Archivo de la categoría: EN – Data Analysis

Why revenue-oriented metrics are ultimate KPIs, and always need to be present when talking about conversion rate.

Introduction

I have been requested several times to create “dashboards” (mind the quotes) containing powerless metrics. Powerless because the client has not thought about the business question/s behind the dashboard in the right context, so data be easily misinterpreted.  

Context is so important. For example if I don’t know the impact of an improvement in the conversion rate (CR) in the revenue, then I cannot know if that’s good or bad news. In theory more CR leads to more revenue, but that’s not true necessarily. That´s why CR should go together with a revenue metric, that can be the revenue itself or at least something like average order value .

I think it was at Congreso Web, in Zaragoza where I heard from one the best analysts I know (Xavier Colomes) that he can “believe” any excuse we analysts claim to justify our lack of action. But there is an exception, he cannot believe that your boss does not want to make more money (more revenue). And I agree with him.

Why revenue is that important?

Because it can tell you whether the business (or business unit etc.) is improving over time or not. Any other metric is not that powerful. Specially CR. I have seen too many times the conversion rate going up and the revenue going down, and the other way around.

Revenue gives you the right context to analyse conversion rate.

Explanation is simple. If you sell i.e. theatre tickets, customers buy normally two tickets (couple) or let´s say four (group of friends). Always one transaction, but units and AOV (average orde value) are going to be twice bigger in the second option.

It also happens when you sell i.e. an offer for a cheap meal, that makes an improvement in transactions and CR of i.e. 10%, but in terms of AOV, or revenue could be just i.e. 2% if that´s product is way cheaper than your average. Still good, as any improvement, but the spike is not as good as we would think if we would only look at the CR.

This does not need to happen in every company. I don’t think anyone buys six pairs of boots in different colours at the same time. But units (number of product being sold) is a very important metric in some industries. 

The key idea is that focusing on just improving the conversion rate (without any context on the real impact in the business) is like focusing on improving just the bounce rate.

Why should we work to optimize the revenue and not the conversion rate?

Optimizing the conversion rate should be a means to an end, and not the end. A means to improve the revenue. Ask your CEO if in doubt 🙂

Some products (or categories, or packages etc.) may have a smaller conversion rates but generate more revenue. And we should identify them by segmenting our data, and then try to improve the conversion rate in these specific segments.

That´s why I think that when we create a dashboard or talk about CR, we should look at the evolution of the CR and its impact in the final goal of the company (that is, making money).

I am not a football fan, but will use it for a clear example.

What´s the ultimate “conversion” KPI in football?

Goals. Full stop.

What micro conversions lead to get more conversions or goals?
Let´s mention a few key football metrics: corner kicks, shots on target. ball possession, dangerous attacks, avg. kilometers per player etc.

 

 

 

Passes are very important, and as a KPI can be segmented

 

 

 

 

 

 

These metrics above may make us think that there wasn’t a big difference between both teams. Let´s take a look to goals, which is the ultimate “Key Football Metric”

Something to say? 🙂

Back to Analytics

I would have been grotesque (even more…:) that after the match, Brazil would have given some importance to the metric total shots, or claimed that their players runned more kilometres than in the previous match, so there’s a positive trend there…

Anyone cares if you have improved your “shots on target” rate if you lose 7 -1. And anyone cares about an improvement in <insert your fave metric here> if the revenue goes down. Specially the CR.

Data need context. And part of the context of a micro conversion is how the macro conversion is affected. Thus, part of the context for the i.e. customer retention metric is how the revenue is affected. Same thing, in a football match. Goals needs to be related with the metric dangerous attacks. Any other thing is pointless and may make us think that something is going well when actually it´s not (or the other way around).

We should care about the revenue and a few KPIs, like CR. But these KPIs need to be segmented and measured together with their impact on the revenue.

Two benefits of giving a special treatment to the revenue are:
– We can know very quickly if we are performing better or nor as a business
And then segment and look the other KPIs in order to understand why, and what groups of customers / products we should focus on
– We will catch & keep more easily the attention of the stakeholders
We are talking about what they care. The language of money is always understood.

Last thought

This same idea should apply when we focus of our analysis in a specific part of the funnel. We should look at the CR, or the next step to get the whole context.

It may happen that the Transition Rate (TR) from step A to step B is working better for a specific product, device etc. but the Conversion Rate (CR) is worst. Or the other way around. And that´s something we need to know.

Adobe Analytics – Cohort Analysis

I heard for the first time about using Cohort Analysis in Adobe Analytics during the talk “The Chef’s Table” from Ben Gaines at Adobe Summit EMEA in London last May (2016). Ben explained that Cohort Analysis was one of the cool things coming with Workspace.

I immediately thought that it’s an “ingredient” that should be present in any analyst’s table in which meaningful insights are to be prepared.

What is Cohort Analysis?

Wikipedia says that “Cohort analysis is a subset of Behavioral Analytics
that takes the data from a given dataset”.

To put it clear and adapt the definition to the context, I will say that a cohort is a group of users who performed a specific action at the same time.

For example: users who came to our site from a PPC campaign and created an account on the first week of May. That’s the cohort, and the cohort analysis will let us take a digging about i.e. the amount of purchase orders generated by that cohort during the next ten weeks.

It will enable us to segment a bit further very easily, and know some characteristics about those users who actually purchased (what do they have in common?) and those who don’t (again, what do they have in common?)

Why is it important?

Looking at conversions / user behaviour over time, cohort analysis helps us to understand more easily the long term relationship with our users or customers.

We can use Cohort Analysis for business questions like:
– How effective was a PPC campaign aiming for new accounts in terms of orders over time?
– How is the purchasing loyalty for a specific product category?
– What segments (cohorts) generate more revenue over time?
– How is the engagement in terms of returning visits generated by a specific action?

We can now easily evaluate the impact and effectiveness of specific actions (or campaigns etc.) on user engagement, conversion content retention etc.

And last but not least, we can apply segments, so we can focus only on a section, referrer, device etc. that is key for us.

How Cohort Analysis can be done in WorkSpace?

Just go to Workspace and select a project. Then select “Visualizations” and drang “Cohort Table” into a “Freedom Table”

 

 

 

 

 

 

A table will appear containing three elements:

  • Granularity

Day, Week, Month etc.

  • Inclusion Metric

The metric that places a user in a cohort.
For example, if I choose Create Account, only users who have created an account during the time range of the cohort analysis will be included in the initial cohorts.

  • Return Metric

The metric that indicates the user has been retained.
For example, if I choose Orders, only users who performed an order after the period in which they were added to a cohort will be represented as retained.

The inclusion and return metric can be the same. For example Orders, to watch the purchasing loyalty.


Cohort Analysis in place

I am changing the example to just visits. Among the users who visited us in a specific day, how many of them returned during the next secuencial days? (remember that this question can be about orders or any other metric/s)

For the question above, I have the table below

 

 

 

 

 

 

 

Every cell, depending on if the numbers are bigger or smaller, show a brighter or softer green colour. And we can be “curious” about both groups (what seems to be working well, to take a digging about “why” who etc. and same thing to understand what seems not to be working as expected)

To know what have in common the users behind a specific cell we want to analyste further, just right click on a specific cell and it will open the segment manager tool containing the cohort.

 

 

 

 

 

 

 

 

Save the segment and take a digging, looking to the entry page, referrer, devices etc. of this specific cohort in which in the example we are not engaging, to know more about that specific type of user

Conclusion

Cohort Analysis can help to analyse the long run. It´s very common analysing how many sales have been generated by the different campaigns and “that´s all”. But Cohort Analysis can help us to know what happened over time with customers who bought for the first time from campaign a and the ones who did for b, o who purchased product type a or b.

And that´s an improvement 🙂

And you? How do you watch your key segments over the time?

Any idea? Any comment? Any complaint? 🙂 Leave your comment and I will get back to you. You can also contact with me via email geekeandoenanalytics@gmail.com o through my Linkedin and Twitter profiles.

 

Adobe Analytics – Anomaly Detection

What´s Anomaly Detection?

Anomaly Detection is part of new & cool stuff from Adobe Analytics and provides a statistical method to determine how a given metric has changed in relation to previous data”

Is an anomaly the same thing as a spike or a dip?

Not exactly 100%

A spike or a dip is what happens when a metric dramatically increase or decrease
for a specific period of time. And it might be “created” or “expected”.
For example, if we run an extra £10000 PPC campaign, then it’s normal we will have an increase in traffic (due to that campaign). Thus, if we have 20% more of traffic and 17% more of conversions, that’s not an anomaly, just a spike.

An anomaly is more about the way that metric has changed and has an statistical approach.

For example, if one day 23% of the orders come from a specific campaign that represents just 3% of the traffic, that’s an anomaly, but can also be a spike or not.

It worth taking a digging, and the results are statistically significant (it’s highly recommend to thick the box “Show Only Statistically Significant Items”)

As we can see in the graph below we can see that there is an anomaly on the 29th of June, but it’s not really a spike
1

How can we get started with Anomaly Detection?

1- Anomaly Detection can be found within “Reports”, and then Site Metrics

2

2- Select the metric/s & the period

Just click on “Edit Metrics” and then choose a “Training Period”

  • Metrics

You can select one or more metrics (so you can see the relation between i.e. two metrics)
You can select every Success Event, and also the Standard Events related to eCommerce (cart additions, views, removals, orders etc.)

  • Period

3

 

 

 

The three training periods available are: 30, 60 and 90 days. Note a bigger training period may reduce the size of an anomaly.

3- Take a digging for a specific anomaly

Once you select the metric and timing, you will see a graph showing the evolution, pointing out the anomalies
4
As soon as we click on an anomaly, we see below the graph the actuals and what would be reasonable for that metric during that period of time. Additionally, we also see its impact on percentage (in green if it’s positive and in red if it’s negative).

Then we should click on analyze (above the graph) to see the “contribution analysis”

4- Check the possible reasons

5
Adobe Analytics suggest a range a “items” (that can be product, campaign etc.) in which an anomaly has been spotted.

Each posibitility has a contribution score that take values from 1 to -1:
1: complete association for a spike or complete inverse association for a dip
0: No association for contribution
-1: Complete association for a dip or complete inverse association for a spike.
In the image can see in the second row: 1% of x has generated 23 of y…

5- Create a segment and inspect it

6

Just click on one of the items (rows) and a button to create a segment containing that item (product, campaign, referrer etc.) will appear.

Next steps? Save the segment and apply it by referrer, device etc. in order to take a digging and know what´s going on..

As you can see, it’s very fast to identify what’s “unusual” and the segments we need for our analysis, and it will save us loads of time.

Any idea? Any comment? Any complaint? Leave your comment and I will get back to you. You can also contact with me via email geekeandoenanalytics@gmail.com o through my Linkedin and Twitter profiles