Regardless of how you feel about this election, today is a life-altering day for those data scientists that develop voter response models. In yesterday’s post (HERE), we discussed how by understanding your customer or potential customers party affiliation can give deep clues as to their persona. We also looked at how party affiliation and certain brands were linked together and then discussed a framework how by asked just a couple questions on each, you can glean a deep array of information that can help you be multiple times more effective in your marketing program. In today’s post, we take a look at how existing voter response models can be leveraged to understand if potential customers could be interested in switching banks.
Voter Turnout Models
If you ask if a person is going to vote, much more say “yes” than do This is why firms like Gallup make a business out of trying to find the right questions to ask to see if a person is likely to vote. For example, if you ask if you voted last year and the answer is “no” then chances are you are not likely to vote this year. However, by asked some six additional questions such as do you know where your voting place is and how much have you thought about the election, survey firms like Gallup can get accurate to the range of 10% to 15% percentage points. Thus, if you ask 100 people just seven voter turnout questions, just using the basic responses, these firms can accurately guess 85% to 90% correctly.
Note here that we make the difference between sentiment and action. Political candidates, like banks, care about sentiment, but they care about action even more. FBI Director Comey’s letter is a classic example that while it served to shift sentiment away from Secretary Clinton, it also served to increase the likelihood of voting for a Clinton supporter. Thus, it was not all that clear what the net impact was.
While 85% to 90% accuracy isn’t bad, in this day and age, we can do better. By adding in third-party data such as checking if the person voted against a publically available database of voters, firms can add validity to the data. Note that this data is often available for free (like the sample below) where banks can look up any given area or individual or lists can often be easily purchased. If the person says they did vote in the last election and they did not, chances are they are not going to vote in this election as well despite how they answered any of the questions. We can also append items like homeownership, household size, age, and number of vehicles owned to make a more robust file about each individual.
This is where things get really interesting. Now, we can look at any down ballot measures such as other candidates, tax assessments, the legalizations of marijuana and such and see if those issues are motivating factors to your persona. Even more interesting, not only do those data points have an impact on your likelihood to vote, but also how you will vote. For example, it is no surprise that if you own your home, you are more than 90% certain not to vote for any candidate or any measure that does away with the interest deduction (which explains why no successful candidate has ever supported that measure).
Firms can then add machine learned algorithms to the data to further increase a model’s accuracy. For example, if we know the response and validity of if your neighbors are going to vote, we can then make an inference as to what you are going to do. The more your neighbors are likely to vote, the more you are likely to vote. By correlating your responses to third party data and then looking for associations and correlations, polling firms can now get accurate in the range of 3% to 5% - a huge improvement for very little extra effort.
The problem is that people switch primary banks just slightly more than they switch religions. Even adding a new banking relationship has a long sales cycle. This is why this topic is important as if you can narrow down your field of potential customers, you can dramatically shorten the sales cycle. Similar to voter response models, particularly voter turnout models, if you ask a person if they are open to switching banks, many more say “yes” that actually would. This is why bank marketing models ask a series of additional questions. By way of example, here is the basics of our bank customer switch model:
- On a scale of 1 to 10, 1 being hate and 10 being love, how do you rate your current relationship with your primary bank?
- In the past 90 days, have you had a problem with your bank?
- Are you planning on moving in the next year?
- Are you planning on applying for a loan in the next year?
- Would you switch to a bank that made your life easier and more productive?
- When is the last time you opened up a new bank account or applied for a new product?
- Are you open to switching banks in the next 90 days?
Similar to voter turnout models, respondents get points for how they answer each question. The more points you get, the more likely you are willing to switch banks. The model is simple, straightforward and while not as accurate as voter turnout models; you can still find out if a customer is likely to bank with you with about 75% accuracy. If that doesn’t seem like much, consider that your average community bank is only about 7% accurate at predicting if a customer is likely to switch without using a model. That type of improvement in accuracy tends to make your marketing dollars go further and be much more effective.
Behind The Model
What is important for every banker to understand is the framework behind the model. On average, a bank product or experience has to be five or ten times better to get a potential customer to switch banks. That is hard to do in this day and age except for our industry’s most creative banks.
Absent of a vastly better product or experience, people, and businesses are reluctant to switch banks and will usually only do so if they have had a problem or are moving their residence or office. Even if a person or business is not open to switching banks and they move locations, statistics show that they are more likely to switch banks. The same is true for a person needing a new banking product such as a home loan or loan to finance a child’s education. Regardless of how they feel about their current bank, they are more open to other products and banks.
By augmenting the above questions with additional data, such as demographic, behavioral, and other, you can further improve your bank’s conversion targeting. Revenue growth over 14% AND employee growth over 10% are an example of two indicators when taken together foretell a business is more open to a new banking relationship. On the household side, babies, the number of kids, job promotions, boat purchases and other items also are predictive actions of being more open to a bank.
The data can also be further broken down to allow for all sorts of granularity. If a person, is likely to switch banks, by knowing their age we can determine the likelihood of switching (see graphic below). If they are 30 to 44 years old, they have the highest probability of switching banks if they are unhappy with their current bank or say they are in the market for a new bank. Further, if the respondent is over 60 years of age, there is the lowest probability of switching.
More to that point, if you purchase or keep enough data, you will also know that depending on your age, different factors motivate you. Those over 60 are more tolerant of poor service than those in the 18 to 29-year-old cohort but are also more likely to stay because of great service. Rae matters more for those over 60 and could alone cause a person to switch banks while if you are between 45 and 60, fees are one of your largest motivators.
What banking products you use also matter. If you are a consistent user of mobile banking, then you are not likely to switch banks, for example. The other major product that is a predictor of bank switching is bill pay.
Finally, there is a whole list of external items that also statistically help in conversion. If a person’s or business’ primary bank is being acquired, that makes a person more open to bank change. New legislation, tax code changes or available products also help cause people and businesses to reconsider their banking future.
This topic gets even more interesting when you consider the Heisenberg Principal-aspect of this. If a potential customer even agrees to take this survey, regardless of what the answers are, they are more open to moving banks. Thus, it pays to either use these questions or develop some of your own that your bankers can either formally or informally ask.
We also point out there is no mention of rate in the questions. Banks have asked questions to the effect of, “If we paid you 1.50% on your 1-year CDs would you open an account?” They might, but what happens is that you obliterate any value of the survey question and end up self-selecting a rate sensitive customer while training all your other potential customers and your employees to be rate sensitive.
Putting This Into Action
The goal with these banking questions isn’t to market the customer, but to lay the groundwork to better understand the customer enabling you to be a more effective marketer. Creating a more methodical framework about client prospecting can reap huge rewards. Using an existing model or creating your own like the one above is a fairly easy step and can help your bank not only individually target potential customers, but lets you do a statistical sampling of areas where you might want to expand.
Developing a bank customer response model may not be for every bank, but given that the methodology and data are widely available, bankers would be remiss to ignore such a powerful tool. This election is going to be bad for banks; it is just a question of how bad, “bad” is defined. However, the silver lining is that we will be getting some great data and even better modeling from this cycle. At least, this is one bright spot.
Submitted by Chris Nichols on November 08, 2016