SOLUTION INSIGHTS Solutions by Industry
Across multiple industries the major categories of customer relationship management, risk management, and planning and optimization are the major focus of predictive modeling.  These are by no means the only applications but they are overwhelmingly the most common.  Our Solutions Insights page gives a brief description of each major use to help you visualize where these 21st century management tools can be of most value in your organization.  The companion page, Solution Insights by Industry shows the range of problems addressed by type of business. Financial Services
      Banking & Mortgage Lending
      Credit Cards
      Securities & Brokerage

Telecom & Utilities
Manufacturing & General Business
Transportation & Logistics

Why they come - Precision Marketing

Predictive Modeling is the ‘smart bomb’ of 21st century marketing professionals; able to put your message or program directly in front of the prospects most likely to accept.  Whether you target your marketing by direct mail, email, web, print, phone, or electronic media, predictive modeling can create precision targets for your message.  Taking an example from the oldest and most time-proven application, direct mail, this approach creates a model from what is known about your prospects (such as age, sex, income, geo coding, and any other known data) to predict which prospects are most likely to accept your offer.  The model creates a scoring algorithm producing a likelihood to buy score from 0% to 99% which is applied to your raw list.  The real benefit comes from creating what is called a ‘lift model’.  In the lift model, the prospects are rated from high to low based on likelihood to buy then typically divided into ten equal segments (deciles) which are evaluated for ROI.  Obviously you will mail to the top decile (top 10%) but will you mail to the bottom 10% or even the bottom 40%?  The cost of your promotion and the potential gross profit of your sale can then be analyzed in detail to determine, for example, that at the 4th decile (from the bottom) the cost to make the offer is equal to the potential profit from that group – and usually means you will make the offer only to prospects with a higher probability of buying.  Results depend entirely on your data and your model, but it is common to see outcomes in which mailing to the top 60% of the list (eliminating the cost of mailing to the bottom 40%) yields 80% to 90% of the sales from mailing the whole list, but now with a decidedly higher ROI.

Why they stay – Loyalty Programs

Retaining your best customers is job 1 after you acquire them.  Predictive modeling will help you identify who your best customers are (life time value analysis is one common technique).  Then it will help you identify which of your ‘best customers’ are most likely to respond positively to the loyalty program you have.  This analysis will also help you with modifications to your loyalty programs to maximize return.

Why they leave – Defection and Churn Management

The last thing you want is to lose a good customer but the reality is that they’re leaving all the time.  If only you had known 30 days, 60 day, or even 6 months ahead that your customer was getting ready to leave.  Predictive modeling can give you sufficient advance notice that you can take a specific set of actions to try to rescue the customer.  This type of analysis is used widely where turnover is high, such as in the banking, credit card, and telecom industries.

What else will they buy - Cross Sell / Up Sell

Now that you have a customer, whether your model is B2C or B2B, how can you predict:

    Who will buy next.
    Who will buy product X, or who will buy product X if they have already purchased product Y.
    Who will place an order over $XXXX (making them the financial basis for sustaining your business).

Using the same scoring principal as above, you can target your promotions, discounts, or other offers to exactly the right group of existing customers to maximize ROI.  An interesting issue here is that you may think you do not have much current information about your customers, perhaps only a shipping or billing address.  You have more information than you think. Frequently patterns can be derived from the value, recency or frequency of purchases, even from method of payment.  However it is often very worthwhile to append external data to your existing customer files.  Based on the address you have it is a regular practice to add the 15 or 20 top characteristics from the last census giving you much richer information on which to build your model.  PreMo can advise and assist you with any appended data that may be helpful.

A special note on model deployment:

You may be thinking of direct mail or an electronic offer to increase cross sell, up sell, or to reduce churn and defection.  However, these models can also be deployed directly in your call center software.  When your call center rep receives the in-bound call (or makes a planned outbound call), they can see the ‘score’ for that caller in real time along with your pre-programmed direction about what to offer or ask to encourage a new sale or discourage a defection based on the score.


Most often we think of risk management in financial terms.  Here are several examples of financial risk management and one from the world of manufacturing.

Fraud Detection:

If you rely on claims or bills submitted to you, you already know that some small percentage of them are fraudulent.  The problem is that the cost of evaluating every claim can easily exceed the value of the fraud.  So what to do?  Audit every claim over a certain dollar value?  That won’t get the results.  Predictive modeling can predict and score incoming claims and bills for the likelihood that they are fraudulent.  Now you can balance the cost of audit against the value of recovery and make an informed decision to audit all claims with a fraud potential score over X.  This is a major application for insurance, particularly workers comp and health care, and for any warranty or claims-based business.  In the public sector this extends to targeting tax returns most likely to yield revenue if audited.

Credit Scoring:

Do you rely on the credit scores of major retail or business credit raters to grant credit to your customers?  In many industries (mortgage lending is a prime example) the commercially available scores don’t line up well with actual credit experience.  Many banks, mortgage lenders, leasing companies, and other B2B commercial lenders use predictive modeling to develop custom credit scoring models.  Just as with deployment of the marketing solutions, these models could easily be embedded in for example your underwriting or origination software to ensure that all your lending or credit agents are using uniform criteria in extending credit.

Billing Review:

Ever think about the profits lost through accidental under billing?  Like fraud detection, a systematic model for reviewing bills before they are ever sent can detect the bills most likely to be accidentally understated and flag them for review.  Unintentional under billing is a major source of revenue leakage in many industries, especially those with frequent billings such as utilities and telecoms, and in the public sector, annual property tax bills.

Time to Failure:

Risk is ultimately financial, but in manufacturing knowing when a component, device, or system is likely to fail can avoid financial risk, allow a proactive sale of maintenance or parts, and create a very happy customer whose business was not disrupted by the failure of your device.  In aircraft this may be a life saving intervention.  With all capital intensive transport equipment (aircraft, buses, trains, trucks, lift equipment) and major manufacturing tools and equipment, anticipated maintenance is always financially preferable to unscheduled maintenance.


Planning and optimization is our catch-all category with some applications seen frequently, and some innovative solutions with high value still seen only rarely.  Here are just a few of the application categories in this domain:


Generally, forecasting occupies a special niche in predictive modeling called time series analysis.  The applications here are endless and can become very esoteric.  Here are just a few examples of forecasting solutions:

    Predict the cost of natural gas or electric power by the hour over the next 24 hours (very important to power producers).
    Predict the output in bushels from the crop planted in each of X specifically identified plots of a farm or region.
    Predict the price of a commodity, interest rate or security in the financial markets.  Predictive modeling is the meat and potatoes of the ‘quants’ on Wall Street.
Examples here are too numerous to list.


Predictive modeling can produce an algorithm that describes price elasticity against competitors in a market.  Using optimization techniques with the model, you can then determine the price appropriate to meet your volume goals or vice versa.  This type of modeling is used widely in new car and truck sales, particularly when trying to set the price of a new model.


This is a very high value application for predictive modeling that is most easily used in continuous process manufacture such as chemical and gas manufacture.  The model is based on the inputs of raw material and further guided by the pressure, temperature, and other measures taken by monitoring instruments at various locations throughout the manufacturing process.  A model can be created which mathematically expresses the output of the plant in terms of the inputs and other variables.  The model can then be used to optimize one or more variables such as maximum output, or maximum output constrained by a certain maximum value for environmentally hazardous output products such as waste gases or effluents.

Claim Denial Management:

Healthcare providers and others whose cashflow depends on the payment of a claim can dramatically enhance the management of their cashflow by anticipating which of their claims are likely not to be paid.  These can then be audited or reworked to meet the standard required by the payer without the long delay of submission, denial, and resubmission.