Archive for the ‘Predictive Analytics’ Category

Applying Predictive Analytics

Posted by: Grant Stanley on April 13th, 2012 Leave a Comment

Predictive Analytics allows people to make better decisions about how to spend their limited time, energy and money. The potential impact of predictive analytics on business will be similar to the personal computer, relational database and Internet. The power of predictive analytics is that it is a scientific business process improvement method that can be used to model complicated and hard to measure actives, such as why people buy something or which employees are likely to leave. Many business executives understand this potential and are excited about applying predictive analytics to their businesses.

CAN has 4 years of experience helping 200+ companies realize the benefits of predictive analytics. We have developed a 6 stage process for applying predictive analytics to our clients’ businesses that maximizes our clients return on investment, increases their chances for success, and makes sure that the results of our research are applied.

 

Six Steps to Applying Predictive Analytics

The first stage is to define the company’s mission, vision and values. We want to know why the company was started, and why it exists. We want to know what they want to accomplish in the future. Most importantly we want to know how they do business; what values they have that are unique and perminant even when the strategy changes. This understanding set the priorities and filters that guide future discussions.

The second stage is to define the company’s goals. Unlike mission, vision and values, goals have clear beginnings and ends and typically can be accomplish in less than a year. Companies typically have one to three goals. Goals should be in alignment with the company’s vision for the future, and should be accomplished in a way that adheres to the company’s values.

The third stage is to define the business question to be answered. The business question is about business process improvement, and should not involve technology or research questions. When answered a business questions should have a noticeable impact on at least one of the three parts of a business; sales, operations and administrative support.

The Business Question

The fourth stage is to determine what resources are available. Resources include political approval, availability of necessary data, and determining research methodology. It is important to note that we only determine research methodology once we have defined the business question. If we don’t have the necessary resources to answer the business question, we go back to stage 3 and try to refine the business question to fit the available resources.

It is also during this stage that we determine which of CAN’s resources are best for the client. There are two basic options; custom solution or CAN’s products. If possible we try to answer the business question using one of CAN’s 5 products. This allows us to minimize cost while increasing chances for success. Our 5 products are designed to answer 5 key business questions that the majority of business owners have:
1. Tracker: Who is most likely to purchase my product next?
2. Capture: Where and when should I spend my marketing budget?
3. Pulse: How do I attract and retain my best customers?
4. Beacon: Which employees are most likely to leave and why?
5. Terrain: What sales are likely to be next quarter?

If a business question can’t be answered using one of CAN’s products, we offer custom solutions. Many of CAN’s clients leverage our custom solutions to develop a competitive advantage in predictive analytics. When developing custom solutions it is essential that we become apart of our clients team and fully understand their business, goals and resources. Before committing to any custom projects, CAN requires that we build a proof of concept. The purpose of the proof of concept is to make sure that we fully understand what we need to build and that we have all the necessary resources.

The fifth stage is to determine how the models will be implemented. While our research is complex, we make sure that our work is easy to understand and use, because that is how it gets implemented. We use 4 methods to implement our research, and often combine multiple methods depending on what the client’s goals are.

Reporting Predictive Analytics

1. Formal Reports help our clients understand the nuances and details of our research. Formal reports are most useful when the results of our research will influence a company’s strategy, will be used by a small and specialized audience, and frequent updates are not required.

2. Marketing Summaries provide our clients with colorful and easy to understand summaries of our research. Many of our clients use these summaries as marketing pieces to communicate quickly with large and unspecialized audiences about research that impact future strategy. Marketing pieces go beyond executive summaries because they can be used be used to inform executives, employees, customers and the community.

3. Dashboards help our clients quickly get the up to date information that need to run their operations. While formal reports and marketing summaries often include data visualizations, dashboards are unique because they can be quickly updated and display key information on a single screen that can be monitored at a glance. Dashboards are typically used by a small and specialized audience that is trained to understand and use the information on the dashboard. Dashboards can also be very useful for sensitive information, because administrators can control access by user on a need to know basis.

4. Workflow Integration provides our clients with the ability to use our research to impact the activities and operations of large number of people through the systems they are currently using. Workflow integration is useful because users don’t have to learn or get in the habit of using a new system. The predictive models and coefficients that CAN develops act as a filter to the current database and either users are presented with familiar data fields or if need new fields.

The method that we choose depends on who the audience will be and how they will use the results. The fewer the people that need access to our research the more important security and control becomes. If the use of our research is for strategy development then we typically publish a formal report. However if the use of our research is to optimize operations then we publish it as a dashboard, marketing piece, or integrate it into the software you are currently using.

The sixth step is to evaluate the model. As part of developing models we run tests to make sure that they are statistical robust. However, it is important to further evaluate a model before and after we implement it. The first evaluation criteria is does the model answer the intended business question. The second criteria is does the model produce results that reflect reality. While the model might be statistical robust, it is useless if it produces misleading results that the experts in your business know aren’t true. The third evaluation criteria is once implemented does the model produce the expected results.

Predictive analytics is a very new field. While the technology is exciting, it is predictive analytics ability to answer hard to answer or previously impossible to answer business questions that is most exciting. What separates CAN from our competition is our focus on making sure that we answer our clients’ business questions, instead of being enamored by the technology. Our hope is that we can help our clients apply predictive analytics to their businesses, and that our 6 stage process helps them maximize their return on investment, increase changes for success, and makes sure that the results of our research are applied.

Occam’s Razor and Model Complexity

Posted by: Matt Dickinson on March 6th, 2012 Leave a Comment

Occam's Razor and Model Complexity

When using predictive analytics to develop a model it is important to understand the principles of model complexity.  Occam’s Razor is a concept that is frequently stated, but not always fully understood.  The basic idea is that “All else being equal, simpler models should be favored over more complex ones.”  It is concept we both embrace and approach with caution so that it is not misused.

First, let’s flesh out the concept of Occam’s Razor beyond the simple aphorism given above as it can apply to predictive analytics.

Suppose I flip a coin ten times, and I get a run that goes “HHTTTHHTTT”.  After observing the coin flips I assess that there are two possible models for the behavior of the coin:

(A) The coin is fair and has a 50/50 chance of getting either heads or tails on each flip.  The observed run was just one of 1024 possible results of the ten coin flips.

(B) The coin flips are deterministic and will land in a repeating pattern of “HHTTT” which perfectly fits with the results of our sample of coin flips.

Without further experimentation I have no certain way of knowing which model is actually true.  If I were to flip the coin five more times, if I got anything other than “HHTTT” all confidence in (B) would be gone, the same cannot be said for (A).  This is because (B) is a much more complex model then (A).  It other words, it would take much more evidence to be confident in (B) over (A).

Keeping this concept in mind is important when developing predictive models.  With the huge volume of information and the massive data sets that CAN utilizes it can be tempting to include as many parameters as possible.  Not paying attention to the model complexity, and the evidence required to support the claims of observed relationships can lead to false assumptions, and predictions that are outside of desired bounds.

Another problem arrises when Occam’s razor gets misapplied.  A common mistake is that people misinterpret this to mean that models should be a simple as possible, when the thought process should be to keep models as simple as they need to be to explain the what is observed.

One of the more famous examples of this is in the history of the understanding of the motion of the planets.  In an attempt to explain the retrograde motion of the planets Ptolemy devised a complicated geocentric model where each of the known planets, including the sun, orbited the earth while also moving in their own smaller circular paths.

Alternatively, Copernicus’ heliocentric model, placing the sun at the center of the solar system, was a much simpler model and it fit the observed motion of the planets, as the planets are orbiting the sun at different rates.

As revolutionary as the Copernican idea was, it does fit Ockham’s premise.   At the time, Ptolemy’s model was a better fit for what had been observed. But, like the more complicated coin flip model the geocentric model fit what had previously been observed yet utterly failed when later observations did not fit.

Ptolemy's Model

Copernicus' Model

But, simpler models alone do not mean that that the truth is revealed.  Copernicus’ model was not perfect.  Later, when Johannes Kepler discovered that the planets move in elliptical orbits, not the circular ones as supposed by Copernicus, it made the model much more accurate at predicting the movement of the planets.  But, ellipses are much more complex then circles, what does this mean for Ockham?

Again, utilizing Occam’s razor is not a search for the most simple model, but the most simple required.  While Kepler added complexity to the model, it was complexity that was supported by decades of information gathering on the inconsistencies in the Copernican model.  More complicated models require more evidence to defend, which is not the same as saying they are indefensible.

Let’s go back to the coin flip example and change the scenario slightly.  Let’s say we were attending a magic show, and the magician claimed that he could flip a coin and make it follow the pattern “HHTTT” (i.e.  model (B) ).  Does this change our perception of the complexity of the models?  Of course it does.  We expect there to be a trick that will cause the magician’s claim to be true.

Like Kepler’s model, there is more complexity in the model that predicts a distinct pattern of coin flips, but the presence of new data (the magician flipping the coins, or the precession of planetary orbits for Kepler) allows us to have more confidence in the more complex model.

At Contemporary Analysis we solve hard problems, and hard problems often necessitate complex models, and complex models are not in themselves bad.  Throwing out informative parameters for just the sake of simplicity would strip away the predictive value of the models. Its about ensuring that the factors of the complex models are both valuable and grounded in sound theory.  This is why we don’t just jump into building models as soon as we have data.  We spend a lot of time working with customers to understand the nature of the environment we are modeling.  Simple or complex, if our theory is misinformed there is no way our models can accurately reflect the real world.

References:  

http://philosophy.wisc.edu/forster/520/Chapter%203.pdf 

http://www.stat.duke.edu/~berger/papers/ockham.html

http://andrewgelman.com/2004/12/against_parsimo/

Using Mean Absolute Error for Forecast Accuracy

Posted by: Branden Collingsworth on January 23rd, 2012 1 Comment

Using Mean Absolute Error for Forecast Accuracy

CAN uses mean absolute error to help our clients that are interested in determining the accuracy of industry forecasts.  They want to know if they can trust these industry forecasts, and get recommendations on how to apply them to improve their strategic planning process.  This posts is about how CAN accesses the accuracy of industry forecasts, when we don’t have access to the original model used to produce the forecast.

First, without access to the original model, the only way we can evaluate an industry forecast’s accuracy is by comparing the forecast to the actual economic activity.  This is a backwards looking forecast, and unfortunately does not provide insight into the accuracy of the forecast in the future, which there is no way to test.  Thus it is important to understand that we have to assume that a forecast will be as accurate as it has been in the past, and that future accuracy of a forecast can be guaranteed.

As consumers of industry forecasts, we can test their accuracy over time by comparing the forecasted value to the actual value by calculating three different measures.  The simplest measure of forecast accuracy is called Mean Absolute Error (MAE).  MAE is simply, as the name suggests, the mean of the absolute errors.  The absolute error is the absolute value of the difference between the forecasted value and the actual value.  MAE tells us how big of an error we can expect from the forecast on average.

One problem with the MAE is that the relative size of the error is not always obvious.  Sometimes it is hard to tell a big error from a small error.  To deal with this problem, we can find the mean absolute error in percentage terms. Mean Absolute Percentage Error (MAPE) allows us to compare forecasts of different series in different scales.  For example, we could compare the accuracy of a forecast of the DJIA with a forecast of the S&P 500, even though these indexes are at different levels.

Since both of these methods are based on the mean error, they may understate the impact of big, but infrequent, errors.  If we focus too much on the mean, we will be caught off guard by the infrequent big error.  To adjust for large rare errors, we calculate the Root Mean Square Error (RMSE).  By squaring the errors before we calculate their mean and then taking the square root of the mean, we arrive at a measure of the size of the error that gives more weight to the large but infrequent errors than the mean.  We can also compare RMSE and MAE to determine whether the forecast contains large but infrequent errors.  The larger the difference between RMSE and MAE the more inconsistent the error size.  The following is an example from a CAN report,

While these methods have their limitations, they are simple tools for evaluating forecast accuracy that can be used without knowing anything about the forecast except the past values of a forecast.

Finally, even if you know the accuracy of the forecast you should be mindful of the assumption we discussed at the beginning of the post: just because a forecast has been accurate in the past does not mean it will be accurate in the future.  Professional forecasters update their methods to try to correct for past errors.  However, these corrections may make the forecast less accurate.  Also, there is always the possibility of an event occurring that the model producing the forecast cannot anticipate, a black swan event.  When this happens, you don’t know how big the error will be.  Errors associated with these events are not typical errors, which is what RMSE, MAPE, and MAE try to measure.  So, while forecast accuracy can tell us a lot about the past, remember these limitations when using forecasts to predict the future.

Why Customer Segmentation Will Improve Your Marketing

Posted by: Nate Watson on January 3rd, 2012 Leave a Comment

Why Customer Segmentation Will Improve Your Marketing

I have been married for almost 10 years.  I have gotten good at buying gifts, even clothes.  I can match blouses and jewelry, dresses and belts, shoes and jeans.  My secret?  I look at the mannequins.  Mannequins were designed to attract your attention in a store window and to lure you into the store.  Then inside, they are designed to show you some of the combinations you could make with their clothes.  Essentially they are designed to get you to by more than one thing.  This is perfect for guys.  All we have to do is point to the mannequin and ask a store sales person where we can find those certain pieces of clothing.  We can buy the mannequin  lock stock and barrel and end up with a complete outfit for our brides.

This is actually what I do when I do marketing for companies.  Marketing comes in two forms.  Marketing to your current clients and marking to people who aren’t your current clients.  I view each as a completely different problems while most marketing companies do not.  They have only have one mannequin with the same clothes for every store.  Children’s clothes, women’s clothes, men’s clothes, all the same mannequin.

Unfortunately most of the marketing plans I see are all the same.  The commercials are different but the plan is the same.  They do direct mail, email, print campaign, radio, and if you have the budget TV ads.  Their view is that all people are potential clients which as you who follow the blog know, is not true.  The problem is that they have no idea who their target audience is, and how to market to them.  They found a campaign that reaches a lot of people and they sell you that one.  It may have different clothes, but it is the same mannequin.

This is critical to understand.  Going are the days when people walk down a street with stores on it, see a mannequin, and make the decision to go inside.  Now, people browse their favorite stores online and are more loyal to store brands like Gap, Old Navy, JC Penny’s, Younkers, and Von Mauer.  We have to get a new mannequin.

Mannequins, models, are going through a huge makeover right now.  In fact, if you do a job search for marketing entry-level positions, they are looking for things like statistics, modeling, and analytics in your background.  What they are struggling to come to understand is the idea of marketing to one person at a time.  Marketing now needs individualized messaging.

I don’t have to go to far back to find when this was still science fiction.  Minority Report (2002) shows a seen where character John Anderton is walking through the mall and cameras recognize, data base search, and present relevant ads, based on his buying history.  To see a version of this now, just go to iTunes.  Any song I buy will create 5 suggestions of other songs I might like, based on my buying history, songs with similar tempos, themes, or by the same artist.  With the combination of a predictive analytics company like ours, we can do the same thing for your product.

First you have to understand something about how we view marketing.  This is the key philosophy that makes us different. Marketing is using a different medium to get in front of your target audience for the sole purpose of selling to them.  Selling to them.  The sole purpose.   EVERYTHING else is branding.  Branding is fine. We do a lot of branding.  We just don’t call it marketing.  That key aspect alone will forever change the way you do marketing.  When you use it as a sales tool, you will no longer accept marketing with no measurement of who looks at it, how it is crafted, and where it is put.  It will focus your marketing on only the people who have an actual chance to buy from you in the marketing cycle.  This does not include:  people who might buy, people you think need to be introduced to your product, or, someone who might have a need, someday.  What marketing with the intent to sell does is only spend your time and money on the people who are ready to buy now.

How do we do this?  We use math and econometrics to understand the buying process.  What causes people to think of buying a product like yours?  What series of events leads to needing a product like yours?  Who are the people in a company that make the decision to buy a product like yours?  These are important things to understand in the process.  Don’t market to someone who doesn’t need what you are selling, isn’t high enough up in a company to make any kind of decision, or hasn’t experienced any kind of problem that your product would fix.  It wastes time.  Why would you ever market a phone system to a sales person.  They can’t buy it.  Why would you ever market a copy machine to a company of 4 people.  They can’t afford it.

I have heard the argument that you need to be in front of those people now so they think of you when he problem arises.  Valid argument.  However, because of the new view of marketing I just gave you, that states that marketing is used as a sales tool to find people who are ready to buy now, you can see that this is branding.  Branding is necessary, we do branding; however, if you have a company like us who markets you correctly, i.e. to the right people, at the right company, at the right time, you don’t even need to do that.  Example.  Quick, name someone who makes shingles.  Not someone who installs them, someone who makes them.  Why don’t you know?  Shingles protect everything in your home and are the first thing damaged in a storm.   If you own a house, shouldn’t you know who the best, worst, and middle of the road companies are in the shingle business?  The reason you don’t need to know, is that you don’t need shingles.  When the time comes and you need shingles, you will do your research and find a company that installs the type of shingle you want on your home.  It’s the same for business.  People don’t really need to know about your product until they start doing research about your product.  Key point:  Up until now, you had to guess when companies were going to need you and you had to brand so that people remembered you when that time came.  Now, with analytics, we can predict when to contact companies because we can predict when they should be beginning research on products similar to yours.

Think about this.  How can a marketing company know where to put your ads either on tv, radio, direct mail, email campaigns,and social media without knowing what causes someone to buy your product.  Its time to start marketing differently.  Instead of putting mannequins in the window, lets know the person that is walking down the street.  Lets advertise to them when they need us, want us, and can afford us, and save our money on tire kickers, time wasters, and spectators.  Its time to spend your hard-earned marketing budget on the people who we need to talk too.  Its time to work smart.

Presenting Predictive Analytics

Posted by: Grant Stanley on September 19th, 2011 Leave a Comment

The nature of forecasting the future makes presenting predictive analytics unique and challenging.  There is no flashy server or dashboard that will make presenting analytics any easier.  There is only a model that tells a story about the future of users’ business, customers, non-customers and competitors.  While models are very valuable they are not your typical business intelligence artifacts.  To produce a meaningful return on investment you need to translate the details of the story into results that can be applied to a specific business question.

While you can not replace sound scientific and statistical methodology, CAN has found that users don’t care about a model’s Durbin-Watson, standard error, or R², or they are not familiar enough to properly understand the statistical nuances.  The key to proving that a model works and getting political support required for implementation is to ask the experts if the story the model tells reflects reality.  It is also valuable to prove that a model works by letting the prove itself over time.

It is important to note that predictive models typically do not provide solutions to business questions, but instead often offer incomplete answers and important insights.  When presenting predictive analytics your audiences epectations should be set on becoming less wrong, instead of finding the perfect solution.  CAN finds that users appreciate our philosophy of Less Wrong.  While it seems counter intuitive, our lack of hubris builds confidence in our models and sets realistic expectations.  The basic principle behind, Less Wrong is that in business winners are not right, they are simply less wrong.  There are no perfect answers in complex sciences, such as data science and predictive analytics, only less wrong answers.  The goal is to reduce the uncertainty of making the wrong decisions, not thinking uncertainty can be eliminated.

In conclusion when presenting predictive analytics don’t be afraid to kill your darlings.  If you can not justify an element of your presentation get rid of it.  This will help you focus your presentation, your audience will listen and the results of your hard work building predictive model will get implemented.

Applications of Predictive Analytics

Posted by: Grant Stanley on September 16th, 2011 1 Comment

The following are some examples of how predictive analytics can be applied in financial services, retail and manufacturing.  This list is not comprehensive, but it provides some interesting applications.

In the financial services the cost of making the right decisions provides marginal benefits, while making the wrong decisions can have significant costs.  Most applications of predictive analytics in the financial services industry help companies avoid making the wrong decisions. For example, credit card companies are able to determine who is most likely to default on their credit cards in the next 6 months by applying predictive analytics to customers purchases and demographics.

In retail understanding customers is essential to success.  Retailers have provided customized shopping experiences by using predictive analytics to understand the drivers of profitability, loyalty and activity for each customer segment and develop specific campaigns for each segment.  This has allowed retailers to wow customers with personalized services while scaling and keep prices competitive.  Predictive analytics have helped both offline and online retailers determine which products to carry, optimize marketing plans, and develop promotional and loyalty programs.  Imagine only offering effective loyalty promotions to profitable customers at risk of leaving, while avoiding offering discounts to unprofitable or already loyal customers.  Another example is knowing what a customer is most likely to purchase next, so that your staff or website can make informed recommendations.

Manufacturing is about knowing what, how, and how much to produce.  Predictive analytics have helped manufactures manage their supply chain and production schedules by accurately forecasting demand, and have helped manufactures produce goods in the most effective way possible by predicting failure of equipment, monitoring workers, and identifying ways to eliminate inefficiencies.  For example, CAN has helped companies with tens of thousands of sales each month forecast sales within a 10 to 50 units, so that they can optimize production schedules and supply chains.  Also, imagine being able to understand why different managers have different levels of employee turnover, employee injuries, and equipment failure.

Beyond specific applications, predictive analytics has the unique ability to help companies become less wrong, scale decisions and systematized learning.

Less Wrong: The basic idea of Less Wrong is that in business, and almost anything in life, you can never be perfectly right, but you can be less wrong and by striving to continually become less wrong you get closer and closer to being right.  By using predictive analytics you will not get the perfect answer, but you can determine what is happening, what most likely happen and most is most likely the right thing to do.  For example it might be really nice to know exactly what commodities prices will be in a month, unfortunately this is not possible.  However, using predictive analytics you might be able to predictive with 70% accuracy which direction a commodity price will trend and if this is better than before predictive analytics then it most likely will be worth the investment.  Another example would be picking which commodity would most likely be the best investment in the next 6 months.  The reason that predictive analytics can’t produce exact results is because it is not a simple science, for more read my post on Simple vs. Complex Science.

Scale Decisions: Predictive analytics has the unique potential to allow executives to scale their decision making as organizations and decisions become increasingly more complex with ever thinner margins for error.  Predictive analytics can be used to create a model of the business based on the organization’s data and executives’ theories.  For example an experienced sales manager will be able to determine which sales leads will most likely respond, purchase and be profitable customers, however he or she does not have time to review every lead for a 200 person sales team.  Also, while a sales manager knows what make a lead worth pursuing, he or she most likely finds it difficult to communicate the rules and criteria to their team.  Using predictive analytics a sales manager can develop a model to score incoming sales leads.  This model can be coded directly into the company’s customer relationship management (CRM) system so that leads are scored as soon as they are entered into the system.

Systematized Learning: In the future, profits will be directly related to your company’s rate of metabolizing new knowledge, as opposed to renting out exisiting knowledge.  Predictive analytics can increase your company’s ability to metabolize new knowledge by continually studying the data produced by your company, customers, non-customers and competitors to find important patterns that will impact your business.  For example, predictive analytics can help executives identify when customers stop responding to a certain campaign and why.

While every company can benefit from becoming analytical, like any other tool, predictive analytics can not fix anything.  However, it is most certainly the next step in the evolution of business intelligence.  If properly applied, predictive analytics has the potential to help businesses work smart. Read a related post on when to and not to apply predictive analytics.

When to Apply Predictive Analytics

Posted by: Grant Stanley on September 14th, 2011 1 Comment

At CAN we love predictive analytics and if we are not careful it is easy to start reducing everything into predictive models.  I have even caught Tadd standing at the window collecting primary data on the smoking habits of the people in our building.  To make sure that CAN and predictive analytics experience continued success, we have developed a guide for when and where to apply predictive analytics.  First, predictive analytics should not be applied if:

The cost of being wrong is low. You should not apply predictive analytics if reducing uncertainty does not provide enough value.  Predictive models should only be applied in situations with a high cost and/or probability of being wrong and where predictive analytics can provide information to reduce uncertainty. To determine if predictive analytics is worth applying to a decision you need to calculate the expected value of information.  In the book How to Measure Anything, Hubbard provides the following formula, expected value of information is equal to the difference between the expected opportunity loss before and after information.  The expected opportunity loss is equal to the chance of being wrong multiplied by the cost of being wrong.

The relationships are obvious. A predictive model is basically a story about why something happens and what will most likely happen in the future.  With this in mind, you do not need to develop models if people are able to accurately describe the relationships between variables well enough that they can tell stories about why certain things are or are not happening.

The model can not be Reproduced. Even if you develop a model, but it can’t be reliably reproduced then you should admit that predictive analytics can not be applied.

Predictive models find and tell stories that are difficult for people to discover because there are either too many variables or the events being studied are too rare.  The value in knowing the story of why something happens is that you can reduce uncertainty, make better decisions and work smart.  The following are situations when predictive analytics should be applied.

Too Much Information. Over the last 30 years there has been an explosion of data, and business intelligence has been focused on collecting and presenting that data.  Predictive analytics is one of the BI tools that is capable of statistically filtering out what data is most important.

Customer management is an example of when predictive analytics can be used to find valuable patterns in an overwhelming amount of data.  Predictive analytics can be used to determine how to group customers, and develop campaigns to improve specific behavior for each client segment. For example, CAN sorted through over 500 different variables and millions of observations to determine the 5 variables that have a significant impact on responses to specific market campaigns.

Rare but Important Events.  When an event is rare and results in either major gains or losses, it can be very beneficial to gain a better understanding of why the event happens.  Predictive analytics can help you develop plans to encourage or discourage specific rare events.

Predicting the failure of essential business processes is an example of when predictive analytics can be used to find valuable patterns in rare but important events.  Using predictive analytics CAN’s clients are able replace equipment and parts before they breakdown so that essential business operations continue to move forward.  Imagine being able to know and replace parts at risk of failing before you spend the time and money to install a piece of heavy construction, mining or logging equipment on site.

In conclusion, the next step in the evolution of business intelligence is to understand what is likely to happen.  Predictive analytics allows executives to learn from the cumulative knowledge of their organization.  This systematized learning has the potential to help businesses and executives to make decisions that are less wrong, so that they can work smart.  However, it is important that predictive analytics is applied in the right applications, so that is produces the most value to the end users.

How to Gain Political Support for Predictive Analytics

Posted by: Grant Stanley on September 8th, 2011 Leave a Comment

The decisions you make in business may never be perfectly right, but, you can strive to become less wrong.  Predictive analytics provides decision makers with a system to continually improve decision-making, while eliminating some of the inefficiencies of non-analytical trial and error.  However, the ability of predictive analytics to systemize an organization’s cumulative knowledge can be threatening to experts who value their accumulated knowledge over continual learning.  When beginning a predictive analytics initiative, the most vital key for gaining political support is to maintain focus on the business problem, and never the technology.  By focusing on the specific problem, you will deploy predictive analytics only if it is the right tool for the job.  This ensures that the initiative has a greater likelihood of success, has the support of key internal stakeholders, and because predictive capabilities are leveled at a specific target the initiative gains executive buy-in.  ”A business problem exists, and this is how we are going to solve it.”

While predictive analytics may be ‘the best man for the job’, expect there to be resistance to implementation.  Often, this comes from individuals who rely on their accumulated knowledge as a competitive advantage against their team-mates.  In light of this situation, quite possibly the most effective method of gaining support, is to focus predictive analytics on a specific, defined business problem.  Your initiative must be dialed in on solving vital, business critical issues.  This way, dissenters to implementation will be seen in the light of hindering the future success of the organization.

Consider the following case study.  One of our clients had a division that is responsible for sourcing materials for production.  They had a group of commodity traders that were responsible for sourcing materials at the best price possible.  While their expert traders had a great track record of forecasting the market, their best traders were nearing retirement.  Also, while the company had made significant investments into business intelligence, the amount of data required to make an informed trade had been growing exponentially for the last ten years.  The future of the organization required developing a system that made learning from the cumulative knowledge of the organization easier.  However, trying to come in with predictive analytics was politically challenging, as it could be threat to both business intelligence and the company’s best commodity traders.  To overcome that perceived threat, we had to focus on the business problem and make a case that eliminating the problem was essential to the continued success of the organization.

Our client had to systematize the knowledge of those traders nearing retirement, and also develop a solution to find valuable patterns in the increasing flow of data.  The solution that we developed was simple.  We built a model that forecasted the direction of commodity prices. Before our model, traders and business intelligence had been spending a significant amount of time determining the direction of the market.  Now, by leveraging predictive analytics, traders and business intelligence have moved up the value chain.  Instead of spending the majority of their time trying to determine the direction of the markets, they spend the majority of their time quantifying the direction of the change.  This resulted in the improved performance and value of both business intelligence and the commodity traders.

In conclusion, the key to gaining political support is to define the business problem in the context of its importance to the continued success of the organization.  If solving the business problem is not essential to the success of the organization, it may not be worth addressing.  If it truly is important, and predictive analytics is the best candidate for the job; internal opposition will be seen as coming from selfish protectionists who threaten the continued success of the organization.

Predictive Analytics, the Evolution of Business Intelligence

Posted by: Grant Stanley on September 6th, 2011 5 Comments

Evolution of Business Intelligence

Predictive analytics is the next step in the evolution of business intelligence.  Most companies, even local small business, have already implemented business intelligence systems that help them understand what has happened, why it happened and what is currently happening.  For example, most small businesses have implemented Quickbooks and Google Analytics that allow them to report, analyze and display data about their finances, operations and marketing.

The Evolution of Business Intelligence

The next step in the evolution of business intelligence is to understand what is likely to happen.  Predictive analytics allows executives to learn from the cumulative knowledge of their organization.  This systematized learning has the potential to help businesses and executives to make decisions that are less wrong, so that they can work smart.  Imagine understanding your customers well enough that you only send discounts to profitable customers who are risk of leaving, and not to customers that are unprofitable or not at risk of leaving.

Predictive analytics is a technology that has proven itself in academics, medicine, government and business.  By combining mathematics and statistical methods to discover patterns in data, predictive analytics differentiates itself from other business intelligence tools by being able to ‘learn’ from experience.  It is the only business intelligence tool that doesn’t rely on the users ability to find patterns in the data and deduce meaningful insights.

When properly implemented, predictive analytics enables your business intelligence to move beyond information to insights about why something happened, what you should do next and what the future might look like.  The following matrix highlights the information that traditional business intelligence provides, and the insights that predictive analytics provides.

The Business Intelligence Matrix

Since predictive analytics produce insights that are customized specifically to your company, customers, employees, and competitors; it has the potential to provide your company with a  unique and non-transferable competitive advantage.  The return on investment realized from predictive analytics depends on the value of the business question that answered.  Most research in the ROI of predictive analytics conclude that ROI is maximized when companies use predictive analytics to improve the effectiveness of their business processes, such as sales, marketing, customer retention, management and strategic planning.  This is why CAN has focused on developing products that use predictive analytics to help people sell, market, retain customers, manage and plan smarter.  Our mission is to reduce the cost and complexity of predictive analytics, so that businesses of all sizes can work smart.

Simple Science vs. Complex Science

Posted by: Grant Stanley on July 27th, 2011 3 Comments

Science is the systematic study of a phenomenon that includes observation and experimentation to explain and understand why things happen.  We can use science to explain almost everything in our universe from the effects of gravity to the impact on sales of your latest marketing campaign.  However, it is important to understand that there are two types of sciences, simple and complex, and that the answers they produce are different.

In simple sciences, such as physics and chemistry, the best possible answer is exact and often is not subject to changes.  For example, gravitational acceleration of objects in a vacuum is 32.2 ft/s2 no matter the size, density, or shape of the object.  While in the complex sciences, such as economics, data science and biology, the best possible answer can never be exact and is almost always relative to time and situation.  For example, under the current situation if inflation increases and interest rates go up in the next 3 month the S&P 500 might increase in value 400 points with a 25% margin of error.  The reason for these more complex answers is because complex sciences can not study things in a vacuum.

People often struggle with the scope of the problems that economists and data scientists answer, because they want exact answers with universal truths that hold across different situations and time periods.  However even though an exact answer is not possible it does not mean that the answers are invaluable or non-scientific, but rather are more complex.  When thinking about problems, it is important to keep in mind the scope of the problem being solved.  Specifically; can the problem be solved using simple sciences such as physics and chemistry or complex sciences such as economics and biology.  The difference between simple and complex science isn’t the level of importance or the difficulty of the problems, but instead the answers.  (Why Predictive Analytics is Important)