How to Predict, with 90% Accuracy, Who Your Best Customers Will Be
So you want to optimize the amount of success you have converting your customers? Well, one approach could be to optimize the customers with which you choose to do business. In other words, only market to customers who can really get value from your product.
Not only can you gain more business, but you can also find customers who are more compatible with your organization. This allows for smoother transactions with a higher success rate, which in turn raises the profit with fewer headaches.
How do you do that? It comes down to some math, namely statistics. If you have a data analyst on your team or in your company, I’m going to show you one tactic they can use to help you choose customers to market to who are much more likely to choose your product.
This analysis can even help you set pricing. After all, customers who can get more value from your product will likely pay more for it as well, or, at the very least, need less incentive to encourage them to buy.
Partition analysis helps you predict who will buy
Partition (or decision) trees are a multivariable statistical approach to identifying and classifying members of a population into groups based on a set of dichotomous attributes that are unique to them. The first step, just copy and paste that sentence into an email to your data analyst to show them that you know what you’re talking about.
All the above sentence really means is that you can use these advance statistical approaches to separate the wheat from the chaff of potential customers.
One of the benefits of this type of multivariate analysis is that on top of classifying groups, they can be used to predict which group a particular individual member of the population will be in. If your current and potential customers make up the population, this method will tell you if the potential client you seek to do business with will be a great fit for your company, or if they will be more hassle then they are worth.
In other words, you’re looking for potential customers who have attributes similar to those who have already bought from you. It’s the marketing equivalent of your buddy asking you if your football-loving girlfriend has any sisters.
What data is available to analyze?
Every company has certain information that is readily and legally obtained before you begin your business relationship. If you can identify variables that tell you the size, attitude or any other characteristic important to you, then you can know before you even get started if you should focus more of your efforts and budget on doing business with them (or if you should charge them more) based on these readily available attributes.
This can save a tremendous amount of time (which is money) and (actual) money you would have spent trying to secure their business.
How many complex sales marketers have invested into a customer just to have the relationship fail before you ever recovered those expenses or made money? How many direct sale marketers have spent time and money going after a customer segment that simply underperformed? Those days could be over, or at least reduced, based on this partitioning technique.
How does it work?
If you can identify those current customers with whom you consider your company to have a great relationship (whether due to higher propensity to buy or lower likelihood of needing expensive follow-up care) and those whom you consider to be … uhh … less than amazing, then you can begin.
To identify what makes them fit into either of the two groups, you can collect the information you would be able to get for any customer (frequency of purchase, price point, etc.) and begin your analysis.
An example analysis
In a recent analysis I did for one of our Research Partners here at MECLABS, we looked at these four attributes:
- Number of employees
- Company revenue
- Type of company
This helped us find the most valuable potential customers. You can find this information in your own data, third-party data and some information that publicly traded companies must report.
Once we had the information, I used the partition analysis to group the customers based on these attributes.
When the analysis was finished, we were able to predict the best customers with 90% accuracy.
If you were 90% sure your new potential client was going to be a great customer of yours, wouldn’t you do more to earn their business?
The other side of the coin showed — with slightly less accuracy at 83% — which customers would be a worse customer with which to do business.
Once again, if you knew there was a high chance that the client or customer segment was going to be one of your worst customers, would you try as hard and invest as much into gaining their business?
So, work with your data analyst to learn and use partition analysis to make sure that your initial investment in a new customer will actually be beneficial in the future, and you will have a happier and more profitable customer funnel to maximize your profits.
Why were we 90% confident?
Using statistics, there are different validation methods one can use to test the accuracy of their predictive model.
In our case, we used k-fold cross-validation, which splits the data used to build the model into sections. Some sections of the data were used to come up with the prediction (known as “training”), and some of the data to test the accuracy of that prediction. Basically, once the model was built, we ran the data of customers whom we already knew the outcome for, and our model came up with the correct answer 90% of the time.
When you talk to your data analyst, please note that k-fold cross-validation is only one type of validation method, and your analyst might prefer another method.
U.S. Bureau of Labor and Statistics (can help you find information about potential customers)
Investigative Resources International (can help you find information about potential customers)
Recursive partitioning (via Wikipedia)