parking Getting Dirty in the Domain Data

[Whizzbang]

We’ve recently been working with a client to better understand the underperformance of a number of their domain names compared to a few months earlier. What we discovered when we conducted an internal forensic analysis of the data was quite surprising.

The domain we will look at in this article has been renamed to A.COM for privacy purposes. Since the beginning of the year it has had a monthly revenue ranging from $183 to $1221 and a normalised RPM (revenue per thousand visitors) of $75 to a high of $488. So what was going on with this domain?

Upon closer inspection we found that A.COM was a domain from the travel industry. People were looking for the services the domain offered from May to July and this dramatically pushed RPM rates higher as advertisers competed more aggressively for the traffic during this time. In fact, the peak numbers were achieved by a direct advertising travel company.

Read More >

 

[frank-germany]

and now : where do you send the traffic to in future and at what percentage
knowing you lose
if you just rotate evenly?

 

[Whizzbang]

This is an interesting question and one that is easily calculatable.
If you split the traffic equally each month across every source then you end up with a total revenue of $1611 for the year versus $2217 or 38% less revenue overall.

There are a lot of strategies around sampling but they basically boil down the single question of what did the information cost? In other words, if I was earning one dollar with one company and then sampled another company and found they were paying 90 cents then the information cost me 10 cents.

Minimising these "information costs" is crucial to optimisation and every single domain requires and different regime to do this.

 

[Hypersot]

This is an interesting question and one that is easily calculatable.
If you split the traffic equally each month across every source then you end up with a total revenue of $1611 for the year versus $2217 or 38% less revenue overall.

There are a lot of strategies around sampling but they basically boil down the single question of what did the information cost? In other words, if I was earning one dollar with one company and then sampled another company and found they were paying 90 cents then the information cost me 10 cents.

Minimising these "information costs" is crucial to optimisation and every single domain requires and different regime to do this.

exactly
that's why I always say that, not every domain is fit for optimisation

 

[frank-germany]

This is an interesting question and one that is easily calculatable.
If you split the traffic equally each month across every source then you end up with a total revenue of $1611 for the year versus $2217 or 38% less revenue overall.

There are a lot of strategies around sampling but they basically boil down the single question of what did the information cost? In other words, if I was earning one dollar with one company and then sampled another company and found they were paying 90 cents then the information cost me 10 cents.

Minimising these "information costs" is crucial to optimisation and every single domain requires and different regime to do this.

yes I know it an interesting question
but I can't see you answered it

 

[Whizzbang]

Actually, every domain can be optimised. One of the inputs into sampling is that each domain needs to its own unique time period over which sampling can take place. For example, low traffic domains should be sample much less often than a high traffic domain to minimise the cost of information. But like I said, this is one of the inputs, there are many other factors that need to be considered.

 

[Whizzbang]

yes I know it an interesting question
but I can't see you answered it

There is a lot of information that appears on the fly that can be used in the optimisation process. A simple example is the IP address and geo-locating it. You can also combine this with external factors such as weather. For example, (these are a contrived example for illustration purposes) if I know it's raining in New York then I am less likely to display advertisements about sun screen compared to if I the person is from Melbourne Australia in the summer. Likewise, if the person is going to a sports domain and is from Melbourne Australia and it's summer then display information about cricket not football as it's the off season in summer for football.

Once you broaden your view of data then there are many things you can do to increase your optimisation potential to extract additional value from the domain traffic.

 

[frank-germany]

Actually, every domain can be optimised. One of the inputs into sampling is that each domain needs to its own unique time period over which sampling can take place. For example, low traffic domains should be sample much less often than a high traffic domain to minimise the cost of information. But like I said, this is one of the inputs, there are many other factors that need to be considered.

My friend we are talking about 1exact domain and its data

Now what is the strategy on that domain

Now that we have the data
Whats next

Evaluation has taken place already
If that data can t lead to a decision
Its useless

 

[frank-germany]

There is a lot of information that appears on the fly that can be used in the optimisation process. A simple example is the IP address and geo-locating it. You can also combine this with external factors such as weather. For example, (these are a contrived example for illustration purposes) if I know it's raining in New York then I am less likely to display advertisements about sun screen compared to if I the person is from Melbourne Australia in the summer. Likewise, if the person is going to a sports domain and is from Melbourne Australia and it's summer then display information about cricket not football as it's the off season in summer for football.

Once you broaden your view of data then there are many things you can do to increase your optimisation potential to extract additional value from the domain traffic.

 

[Whizzbang]

My friend we are talking about 1exact domain and its data

Now what is the strategy on that domain

Now that we have the data
Whats next

Evaluation has taken place already
If that data can t lead to a decision
Its useless

Based upon the data, decisions are literally being made on a milli-second by milli-second basis. We use dynamically changing data from multiple inputs to alter not only the routing decisions of traffic but what is displayed on the page and which advertisers are engaged.

As an example, we track over 250 different metrics for every domain every day and we process this data to alter how the traffic is routed. Layered over the top of this daily data we then incorporate the dynamic data that I mentioned in my previous post.

Everything must lead to a decision....otherwise it's just intellectually interesting but pointless.

 

[frank-germany]

Based upon the data, decisions are literally being made on a milli-second by milli-second basis. We use dynamically changing data from multiple inputs to alter not only the routing decisions of traffic but what is displayed on the page and which advertisers are engaged.

As an example, we track over 250 different metrics for every domain every day and we process this data to alter how the traffic is routed. Layered over the top of this daily data we then incorporate the dynamic data that I mentioned in my previous post.

Everything must lead to a decision....otherwise it's just intellectually interesting but pointless.

exacly
so on that domain
where would you prioritise the traffic
based on that data?

 

[Whizzbang]

exacly
so on that domain
where would you prioritise the traffic
based on that data?

Hmmm.....let me see if I can answer the question.
Let's imagine a piece of traffic comes in from the USA right now. It's initially offered to realtime advertising bidding networks that try and "win" the traffic in an auction process. Assuming they aren't offering enough for the traffic it will most likely flow through to Voodoo as they are paying the most for the traffic at this point in time with a normalised RPM of 174.84.

I should say that the data I provided in my blog post was aggregated to months......we typically work on seconds or at the most days. This means that routing is a lot more dynamic to the solution that will pay the most for the traffic.

 

[frank-germany]

Hmmm.....let me see if I can answer the question.
Let's imagine a piece of traffic comes in from the USA right now. It's initially offered to realtime advertising bidding networks that try and "win" the traffic in an auction process. Assuming they aren't offering enough for the traffic it will most likely flow through to Voodoo as they are paying the most for the traffic at this point in time with a normalised RPM of 174.84.

I should say that the data I provided in my blog post was aggregated to months......we typically work on seconds or at the most days. This means that routing is a lot more dynamic to the solution that will pay the most for the traffic.

here is what we got as data:

SE 53.93
DS 48.56
VD 80.10
BD 83.61
PC 64.01
Ad Networks 593.54


so if you have no direct advertiser you send the traffic to voodoo right?

why not Bodis?

do you think you lose much in the course of a year
when you only rotate Bodis and Voodoo
based on that data?

 

[Whizzbang]

Take a look at the monthly data. It changes a LOT from one provider to another. In fact, the winning solution is highlighted each month in the last table I provided.

What you quoted in your post was the average data for 10 months. If you only routed the traffic on a 10 month basis then you would have a VERY sub-optimal result and lose a lot of money.

 

[frank-germany]

Take a look at the monthly data. It changes a LOT from one provider to another. In fact, the winning solution is highlighted each month in the last table I provided.

What you quoted in your post was the average data for 10 months. If you only routed the traffic on a 10 month basis then you would have a VERY sub-optimal result and lose a lot of money.

are you kidding?

if you present 10 month data -
we talk about 10 month data right?

 

[Whizzbang]

Not at all.....the traffic dynamically routes differently every single day.

The top row in the table was simply to display who was winning on average over the ten months. If you only route to the average winner then you will lose massively overall.

If we routed ALL of the traffic through to Voodo and applied Voodoo's payout rates each month then Voodoo would have paid out $436 for the ten month period. The domain actually earned $4531 for the same period of time. The reason for this was a combination of an advertiser paying a lot for the traffic in May-Jul and other parking solutions beat Voodoo the majority of the time.

I think what you are wrestling with is the concept of a spot price versus an average price. We largely deal in spot prices (what will be paid now). What you are getting focused on is the average price over a long period of time.

Here is the tables that we are talking about for other readers: