknown-unknown into a known-unknown, which also means we can target our analysis.

How awesome is that?

Taking it one step further

One amazing thing that is happening today is that we see a huge shift in how we are doing analytics, from single funnel to multi funnel analytics.

It's a truly fascinating shift.

We are learning that our customers don't follow single paths from one point to the next. Instead they move through different stages more or less at random.

But here is the thing. This is nothing new. People have always behaved this way, long before we even invented the internet.

We have never lived in a single funnel. But it isn't until now that the digital world of analytics have started to realize that everything we have been doing so far was wrong. We have been trying to find answers by looking at single points, leading through single funnels, in a world that is not only multi-channel, but also multi-directional.

The main problem is that the analytics community used to be (and largely still is) obsessed with this concept:

It looks logical. First we have to make people aware that we exist and what options we have to offer. Then, we have to help people decide which one of those options is right for them. And finally, convince them to buy the product.

Sounds good, right?

We assume that every interaction *starts* with people not knowing what they want, and it's therefore our job to give them options that help them narrow down to the point of sale.

It's the old model of the web. The one that starts with your homepage leading to your product category pages (awareness), asking people to decide what they are interested in (deliberation), until finally choosing a product they want to buy (conversion).

But we don't create web strategies like this anymore. It doesn't reflect how people actually behave. Instead, today, we are telling people to make every page your front page. Why? Because we know that people don't follow this linear approach.

So why are we still basing our analytics on it?

Even our new fancy multi-channel funnel reports are based on this old single funnel concept. And we see this when our reports start to tell us about assisted conversions versus last click.

Here we have a multi-channel interaction that spans across multiple channels and sources. But we are still trying to group them into a single-funnel and linear flow from awareness, to deliberation, to a conversion.

Don't get me wrong. I love multi-channel funnel analytics - it's a brilliant concept. But it makes little sense to measure it this way.

Think about it in relation to the physical world:

Some people will follow the above model. They will walk into your store. They will have no idea what they are looking for, and are just milling around hoping to find something interesting to buy. This usually happens when people are looking for a present for a friend's birthday.

In this case, the model of 'awareness, deliberation, and conversion' makes perfect sense.

But, most people don't behave that way. Most people have some kind of idea of what they want before they even arrive. A person would walk into your store looking for "a pair of dark blue jeans that would work with this shirt, that he can wear for meetings at work".

In this case, the person *starts* with a conversion. Yes, starts!

This person already knows what he wants. Trying to make him aware of who you are and all the choices you have to offer is not what he is asking for. He already made his choice. What you have to do is not give him options, but to find that one product that absolutely perfectly matches what he has already decided he wants.

In other words, we have this flipped model. We start with a person who has already converted. You understand his needs, and to complete the sale you make him aware of that one perfect matching product.

But this is not how our analytics systems work today. They are all based on the old model, and they always *end* with a conversion and attribute the most value to the interaction that took place just before it (last click).

In this case, the real value happened in the very first step (first click), and every click after that had the potential to lose you the sale altogether (if you got it wrong).

But the point is not that it's one model or the other. The point is that we have to think of our interactions as being 'multi' by default. Not just multi-channel. But also multi-models and multi-behaviors.

This is why we say that you should make 'every page your homepage', and why we need to approach analytics the same way.

We can't assume that the first interaction is always one of 'awareness'. Or that the following interactions are always ones of deliberation (assisted conversions). Nor that the final interaction is more important than the others because it's the last click... and that every path ends with a conversion.

That's not how people behave. If it was, we would still be making hierarchical websites. Instead, we need to approach analytics as being multi by default. That means that instead of analyzing and optimizing funnels, you optimize for maximum effect at any point.

You assume that any point may lead to either a quest for more awareness, a decision, or a conversion (in any order). Exactly the same way as you are optimizing every page as if it was your homepage.

When people visit a page, you accept that the next action might be anything. Some people might go to another page to see what else you have. Some people might look up more information in order to make their decision. Some might leave and hopefully return another day (and not even to the same page), while others will simply buy the product right there and then.

That's what 'multi' is about. It's about giving people the best experience possible regardless of where they are. And allow people to stay in control and choose their own paths.

And remember what I said in the beginning of this article:

"A conversion doesn't end when people buy something. It ends when they no longer return for more."

Let's bring all this together: In order to really do analytics for web shops, you need to do three things:

  1. You need to *start* by understanding what questions you are looking to answer. You do that by looking at your macro analytics. How are you doing as a business, in relation to your overall strategies, for each of your important seasons (based on people's behavior), for each customer group (new, returning, repeat), by product group (new vs existing).
  2. Once you know what your real problem is, you also know what the question is. So now you have to target your analysis to figure out what the answer is. What you look for specifically depends on what question you are asking. There is no universal 'just look at this report'.
  3. In doing this more detailed analysis, you need to remember that most analytics reports today are still based on a single-funnel concept. Even those in our multi-channel reports. And when you look at the data, you have to keep in mind that people don't behave that way. We don't move from one state to the next. So you have to look for patterns, not specific data.

I admit, the last part is really hard, especially because we lack the tools to make this easy. But the analytics world is in the midst of a dramatic transformation. Almost every week we see new and amazing concepts, many of which are trying to embrace multi interactions.

The problem is that most of these concepts are still being based on the old 'awareness, deliberation, and conversion' model and this makes them somewhat useless.

But we are now seeing a new world of 'universal' analytics and similar from other services, that... well... are universal. This too is mostly based on the old ways, but it's the first step to creating analytics in which 'every point is your homepage'.

We are starting to see the future, but our reports are still based on the past.

All I can do today is help you understand these changes, make you aware of their shortcomings, and to give you pointers on how to 'flip the model' and think about analytics in a new way.

Start with the end (your results). Identify what the question is, and then look for answers in the data. Not the other way around.


The Baekdal Plus Newsletter is the best way to be notified about the latest media reports, but it also comes with extra insights.

Get the newsletter

Thomas Baekdal

Founder, media analyst, author, and publisher. Follow on Twitter

"Thomas Baekdal is one of Scandinavia's most sought-after experts in the digitization of media companies. He has made ​​himself known for his analysis of how digitization has changed the way we consume media."
Swedish business magazine, Resumé


—   analytics   —


How can publishers measure trust and other editorial metrics?


A guide to analytics for independent journalists


Why producing less news leads to a boost in subscriptions


GDPR: How publishers can track things without tracking people


Machine Learning is like black-magic for publishers


Dwell time, watch time, and the new world of audience analytics