We've been busy at Inuvo building all sorts of new models. I'll talk about those in the next few posts. I also had a new baby girl, so its been busy around here!
So how do predictive analytics work? The basic idea is to build a statistical model that looks at a variety of input data and produces a numerical score as the output. The input data could be the age of the person, their gender, their income and their occupation. The output could be the probability that they buy a Toyota Prius. The input data could be the time of a click on an advertisement, the number of previous clicks from the IP address and the click through rate on that ad. The output could be the probability that the click is fraudulent.
Most simple predictive models are a mathematical equation that combines the various input data to compute the output value. Numerical data, such as age, might be input simply as a number into the model. Categorical data such as occupation might be grouped into groups, and then a numerical weight assigned to each group. There are dozens of techniques for managing different types of input data.
But how do these equations get built? The simplest way to build predictive models is to take a large set of past data where we know the outcome. At Inuvo, we recently built a very powerful fraud model by examining many months of previous transaction data. Because time had passed on that data, we knew exactly which transactions turned out good, and which turned out bad. We used this good/bad data to build the equation inside the model, and now it can monitor a publisher's transaction traffic, and predict which transactions are fraudulent.
How accurate is it? Well, like any statistical system, it only prescribes a probability of a transaction being fraudulent. But every day we review the output of the fraud model, checking through the most suspicious activity. We've already shutdown a handful of bad publishers, including some that were caught after just four transactions! That's pretty powerful.
Building predictive models has all sorts of pitfalls and techniques, smart rules of thumb and clever approaches to use under different circumstances. Over the next few posts, I'll talk about some of the ones we use at Inuvo, why and what the benefit is to our customers.
Monday, August 16, 2010
Tuesday, March 2, 2010
Predicting The Future
"DISCLAIMER: Past results may not reflect future performance"
How many times have you read that? The truth is that past behavior is highly predictive of future performance, at least in a statistical manner. The science (and art) of predictive analytics is all about using past data to better predict the future... and it works amazingly well.
There are some great examples at play today. One of my favorites is credit card fraud detection. Most banks today have such systems. Every time you make a transaction, the system compares that transaction to your past behavior as well as to known fraudulent patterns. In a fraction of a second, these predictive analytic systems compute the probability that your card has been stolen. Transactions that exceed certain risk thresholds are immediately flagged, and a bank employee will call you to check to see if everything is okay. These systems have stopped over half of all credit card fraud, often within the first few transactions. Because these systems work in real time, they have even aided in the apprehension of thieves at the checkout!
I have had the pleasure of working on the analytic engines within these systems. The pattern detection algorithms are extremely interesting, but they must remain secret. However, I can tell one interesting story. Humans are terrible at making up numbers. So when a fraudster tries to create a "seemingly random" dollar amount, there are certain psychological behaviors that make some digit probabilities more likely than others. And using truly random numbers also doesn't work because legitimate business transactions do not follow true randomness either. This fact alone can be used to help predict whether a transaction is fraudulent or not. In modern fraud predictive analytics, there are literally thousands of such statistical qualities that are used in the models.
Predictive analytics are used in myriad areas of commerce. Every checking account you have can result in up to a dozen or so predictive models run every month. They test for everything from your profit potential for the bank to likelihood of going into bankruptcy, from likelihood of leaving for another bank, to likelihood of accepting an offer of overdraft protection. Every time you swipe your credit card when you checkout, between the time you swipe the card and the "Approved" response (which is usually 2 second or less), a couple of companies thousands of miles away run predictive analytics to make sure you are the card holder, and if you are near your credit limit, is it worth approving you beyond it, and even is this merchant who they say they are.
Online auction sites and online payment companies use similar fraud and risk predictive analytics on every transaction. You will be surprised to learn that most of what we call "junk" mail is actually run through predictive models. Marketing is tough, because response rates are always so low, but if predictive analytics can distill responders from 0.25% to 1.0% the profitability of that campaign just went up by a factor of four! Not to mention, that is a lot of trees saved.
Predictive analytics are used everywhere in our lives. Every time you visit the doctor, paperwork gets filed on which tests you did and which drugs were prescribed. A number of different predictive analytics are run to see if the doctor is over-billing, and even if the doctor's billing address is correct.
Next time you buy groceries, turn over your receipt and look at the coupons printed on the back. These were selected for you, in part by a predictive analytic system. If you shop at major retailers like Walmart, Target, Best Buy and so on, you will notice that they almost have everything in stock. Their supply chain management systems are packed full over predictive analytics that monitor sales data, social trends, weather, product lifecycles and many other factors to predict exactly how much of each product to send to each store. See a new CVS, McDonalds or Starbucks opening near you? Predictive analytics were used to asses the viability of that site for that business.
So how does these ever-present predictive analytics work? That is what I will cover next post. However, I will say that predictive analytics are still very new in online marketing, which is why I am very excited about where Inuvo is heading. More on that later!
How many times have you read that? The truth is that past behavior is highly predictive of future performance, at least in a statistical manner. The science (and art) of predictive analytics is all about using past data to better predict the future... and it works amazingly well.
There are some great examples at play today. One of my favorites is credit card fraud detection. Most banks today have such systems. Every time you make a transaction, the system compares that transaction to your past behavior as well as to known fraudulent patterns. In a fraction of a second, these predictive analytic systems compute the probability that your card has been stolen. Transactions that exceed certain risk thresholds are immediately flagged, and a bank employee will call you to check to see if everything is okay. These systems have stopped over half of all credit card fraud, often within the first few transactions. Because these systems work in real time, they have even aided in the apprehension of thieves at the checkout!
I have had the pleasure of working on the analytic engines within these systems. The pattern detection algorithms are extremely interesting, but they must remain secret. However, I can tell one interesting story. Humans are terrible at making up numbers. So when a fraudster tries to create a "seemingly random" dollar amount, there are certain psychological behaviors that make some digit probabilities more likely than others. And using truly random numbers also doesn't work because legitimate business transactions do not follow true randomness either. This fact alone can be used to help predict whether a transaction is fraudulent or not. In modern fraud predictive analytics, there are literally thousands of such statistical qualities that are used in the models.
Predictive analytics are used in myriad areas of commerce. Every checking account you have can result in up to a dozen or so predictive models run every month. They test for everything from your profit potential for the bank to likelihood of going into bankruptcy, from likelihood of leaving for another bank, to likelihood of accepting an offer of overdraft protection. Every time you swipe your credit card when you checkout, between the time you swipe the card and the "Approved" response (which is usually 2 second or less), a couple of companies thousands of miles away run predictive analytics to make sure you are the card holder, and if you are near your credit limit, is it worth approving you beyond it, and even is this merchant who they say they are.
Online auction sites and online payment companies use similar fraud and risk predictive analytics on every transaction. You will be surprised to learn that most of what we call "junk" mail is actually run through predictive models. Marketing is tough, because response rates are always so low, but if predictive analytics can distill responders from 0.25% to 1.0% the profitability of that campaign just went up by a factor of four! Not to mention, that is a lot of trees saved.
Predictive analytics are used everywhere in our lives. Every time you visit the doctor, paperwork gets filed on which tests you did and which drugs were prescribed. A number of different predictive analytics are run to see if the doctor is over-billing, and even if the doctor's billing address is correct.
Next time you buy groceries, turn over your receipt and look at the coupons printed on the back. These were selected for you, in part by a predictive analytic system. If you shop at major retailers like Walmart, Target, Best Buy and so on, you will notice that they almost have everything in stock. Their supply chain management systems are packed full over predictive analytics that monitor sales data, social trends, weather, product lifecycles and many other factors to predict exactly how much of each product to send to each store. See a new CVS, McDonalds or Starbucks opening near you? Predictive analytics were used to asses the viability of that site for that business.
So how does these ever-present predictive analytics work? That is what I will cover next post. However, I will say that predictive analytics are still very new in online marketing, which is why I am very excited about where Inuvo is heading. More on that later!
Thursday, February 25, 2010
Business Rules
Last post I talked about reporting analytics - the process of data tabulation, data aggregation, and other reporting techniques. In the online marketing world, reporting analytics tell us page views, unique visitors, impression volumes, click through rates, conversion rates, time on page and other interesting data.
But acting on that data requires a human to first read the report, figure out what to change, make the change, then remember to go back and check that the change worked. When you have hundred of web pages, thousands of keywords, dozens of page tags... how can you humanly manage this?
This same problem was solved by the financial services industry many decades ago. The key came when they implemented credit policy in the form of strict business rules. As economic forces changed from year to year, the credit policy was changed at a macro level, and passed down to local lending officers. This replaced individual ad-hoc decisions made at the branch level. Since that time, business rules and platforms that implement them have created an entire branch of analytics. Business rules are now used in almost all industries.
So how can we apply business rules to the online world? There are SEM bid management platforms that allow you to put in rules to increase or decrease bid prices automatically based on ad position and click through rate. Some display advertisers are now starting to team up with online list providers (in the form of cookie data) to do behavioral targeting. This is a simple form of business rules - if the user is in a certain segment, show them a certain offer. No doubt over time these business rules will get more sophisticted.
One of the maxims of behavioral targeting is that humans are terrible at stereotyping "who buys what product". For example, what is the demographic of people who will pay $15,000 for a bicyle? What is the demographic for mortgage applicants of condo's in London. In the first case, its men between 24 and 30 with an income between $25,000 and $35,000. In the second case, its predominantly professional single women. If you know this, you can implement business rules around these attributes to better target offers. That's behavioral targeting.
How are we to discover this in an automated way? What happens when we have hundreds of attributes that we can use, and dozens of ways of slicing and dicing and combing them? How can we possibly sift through all this?
Back in the 70's, statisticians invented the next type of analytics to address this... predictive analytics. We will talk about that next post!
But acting on that data requires a human to first read the report, figure out what to change, make the change, then remember to go back and check that the change worked. When you have hundred of web pages, thousands of keywords, dozens of page tags... how can you humanly manage this?
This same problem was solved by the financial services industry many decades ago. The key came when they implemented credit policy in the form of strict business rules. As economic forces changed from year to year, the credit policy was changed at a macro level, and passed down to local lending officers. This replaced individual ad-hoc decisions made at the branch level. Since that time, business rules and platforms that implement them have created an entire branch of analytics. Business rules are now used in almost all industries.
So how can we apply business rules to the online world? There are SEM bid management platforms that allow you to put in rules to increase or decrease bid prices automatically based on ad position and click through rate. Some display advertisers are now starting to team up with online list providers (in the form of cookie data) to do behavioral targeting. This is a simple form of business rules - if the user is in a certain segment, show them a certain offer. No doubt over time these business rules will get more sophisticted.
One of the maxims of behavioral targeting is that humans are terrible at stereotyping "who buys what product". For example, what is the demographic of people who will pay $15,000 for a bicyle? What is the demographic for mortgage applicants of condo's in London. In the first case, its men between 24 and 30 with an income between $25,000 and $35,000. In the second case, its predominantly professional single women. If you know this, you can implement business rules around these attributes to better target offers. That's behavioral targeting.
How are we to discover this in an automated way? What happens when we have hundreds of attributes that we can use, and dozens of ways of slicing and dicing and combing them? How can we possibly sift through all this?
Back in the 70's, statisticians invented the next type of analytics to address this... predictive analytics. We will talk about that next post!
Monday, February 22, 2010
Types of Analytics
Like all great things, there are many flavors of analytics. There is a general trend of complexity, where each level of analytics builds on the previous ones. So lets start with the basics.
Reporting analytics is about the simplest form of analytics. People generate reports all the time in all sorts of business. You can tally sales by region, losses by credit vintage, fraud by zipcode, and so on. There is an old adage that you can't fix something if you can't measure it. Reporting analytics is exactly that - measurement.
In the online world, Google analytics is one of the most common reporting analytics systems. This system tracks page traffic, conversions, keyword and ad performance. In the hands of a sophisticated marketer, tools like this are useful in fine tuning marketing campaigns.
But reporting analytics is such entry-level, that many purists don't even like to use the word "analytics" in the same sentence as "reports". I've had people tell me that they have fully implemented analytics, just to find out that all they have done is put in Google analytics. To some extent Google has co-opted "analytics" in the online world. If only they'd called it Google Reporting, alot of us would be more content.
In the offline market world, reporting systems were built starting in the 1960's. Essentially as soon as people could put enough data, like mailing lists, into computers, they created reports as their first step down the analytic journey. Reporting systems are still an important component of any analytics platform today, and there are very sophisticated tools that bleed into data visualization.
However, 99% of reports are never looked at. Relying on a human in the loop to digest the reports and make adjustments to campaigns is well proven to be unreliable, error-prone, and inefficient. This was realized in the offline world 40 years ago, and resulted in the development of the next stage of analytics...
Reporting analytics is about the simplest form of analytics. People generate reports all the time in all sorts of business. You can tally sales by region, losses by credit vintage, fraud by zipcode, and so on. There is an old adage that you can't fix something if you can't measure it. Reporting analytics is exactly that - measurement.
In the online world, Google analytics is one of the most common reporting analytics systems. This system tracks page traffic, conversions, keyword and ad performance. In the hands of a sophisticated marketer, tools like this are useful in fine tuning marketing campaigns.
But reporting analytics is such entry-level, that many purists don't even like to use the word "analytics" in the same sentence as "reports". I've had people tell me that they have fully implemented analytics, just to find out that all they have done is put in Google analytics. To some extent Google has co-opted "analytics" in the online world. If only they'd called it Google Reporting, alot of us would be more content.
In the offline market world, reporting systems were built starting in the 1960's. Essentially as soon as people could put enough data, like mailing lists, into computers, they created reports as their first step down the analytic journey. Reporting systems are still an important component of any analytics platform today, and there are very sophisticated tools that bleed into data visualization.
However, 99% of reports are never looked at. Relying on a human in the loop to digest the reports and make adjustments to campaigns is well proven to be unreliable, error-prone, and inefficient. This was realized in the offline world 40 years ago, and resulted in the development of the next stage of analytics...
Friday, February 19, 2010
Welcome!
Inuvo is breaking the mould on affiliate marketing technology for the web. Our recently launched Inuvo Platform takes affiliate marketing to an entirely new level, with features, controls and reporting that are incredibly compelling.
At Inuvo, we are also changing the performance of affiliate marketing with a big investment in analytics. This blog is all about analytics, what they are, how they work, and how we are using them at Inuvo to make a measurable differences.
So, what is Analytics?
Analytics is the art and science of taking data and using statistics to try to describe a situation or improve a decision outcome. Analytics are all around us: when we apply for a loan of any type, dozens of different analytics are invoked. They check out credit worthiness, they try to figure out how profitable we will be for the bank, they even try to determine if we are truly who we say we are. Analytics are used to target direct mail offers, detect fraudulent behavior, and to optimize the prices of products.
In the offline marketing world, analytics have been used extensively for 40 years. Even though there is a lot of buzz around analytics in the on-line world, we are still very early on its adoption. Like in so many other industries, analytics has become a key part of how a business functions. For example, you could not start a credit card company today without first hiring about a hundred statisticians to figure out all your analytic systems.
I believe that analytics will fundamentally change online marketing by dramatically improving response rates and reducing bad traffic. At Inuvo we are assembling a team of very experienced statisticians to bring this to a reality.
At Inuvo, we are also changing the performance of affiliate marketing with a big investment in analytics. This blog is all about analytics, what they are, how they work, and how we are using them at Inuvo to make a measurable differences.
So, what is Analytics?
Analytics is the art and science of taking data and using statistics to try to describe a situation or improve a decision outcome. Analytics are all around us: when we apply for a loan of any type, dozens of different analytics are invoked. They check out credit worthiness, they try to figure out how profitable we will be for the bank, they even try to determine if we are truly who we say we are. Analytics are used to target direct mail offers, detect fraudulent behavior, and to optimize the prices of products.
In the offline marketing world, analytics have been used extensively for 40 years. Even though there is a lot of buzz around analytics in the on-line world, we are still very early on its adoption. Like in so many other industries, analytics has become a key part of how a business functions. For example, you could not start a credit card company today without first hiring about a hundred statisticians to figure out all your analytic systems.
I believe that analytics will fundamentally change online marketing by dramatically improving response rates and reducing bad traffic. At Inuvo we are assembling a team of very experienced statisticians to bring this to a reality.
Subscribe to:
Posts (Atom)
