E-commerce Internet marketing: The crime and penalty of ROI (the next)

Intermediary transaction SEO diagnosis Taobao guest Cloud host technology Hall

"Each issue" the human only super-delegates of the fear time side to IS Inc. to be brave. Man can be brave only by fear. )--The Song of Ice and Fire

"What does this article talk about?"

Multi-dimensional subdivision flow channel value measurement including ROI and indicators

Interaction between flow channels and its impact on flow channel evaluation

Case analysis of assists flow and score flow

Deep excavation of the relationship between attribution modeling and flow

"Preface" (This article is invited by Netconcepts CEO Allen). The last part of this article see: http://www.chinawebanalytics.cn/roi-sin-1/

Since 2008, we have had a problem that has been difficult to articulate, and the problem has been haunting us for a long time. We explain this problem for the time being by "fuzziness"-that is, it is impossible to trace through technical methods to obtain accurate data and accurate analysis. This kind of problem still exists until today, for example, to explain the exact impact of an offline marketing campaign on website Traffic, we are far from "precise" and can only approximate and accurately explain the impact by indirect means. But 2008 years of this problem, in recent years, through simple technical methods and the upgrading of tools very good solution. The question is the assessment-do we need a portal like Sina as a traffic source? What are the values of these portals? For customers, these portals look like traffic is not exaggerated, and it seems that the quality of traffic is not ideal, what is the value of them? But the customer eventually used these flows for a long time, although we did not have sufficient proof of their full value.

Until the time comes 2010 years ...

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

Body】

In the upper part of the article, we emphasized that ROI is not the only KPI that deserves attention, and the higher the ROI, the better. We need to first identify what we want, because the pursuit of ROI is definitely not a goal, but a means. In addition, the application of ROI, from a macro point of view (similar to the micro-economic research methods), seems to be more useful. This does not mean, however, that ROI is meaningless at the microscopic level, but that we have to look at ROI in a different way than in the past. What's the difference? This is the question that this article wants to explore. To figure this out, let's take a look at how traffic performance approaches evolve.

Measurement of subdivision channels of single latitude

I interviewed a lot of clients and asked them how to measure the performance of different traffic, most of the answers are the following tables. For example, customers who do not have a part of E-commerce to brand marketing, they measure the flow performance of the table is the following (the data is purely fictitious, if there is a coincidence, purely coincidental).

This form can tell us that some of the source of traffic benefits (effectiveness) is not particularly ideal, such as Sohu, and the use of Baidu is very cost-effective. In fact, it's a good way for CPA to measure traffic performance, it has no difference in nature from ROI, except that the R (return) becomes a specific transformation event (registered in this case), while the formula is cost-specific--but the ROI is compared to the cost (investment) .

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

Of course, some customers are relatively simpler, and they may not define an important transformational event as an action, so they can even save a CPA, just to calculate the CPV (the cost per flow) or look at bounce rate. But I do not advocate this method, too lazy, after all, the amount of traffic is only a measure of the flow of a property, and through some specific behavior or transformation to measure the quality of traffic, it is more important.

If it is an E-commerce site, the table will be slightly changed (the data is fictitious, if there are similarities, it is a coincidence):

See this table, we must be enlightened, hey, this is not our company use it!

Yes, 10 of the E-commerce companies I've seen have nine use this table. I can't say what's wrong with this watch, especially if you have the ability to do this on a daily basis (the tools developed in-house are actually not difficult), and it can quickly position the flow source of ups and downs. This kind of form in fact the bosses are also fondle admiringly.

But it always feels like something is missing. In accordance with the logic of these two tables, we must immediately stop to such as "Sina", "NetEase", "Sohu" the flow of purchase, and increase Baidu PPC, Baidu SEO and Baidu EDM flow. How many E-commerce sites do this? Or, how many of my previous clients really stopped "Sina"? It's strange that people are painfully seeing their "performance" so low, but they "have to" continue to "buy" them. "That must be because of the need for PR!" A voice came over. Maybe, but not quite. If it is purely for PR, we can actually have better four and two ways to dial the jack. They may have other value in their traffic channels, which have very high or low ROI. So I used another model to understand their value. Please keep looking back.

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

Multi-dimension Subdivision channel measurement

This model is not complex, based on one of the following assumptions:

All traffic channels are different, because the marketing process contains various stages of value, namely awareness (cognitive) –> interests (interest) –> preference (preference) –> Purchase (purchase) four stages. The value of each channel at each stage is different. In some stages, the value of this flow is greater than that, but in other stages the value of that flow is likely to be greater than the traffic.

In fact, this assumption is entirely "nonsense". Isn't it? If the ability to transform than the final, Sina in any case can not exceed Baidu "brand area" (or the search for brand words), but to discuss and increase brand awareness, Sina home best place to do a period of advertising, certainly more than just pay to buy Baidu "brand zone" more reliable. According to the process logic from the awareness (cognitive) –> interests (interest) –> Preference (preferences) –> Purchase (purchase), some traffic channels may only help to increase awareness or interest, or most help form a preference, But not enough to eventually buy, we cannot say that they must be worthless, and perhaps only in the early stages of the marketing process. This truth is very simple, we through a portal ad first contact with an E-commerce site, we can not be in this site on the purchase of a single item (Taobao on the other, we would like to see a merchant on Taobao for the first time because we are familiar with Taobao rules and recognition), But we may be interested in this site, and after some time through other channels (way) to the site again. Portals are only at an early stage in helping to transform our marketing process, but they are still valuable.

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

How do you judge this value? This is the question that haunts me for a long time, I use the following method. This method is based on another assumption:

If some channels can bring the early value of the marketing process, it must be able to express the user's behavior behind the traffic. These behaviors reflect your interest in the content of your website.

This assumption is another "nonsense". Interested in your site, you should always click on the link, read more pages, stay a few minutes. These actions are known as indicators. Indicators seems to be more than just the words that young women like.

(If you want to know what indicators is, click here.) Sometimes, for simplicity's sake, we boil down the indicators to three metrics: page view/visit,time on site and bounce rate. In this way, we measure the value of a traffic channel to broaden the integration of ROI (or CPA) and indicators. The following figure (the bubble size is the amount of flow, author note):

The indicators of the above figure makes the mathematical conversion of percentile, selects bounce rate,non-bounced time on site and non-bounced Pv/visit as metrics to measure bounce and non bounce in a more comprehensive way. There is no more detailed explanation, if you want to know how to convert, please give me a message in the message box at the end of the text-if the demand is large, I will write a small article.

It is not surprising that there is a higher indicators flow source that generally has a better ROI. But for analysts, the unexpected outliers are doubly exciting. For example, we see those with high indicators scores but low ROI (pictured in Tianya and Sina), which show the characteristics of value in the early stages of marketing. And if we can compare the proportion of new flows that they bring, we might be more sure of that. On the other hand, in the lower right corner of the "Beautiful said," from where the flow seems to your site has already been familiar, at the bottom of the marketing process, it's the flow is probably "old fried dough sticks".

The new and old proportions of traffic confirm our view:

The increased indicators latitude broadens our understanding of flow. If only the ROI is the standard, then Tianya and Sina car channels must be excluded from the purchase flow, but with the indicators, coupled with the ratio of new and old visitors to do the reference, we know that these flows are valuable. Just, I still can not clearly explain how much value they have, indicators not money, it is difficult to be monetize (that is, translated into the actual value of money), so do this step, we can only say that we know it good, but in the end how good, we can not tell.

Alas ... I hate the fucking ambiguity.

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

Interest reading: I summed up the marketing phase, the various stages of the appropriate flow, and how to measure these flows.

Interaction between flow channels

The time has finally arrived 2010 years, this year I joined the Adobe Omniture Business, Omniture also in this year the violent promotion their marketing channels. The role of this multi-channel is the same as that of today's GA, which is used to help us understand the relationship between flow channels. The appearance of this report, let me just can't help cheering for a while, I think the outstanding "fuzziness" problem in the past years is finally expected to solve!

If we do the marketing on the front lines for a while, we know that we are not dealing with traffic, but dealing with people. The nature of the traffic we buy is actually in the "buy people". The so-called a traffic is very effective, is actually these traffic behind the user and our marketing purposes is very consistent. With that in mind, we know that the idea of measuring separate channels in the past is not right. Reason without him, people do not cling to the internet on a website, but East tour west stroll, so people are very susceptible to a variety of marketing channels of the common influence. In class, I always cite this example:

I want to buy a printer, so I searched Google for a while, found a printer, and click on Google's link to enter the sales of this printer's website, but I did not order. After a few days, I saw Sina on the printer's advertising, so I clicked on the ad, found that I once again came to this site, but I still do not order. A few days later, I received an EDM from this website and told me that the printer was being promoted, so I clicked on the link in the email to the website and I still didn't order it. Until two days later, I realized that I would not buy this printer promotional activity is going to pass, so I search the name of the site in Baidu, and then finally entered the site to purchase the next single printer.

Now, the question is, what traffic channels do I contribute to this transformation on this site? or if they all contribute, then which contribution is greater?

This example is a very extreme example, I know a lot of people don't wait so long to buy a printer as I do, but something like this--for example, I saw the ad for this product, and I went to the website to get to know the product, but I didn't order it, but I decided to buy it the next day. And then through Baidu to find the site after the order of the situation-is it universal? If this is the case, then at least it solves some of the problems in front of me--those with very good indicators but little ROI levels, and they may end up with some other channel-like Google, Banner banner, and EDM, as in the example above ( For example, in the above picture of Baidu)-because before we statistical channel conversion contribution, all only statistics the last Channel (consortium) role. If this is a common situation, it is proving that these flows are actually valuable.

Google Analytics's multi-channel funnels has given us a good answer. If you open this report and choose the first subreport-"Overview"-you can see how much of the conversion in all of the transformations that spanned multiple channels accounted for the total transformation. Through this report, I found that the situation of each site is not the same, some sites generally exist a transformation across multiple channels of the situation, but some sites do not have any channels between the interlinkages.

The picture above is a case of a foreign electric dealer's website, the overlap of the pie chart tells us: More than half of the transformation has a number of traffic channels to intervene together-the site of the same visitors through different channels of traffic access to the site is very common.

Another Web site (pictured above), the level of overlap of pie charts is very interesting-they have little overlap. This means that users of this site basically only through the fixed flow channel type access to the site.

Therefore, I think the following conclusion is very important: for some sites, for the conversion of the site to contribute to the flow of channels there is not much correlation between the use of a single channel measurement is reliable. But for some other websites, their final transformation to a large extent by the impact of a number of traffic channels, so we for these traffic value of the determination of--roi or, or bring the transformation is good, can not be separated by a single channel to measure, but should be the interaction between the channels to take into account. So before we measure the performance of traffic, we should first understand whether there is a correlation between these flows and how well they relate to each other. This overturns the "classic approach" we used to go straight to the analysis of a specific flow. As shown in the following illustration:

Past methods:

Now the method (I have to temporarily hide the next few steps):

We need to change the "worldview" of the flow channel. But even if we knew there was a correlation between the flow channels, what would it be?—— keep looking.

Assists and scores, the flow value is overestimated and underestimated

If the flow of traffic is determined to have considerable interaction, then the ROI analysis we face will be completely different from the previous one. In the past, we could not directly determine the value of those flows that could not directly generate order conversions, but today we may be able to open the black box. Since GA can tell us the extent of the "overlap" between the flow of transformation, it should go a step further and tell us what the relationship is. At this point GA's multi-channel report did not disappoint us, multi-channel funnels reports below are used to help us solve these problems. For example, the Assisted conversions report tells us how a traffic channel can be "assists" or "scores". This is an interesting report that helps GA to first differentiate between the direct conversion value of traffic (scoring) and the value of helping others to gain transformation (assists). In the example above, the channel for assists is Google, the banner on the portal, and the EDM, while the scorer is Baidu. As shown in the following figure.

Assisted conversions Report using data to describe the above situation, assisted the word "assists" in English. For example, the following example:

The Assisted conversions list shows how many times the channel assists others (including themselves) to convert, while the last consortium conversions means the number of conversions (scores) that occur directly to them. High assists/scoring ratios (those in the red box) mean that these channels are more "selfless". On the contrary, there are some channels of assists/scores compared to 1 lower, which shows that these channels have a greater chance of conversion as the final scorer, they are marketing value process in the "net Channel." A channel, if it is good at collecting nets, it is valuable, another channel, if it is able to assist, it is also valuable. What we are afraid of, however, is that we do not see these assists, or even let the value of these channels be ruthlessly underestimated. Now, let's look at a real example:

This is a traffic channel on the site performance, traffic is about 50,000. It can be noted that the relevant indicators metrics on the site (pv/v,time on site,bounce rate) are pretty good. However, if it merely calculates its contribution as a "scorer" (last Interactive conversions), it only brings 158 conversions, a conversion rate of not more than 0.3%, very unsatisfactory, and it is simply impossible to match the indicators with which it is so strong. However, its "assists" (Assisted conversions) Report card is 643, assists conversion rate is 1.3%, and the site's overall conversion rate is almost the same. With this data, I was thinking that if the bosses cared only about the final finishing touches, then I'm afraid Harvey (Xavi, the Spanish International, is known for assists, the author notes) would not header or be Iniesta so fast, or even Beckham might not have played! Just use the final ROI to judge the effect of traffic, we will underestimate how much of the traffic is useful!

So this is the biggest problem with the second way of "success", as we said in the previous article. In this way, we first sweep the highest ROI traffic, then we look for the high ROI traffic, then we look for a slightly lower ROI, and so on, but when we find the original high ROI and low ROI traffic (both with last consortium Conversions calculation) is in fact inextricably linked to the time, the second way of success is completely inapplicable.

This may be why we sometimes take away the seemingly "innocuous" but expensive traffic, thinking that we are "saved", but surprised to find that the income also strange decline in the reason.

If this story is over, you will feel a little less. Since we already know that a certain traffic is better at "assists", while others may be better at exploiting other traffic "assists," if you have curiosity, you will ask, who did this traffic assist to? Other traffic, who assists and scores?

Keep your curiosity, it pushes the world forward.

Attribution modeling or case by case (concrete analysis of specific issues)?

I have always had the same curiosity, and in the past we could not see the assists, I really want to know why a flow has such a good indicators, but only poor transformation. Now, my curiosity is--those who are good at assists and who give their assists to score? Who is good at scoring traffic, and who assists? Harvey assists Lionel Messi or Pedro or his national team mate Fernando Torres? The good news is that GA can at least help us satisfy this little curiosity.

Go to the top conversion Paths report, and then select the Source/medium Path report in primary dimension, and you can see a detailed breakdown of the assists and scores. In the above example, a number of Sina's advertising and Baidu has a lot of association. I analyzed these associations, and the creative relationship with the advertising, "Sina/yuleyouceqingtianzhu 1/2" The idea of the advertisement is very good highlighting the site's brand, so that people in Baidu can continue to search. But what's interesting is that another advertisement on Sina, which is not related to other media, is a research idea that highlights promotional information about specific items, but does not mention the brand of the site. I'm sorry I can't show the creative screenshots because of the business confidentiality. But this kind of thinking can tell us what is the relationship between the media and what might cause it. It was hard to imagine in the past.

The path of assists and assists is the most hotly discussed area in the industry today, the attribution Modeling (Contribution model) issue. The research of contribution pattern is used to help us to find the value of each traffic channel in assists through some fixed algorithm. Attribution modeling has a number of common computing patterns that are easy to understand.

Last touch credit: The final (scorer) of the assists path has the greatest value, before the value is calculated;

First Touch Credit: The number one in the assists Path (consortium, Google in the previous example) has the greatest value, followed by no value;

Linear: Average distribution value;

Time Decay: The first one has the greatest value, then descending after descending;

Position Based: The first and last value is the largest, the more to the middle value the smaller.

These algorithms are designed to help us more easily compute the value of the media, however, I think that this precise (but not accurate description of the facts, see this article for accurate and accurate discussion) is only useful for the performance of each subdivision of the evaluation marketing, or to help us get a "maybe real" The media channel ROI to make it easier for our traffic noun (suppliers) to checkout, but it is difficult to help us carry out an effective analysis. If we really want to analyze and optimize, I would suggest case by case, open the report I showed you above, do the filtering, subdivide, see what we really care about and what happened between them.

You ask--if you don't use a fixed model to do calculations, how do you know who is more valuable? The question itself, in fact, did not jump out of the old thinking. For the determination of the flow value, we must establish who is more valuable than who, but now, we find that the value of the flow is probably because they are together to truly manifest. We can no longer measure the ROI of a single traffic in isolation, but we should look at them in a more holistic perspective. are isolated traffic valuable? I don't know, because they're not isolated.

Finally, the current analysis of the flow of new ideas are all uncovered.

Conclusion

The marketing process value path helps us to measure the value of traffic with a new vision, and the ROI approach looks pretty naïve in these new, more realistic models. ROI of sin and penalty, in fact, the ROI is not innocent, but if the ROI as the only truth will surely be punished. However, our understanding of online effect marketing is still in a relatively basic situation, many problems are still difficult to figure out, for example, the relationship between traffic (i.e., the traffic path of the marketing transformation mentioned above) must depend on the user's clicks on the traffic entry (for example, clicks on ads) for GA, but the real thing is that Sometimes we don't click on the ad, we just see the ad and it makes a deep impression, and then we search the search engine for something interesting in your impressions. But unfortunately, impression we still do not have a truly revolutionary technology to open the black box (although many people claim they can), or rely on technology to solve the problem of impression itself this may be wrong. Because, the advertising "display" is not equal to "people's impression", and the impression of this psychological level of things, not we use cookies can be said clearly.

Another big challenge is cross-platform tracking, where it is difficult to accurately record a user's path from INTERNET/PC to mobile (pad, mobile, etc.). But people use mobile devices more and more, if we can't accurately describe the process, how can we better analyze and optimize it? In any case, our today is stronger than yesterday, the future will always be better.

To be curious, to try!

Old way, have any idea, or question, please give me a message. Thank you!

This article copyright return "the website Analysis in China CWA" and its author sing, wants to reprint, please contact the author

E-commerce Internet marketing: The crime and penalty of ROI (i)