Please contact us if you would be interested in joining one of our courses or developping a customised in-house training session on predictive analytics.

Predictive Analytics - course outline

An Introduction to Data Mining and Predictive Analytics is a one day workshop covering the foundations of this innovation marketing analytics discipline. During the course of the day you will gain a thorough familiarisation with some of the key principles and methodologies of data mining and predictive analytics and learn how to apply them to common marketing problems such as:

• How can I predict campaign response?
• How do I segment my website visitors or customers?
• How can I anticipate possible customer defections?

In this one day interactive course we will cover the following topics:

Introduction:

• What is data mining and how is that different to predictive analytics?
• How organisations are currently using data mining and predictive analytics across their businesses and to solve particular marketing problems

Processes and implementation

• How to go about a data mining/predictive analytics project
• An overview of a standard industry process (CRISP-DM)

Methods and applications

• An overview of the main types of data mining and predictive analytics applications:
  • Forecasting
  • Segmentation
  • Classification
• An introduction to main methodologies such as:
  • Time-series forecasting
  • Regression analysis
  • Decision trees (CHAID, CART and so on)
  • Cluster analysis
  • Neural networks
• Case studies and examples of how these techniques are used and deployed in both online and offline marketing is areas such as:
  • Retention modelling
  • Conversion propensity modelling
  • Visitor segmentation

The presentation will be looking at the application of techniques such as segmentation and propensity modelling to better understand website visitor behaviour.

You can download a copy of the presentation here…

Quite often when we use Web analytics systems, we focus on what visitors do during a particular visit. The classic conversion funnel is a good example of this trendMost Web analytic systems look at the conversion funnel in the context of a single visit, that is, they report on how people got to page A, then B, then C, and so on within a single visit. This information is useful because it helps identify potential process areas that need improvement. But if we think about those times when a visitor might make multiple visits to a site before a conversion, the classic conversion funnel might not give you a true perspective on what’s happening.Take the example of buying car insurance online. In the U.K., it’s a very competitive business. Consumers typically shop around for quotes and go for the best value proposition. As a result, it’ s very unlikely people will arrive on a site and buy car insurance on their first visit. Maybe they’ll arrive from a search engine, check out the proposition, and bookmark the site for future reference. Maybe later they’ll come back, get a quote, and leave to compare it to other quotes. Hopefully they’ll return to complete the policy application process, and a sale is made.

A generic conversion funnel analysis will contain an amalgam of all three types of behavior: research, quote, purchase. As a result, you’re not seeing a true reflection of your ability to convert opportunity into value unless you analyze visitor behavior over sequences of visits, rather than just within the single visit.

If you work with Web analytics data, you know it’s hard enough to understand what’s going on when examining a person’s behavior in a single visit. Analyzing behavior over multiple visits adds complexity. Here, data mining and predictive analytical techniques come into play.

If we accept (as in the car insurance example) that conversion is often a multivisit process, we must understand the process’s key drivers over time if we are to influence that visitor’s behavior. We must find out what behaviors over multiple visits are most likely to lead to a successful outcome.

Using a decision-tree technique like CHAID can help you understand how different visitor behaviors over multiple visits may increase or decrease the likelihood of converting a browser into a buyer. CHAID, which is highly visual, shows factors that influence conversion in a tree diagram in the order they influence people.

) can help you understand how different visitor behaviors over multiple visits may increase or decrease the likelihood of converting a browser into a buyer. CHAID, which is highly visual, shows factors that influence conversion in a tree diagram in the order they influence people.) can help you understand how different visitor behaviors over multiple visits may increase or decrease the likelihood of converting a browser into a buyer. CHAID, which is highly visual, shows factors that influence conversion in a tree diagram in the order they influence people.) can help you understand how different visitor behaviors over multiple visits may increase or decrease the likelihood of converting a browser into a buyer. CHAID, which is highly visual, shows factors that influence conversion in a tree diagram in the order they influence people.As with the segmentation approach described in part one, data must be in the right shape before an analysis is started. That requires extracting and summarizing data to key activities and events in each visit of the visitor lifecycle. I often think that data mining and predictive analytics are part art, part science. The art requires possessing the right data in the right format for algorithms to provide meaningful and useful results. In these days of automated analytics, anyone can produce a model. It’s a question of whether the model is good or not.

In working with these techniques, we commonly find there are a small number of highly influential conversion drivers over multiple visits. Naturally those drivers vary from site to site, but the importance of time is usually one thing they share in common. The time between the first and second visit, and the second and third visit, and so on, are quite often a good predictor of the subsequent outcome.

As the need to tune the online marketing processes continues, organizations must add capabilities to their analytics tool kit. Data-mining and predictive analytical techniques are firmly established within other marketing disciplines. Perhaps their time is now coming in the online world.

A simple definition of predictive analytics

A simple definition of predictive analytics is that it’s an activity which allows us to quantify future events or actions. This quantification could be as straightforward as generating a list of customers who are likely, at a point in time, to behave in a certain way, e.g. to churn, to register, to buy or to respond to a particular mail, etc. Typically this list will be accompanied by a score which gives us the probability (or propensity) that an event will occur.
Alternatively the technique may give us a value, or set of values, e.g. a set of sales forecasts for a given product line in the coming week; furthermore the forecasts are often presented with confidence values, e.g. we might predict that we will sell 50 widgets on Monday but we can say, with 95% confidence, that we will sell between 45 and 55.

So what’s new?

One of the main characteristics of PA is that many of the tools, techniques and applications which it comprises are not actually particularly new. Credit scoring is one of the most well-known applications of PA. and credit scoring has been around for over 50 years.

Wikipedia details  many of the other uses and techniques which can be described as “predictive�? – though the current list there is by no means exhaustive. You will probably recognise a lot of them. In fact many of them come from the world of Data Mining and there is some significant overlap between Predictive Analytics and Data Mining, but there are also many differences; more on that later.
So is “predictive analytics�? just a new bottle for a lot of old wine? One of the new things is that there is an abundance of relatively new technology which can:

• Take PA to a wider audience by making the often complex algorithms more usable for less statistical/technical users.
• Provide access to a broader range of techniques through smarter user interfaces which map more closely to the analytical process in such a way that modellers (analysts, statisticians, data miners, demand forecasters, etc.) can more productively access data , test and develop models and ultimately deploy the best ones.
• Allow the results of PA (e.g. customer lists, scores, models, etc.) to be used more easily in decision making processes going forward.

For example, PA is being used to enable customer services representatives in a call centre to prevent customers churning with appropriate offers. It’s also be used in web site recommendation engines which can serve up relevant content based on what is understood of a visitor’s needs, preferences and previous browsing/buying behaviour.

This last point could mean generating models in a format (such as PMML) which can be more easily plugged-in to an operational process or integrating PA into other tools like CRM platforms.

There are also many new techniques and new applications for those techniques. The theories around Naïve Bayes and Robust Regression, for example, may have been around for a while but it is only recently that they have been available in an accessible commercial format. Techniques to automate the search for the best fitting time-series model are also quite contemporary. Applications such as SPAM detection and multivariate testing to optimise on-line marketing campaigns and page content are arguably the latest in a long line. Even if your potential application doesn’t seem to feature in any list of previous ones it could be worth exploring the potential that PA offers. There is a first time for everything!

So what can predictive analytics do for me?

PA is more commonly applied within organisations to address specific issues like preventing customer attrition or to identify segments of customers who will be more responsive to specific campaigns. One of its advantages is that it is usually possible to demonstrate potential ROI through the model, and if the model is a good one, the actual ROI when the model is deployed should not be too far from the prediction.

Increasingly though these techniques are being used more strategically to inform a range of organisational decisions and actions. Effective CRM programmes, for example, often use PA in a number of ways to anticipate customer needs and behaviours across channels and at different points in the customer lifetime.

So more sophisticated applications will not only identify good prospects for acquisition they will help define a series of interactions as the prospect develops into a loyal customer which enhance the customer experience and ultimately drive mutual benefit.

Getting more into predictive analytics

Despite the publicity I feel that applications of predictive analytics, particularly in business, are only scratching the surface today. There are a number of reasons for that; among these is the age old one that there is a gap between the available tools/techniques and the appropriate circumstances for business adoption. There is also a dearth of resource to help bridge that gap!

In future blogs I’ll explore predictive analytics in detail; looking more closely at the tools, techniques, applications and vendors.  I’ll give a view on the process of applying PA – particularly aimed at those who haven’t started yet. I’ll also look beyond the hype into cases where it has been used to achieve significant results.

I’ll even discuss how PA relates to BI, CRM and ERP. Though we shall try and do that in a more understandable way than I did in that last sentence!

But what happens if you’re about to launch a major new TV advertising campaign next week? Won’t that possibly affect the amount of traffic to the site? You would certainly hope that it would but by how much?

This is where explanatory forecasting techniques come into play and have been used for many years in offline marketing analysis to understand the impact of marketing activity on important outcomes such as sales or brand awareness. These techniques build a model where the thing that you are interested in, such as visits, registrations, leads or sales, is explained quantitatively by external factors such as TV advertising, promotions, price and so on. This branch of techniques is often called Econometrics and one of the most popular methods is regression analysis.

A constant theme throughout these articles on more advanced analytical techniques is that the use of these methods is as much an art as it is a science. This is certainly true when it comes to techniques such as regression analysis. Regression analysis is widely accessible through programmes such as Excel and even in PowerPoint you can do some basic linear regression. An old colleague of mine used talk about these types of tools being like “putting guns in the hands of children”. Whilst I thought it was an arrogant way of making the point, the point he was making was quite valid. These algorithms can be quite dangerous if there isn’t the right kind of care and thought about how they are being used and what the results are saying.

The trouble with having these kinds of techniques available is that there is a tendency for them to be used in ways that are inappropriate or when the construction of the model or forecast hasn’t been fully thought through.

I got myself into trouble many years ago when I was cutting my teeth in the world of econometric modeling. We were doing some modeling work on price elasticity. We were trying to forecast the effect on sales of a proposed price increase on a client’s brand. The results of the model suggested a massive detrimental impact, bigger than anything that they had ever seen before. The client didn’t like the results (naturally) and said there must be something wrong with the model. I said that the model was technically correct according to all the diagnostic statistics but that I would go back and look at it again. At this point I noticed an event which I hadn’t picked up the first time round and which I hadn’t taken into account in the model. This event had made it look like that the brand was far more sensitive to price changes than it really was. When we factored this event into the model, the price sensitivity became something more appropriate and we were able to make a much more sensible forecast.

I took two key lessons of that experience:

1. If the model looks wrong, it probably is wrong
2. Modeling is like baking a cake

The first lesson was really the law of common sense. Whilst you are trying to look for insight through the use of these more advanced analytical techniques, the results should still make some sense at the end of the day.

The second lesson was that you need to make sure you have all the right ingredients in order to get the right result. If you are looking to forecast sales for example, you need to ensure that you are capturing as many of the likely impacts on sales as possible in your model. If you don’t then you will either inadvertently have the wrong effects coming through or there will be large errors associated with your forecasts.

Is there any use for these types of techniques in the world of online marketing? Well, much of online marketing analysis is based upon direct response or the tracking of individuals over time. This level of granularity is fantastic and allows us to get deep into the analysis of individual visitor behavior. I wonder though whether sometimes we can’t see the wood for all the trees. How do TV advertising and press ads influence online behavior. What can we infer about the synergistic effects of multi-channel marketing? These are all areas where I believe that modeling techniques will help us better understand the return on investment on marketing spend. As they in the offline world for a number of years.

As businesses build up their data trends then the trends become more interesting and useful. One of the problems with a fast growing environment where all the charts are showing lines that go “up and to the right” is that is difficult to know what the underlying trends are and whether marketing activity is having any effect on this growth.

When it comes to forecasting there are two broad categories of forecasting techniques: Quantitative methods and Qualitative methods. The quantitative methods are based on algorithms of varying complexity and qualitative methods are based upon educated guessing.

Quantitative methods of forecasting come in two main types:

• Time-series methods
• Explanatory methods

Time-series methods make their forecasts based purely on the historical patterns in the data that have been observed in the past. Say for example you wanted to make a forecast of visitors to your site over the next few weeks; time-series methods will only use the historical data of visits to your site in the past to make that forecast.

Explanatory methods use other data as inputs into the forecasting data, so in my example above you might include marketing data as inputs into a model to understand how they impact on visit levels and use that data to also make forecasts for the future. These types of techniques have been used for ages in the offline world to evaluate the effect of marketing activity on brand awareness or sales.

Time-series methods are probably the simplest methods to deploy and can be quite accurate, particularly over the short term. Most quantitative forecasting methods look to explain patterns in historical data as a means of then using those patterns to make forecasts for the future.

Simple time-series methods include moving average models. In this case the forecast is the average of the last ‘x’ number of observations, where ‘x’ is some suitable number. For example if you were forecasting monthly sales data you might use a 12 month moving average where the forecast for the next month is the average over the past year.

The trouble is that simple averaging methods don’t tend to work well when there is either a trend in the data or there are seasonal effects. This tends to be the case in a lot of marketing data! In the case, other techniques such as exponential smoothing techniques may be more appropriate.

With moving averages every point in the data carries equal ‘weight’ in making the forecast. With smoothing methods more importance is placed on the most recent data than on the historical data for forecasting. So, if there is a trend in the data then it is going to use the recent observations to make up the bulk of the forecast and the forecast is more likely to reflect the trend.

Moving averages and simple exponential smoothing techniques are available in Excel and so are easy to execute. That’s one of the great advantages of time-series methods, they are generally simple, cheap to run and relatively easy to interpret.

There are more complex time-series techniques out there as well such as ARIMA or Box-Jenkins models. These are more heavy duty statistical routines that can cope with data that have trends and seasonality in them. You would probably need to invest in a statistical analysis package or a dedicated forecasting package to use these more powerful algorithms. Like any analytical technique though, you shouldn’t use them ‘blind’ and treat the results as gospel. In fact, all forecasts are invariably wrong. It’s just a question of how wrong they are!

So why would you use these types of forecasting techniques?

Forecasting techniques are often used as much for their explanatory power as for their predictive power. Understanding the trends and seasonal behaviour of your business gives you a better understanding of the underlying health of the business.

In consumer goods marketing for example, these types of forecasting techniques are often used to assess the “baseline” performance of the brand. A forecast is made of what the sales would have been in the absence of certain types of promotions or advertising, so that the underlying trends can be assessed.

Explanatory forecasting methods take the process a step further and allow you to relate changes in marketing activity to changes in outputs such as sales, brand awareness, registrations and so on. Here we are looking for “causality” and can use that to feed into forecasts but as a way of evaluating marketing response. We’ll take a look at this in more detail next time.

Till then…

There are a variety of other reporting and analysis tools that we might want to use on the data from our web sites to get a better understanding of online business performance and customer behaviour. It is fair to say that web analytic systems have significantly improved their analytic capabilities over the past few years and will no doubt continue to do so. These days there is a far greater ability in a number of the systems to be able to filter and segment data on the fly to look at the behaviour or characteristics of particular groups.

However, as the needs of the organisation continue to develop so too might the need for different or specialist reporting or analysis tools. Other systems for reporting and analysing web and customer data can be grouped into three broad categories:

• Business Intelligence (BI) or OLAP tools
• Visualisation tools
• Statistical analysis and data mining tools

BI or OLAP tools are often found in the corporate reporting environment and this class of tools includes systems such as Business Objects, Microstrategy and Cognos. Databases such as Oracle and SQL Server also either come with BI functionality or it can be bolted on. Underpinning many of these tools is the concept of a data cube that allows the analysts to drill through the data in a hierarchical manner. In a commerce environment I might start looking at say at total sales for a year and then drill down into product categories, then into sub-categories and then down to the product level.

Some web analytics systems do have the ability to drill through data in this way but a feature of the family of BI tools is the ability to handle multiple hierarchies across multiple dimensions. So, in addition to being able to drill through on the product dimension, you can also drill through the data say in terms of geography and also time. BI tools could also be used to report on web data in the context of other channels, for example comparing the profile of leads or enquiries generated online against those generated in the call centre.

As the saying goes, a picture tells a thousand words and visualisation tools can be a valuable weapon in your analytical arsenal. Again some web analytical tools such as Visual Sciences and Site Intelligence have some powerful visualisation capabilities. Whilst many web analytics systems have improved the visual reporting of web data through developments of click overlays for example, for the analyst a visualisation tool might add another dimension.

Visualisation tools can range from add-ins or add-ons for Excel through to complex applications that are commonly integrated in with data mining tools. At the desktop level, Excel add-ons such as MM4XL extend the scope of the charting abilities of Excel and allow the analyst to present data in different ways. More sophisticated tools can produce three dimensional rotating images that allow the analyst to explore and look for patterns in the data. The human brain is still one of the most powerful tools available for spotting patterns and trends in data when presented in the right way!

The final set of tools that might be useful for analysing web and customer data are statistical analysis and data mining tools. What’s the difference between statistical analysis and data mining? The way that I tend to view it is that statistical analysis is predominantly about exploration and data mining is about discovery. With statistical analysis you are often looking to test an assumption or a hypothesis. For example, you may be looking to prove that one group of customers rate your product or service more highly than others. With data mining, you are looking for patterns or relationships in the data that you may not know about.

Statistical analysis and data mining covers a wide variety of approaches, methodologies and techniques that might be useful for the web analyst. The can be broadly be classified as follows:

• Statistical analysis
• Classification techniques
• Clustering and segmentation methodologies
• Forecasting
• Text analysis

Increasingly many of these techniques are being used for making predictions and so the phrase “predictive analytics” is a term that is often used as well to describe these various methodologies.

Some of this stuff may seem like a long way from the current day to day analysis of conversion funnels and the like. But as the market continues to mature and growth comes from optimisation and improvements in marketing efficiencies, some of these techniques will have a place on the analyst’s workbench. Over the next couple of weeks, I will take a look at some these techniques in more detail and how they be used in the context of analysing online visitor and customer behaviour.