Machine Learning (also known as Data Mining) tells you what you don't already know about your data. It is a form of artificial intelligence that takes your historical data and learns how your organisation works. It finds hidden patterns in your data which you can then use to determine why something happened, or whether something will happen.
Perhaps the most common use of Machine Learning today is in the financial industry, where among other things it is used to determine whether someone passes or fails a credit check.
Models are generated using historical information of who has and who hasn't defaulted on loans, and then the model learns what common factors have influenced each outcome. When a new loan application is received, the details of the individual are entered into the data mining model, which in turn predicts the likelihood of that individual defaulting on payments.
No specific rules are entered that say how old someone should be, or what their income should be; the data mining model determines its own set of rules based on previous experience.
We can also use Machine Learning (ML) to predict the short or long term future of a value, such as sales, stock levels, energy usage or stock prices.
There are so many factors that influence these values, that a traditional approach of trying to map them is usually too unrealistic. Data Mining however, provides an ability to enter the results, and let the AI routines determine their relative influence. If time is a factor, then the model can propagate the results forward, and provide a future prediction of the value.
A more generic example, and one that applies to virtually any organisation is that of minimising marketing costs.
The standard expected response rate for a direct mail campaign is only 0.5%. Data Mining can analyse the response from previous campaigns, and create a model of the various successful demographic groups that are most likely to convert into customers.
The model can then be applied to any direct marketing lists, in order to isolate the consumers that are most likely to respond.
If the same sales can be generated from half the direct mail costs, significant savings can be made. For example in the chart opposite, a 50% reduction in marketing spend has resulted in only a 7% drop in converted sales.
It's commonly accepted that gaining new customers is more expensive than retaining existing customers. Therefore successfully predicting which of your customers are likely to leave you can provide significant return on investment in any Business Intelligence platform.
By looking at past customer behaviour we can look at their purchasing or activity habits and use ML algorithms to build a model that can be applied to current customers, and to predict their propensity to churn.
Using this to focus marketing activity at these customers can increase your chances of retaining their business in the future.