Understanding Choice in Banking: Use of Discrete Choice Conjoint Analysis

Hero Image: Understanding Choice in Banking: Use of Discrete Choice Conjoint Analysis

An important consideration for banks is how customers choose them for opening checking accounts. Given that checking accounts can often be the keystone of an entire relationship with a banking customer, the importance of this decision becomes even higher. How then do we understand choice in banking?

Traditionally, the focus has been on directly asking customers what factors they consider when opening a new checking account at a bank. This can be done qualitatively or quantitatively, and often both are done sequentially. In quantitative questioning, respondents are usually provided a series of reasons (or features of the bank) and asked about the importance of each one using an importance scale. There are several drawbacks to using such a scale [please refer to the article Identifying Feature Importance: A Comparison of Methods for a more detailed explanation]. The primary problem is that since no trade-off is involved, the respondent has no incentive to clearly discriminate between features. Hence, many features show up as being important without clear separation among them. A solution to this problem is provided by conjoint analysis.

Conjoint analysis uses a trade-off approach to get at respondent preferences. Instead of asking about features one at a time, respondents are shown entire products (described as bundles of features) and asked for their evaluation. In the particular form of conjoint analysis that we will be looking at (discrete choice), respondents are shown sets of product descriptions (bank checking accounts) and asked to choose the one they like the best in each set. This approach more closely mirrors what happens in the real world and hence provides better answers. As we shall see shortly, conjoint results are also provided in a form that is particularly useful.

To conduct a conjoint analysis one needs to pay careful attention to the features to be included in the study. In this case, specific attention needs to be paid to the features that consumers may consider when choosing a checking provider. Each feature needs to be defined with two or more levels, such that respondents can clearly see the difference between the levels.

An Example

In order to study checking provider choice, a conjoint study was run on a web panel. Eight features were chosen for the study with two to four levels per feature. Before looking at the results, a few caveats are necessary. While the respondents are distributed nationally, the sample is not representative and hence the results should be seen as illustrative and not prescriptive. Further, no allowance has been made for the fact that all of these respondents have online access and hence may be especially predisposed to online banking. Finally, only type of bank was included and not brand names. The features chosen for this study and the results are shown in Table 1.

Table 1:

Features Levels Utilities Importance
Type of Bank National -7 11%
Regional 7
Local Community 7
Credit Union -7
Balance/Fees No min balance and No monthly fees 138 34%
No minimum balance and $5-10 monthly fees -92
Minimum balance of $300 and no monthly fees 47
Online Banking/Billpay No online banking -99 24%
Free online banking 26
Free online banking and bill pay 72
Nearest branch Close to home 11 7%
Close to work -4
Supermarket where you shop 7
Branch hours Weekdays 9am-3pm -29 9%
Weekdays 9am-7pm 7
Weekdays 9am-3pm/Open Saturday 6
Weekdays 9am-3pm/Open Saturday and Sunday 16
ATM Network ATMs at branches only -7 4%
ATMs at branches and other places 7
Customer Service Reputation Excellent 15 7%
Good 3
Average -19
Prior Relationship You had prior relationship with the bank 0 5%
No prior relationship -10
Someone recommends it 10

Conjoint results are typically displayed as utility scores and importance scores. The utility scores are attractiveness scores associated with each level of each feature. The higher the utility associated with a level, the more it is preferred compared with other levels of the same feature. A negative utility score does not mean that level is unattractive per se. It just means that level is less attractive when compared to the other levels in that feature. It is entirely possible that consumers found a level to be acceptable even if it has a negative utility score. Importance scores of a feature are calculated as a function of the range of the levels within that feature. Hence, features where the levels have a wider range will be more important.

Table 1 (see above) shows the utility score of each level and the importance score of each feature. It must be kept in mind that these are aggregated scores. That is, the utility scores were calculated for each respondent individually (using an advanced statistical method called Hierarchical Bayes estimation) and then averaged to produce the scores in the table. To calculate the importance scores, the individual utilities were used and not the aggregated utilities shown in the table. That is why Type of Bank has an importance score that is well above zero (11%). If calculated from the aggregated utility scores, its importance would have fallen to near zero. This type of result indicates that there are clear differences between respondents in terms of how important the type of bank is and shows the usefulness of using individual level utility estimation.

Implications of Conjoint Analysis Results

As can be seen,  Balance/Fees and Online Banking/Billpay are the most important features to these respondents overall. The former is not surprising and the latter could be affected by the fact that this is an online sample. Separately estimating the utilities for those who currently use Online Banking or Billpay and those who don’t could produce different results for the two groups.

The location of the nearest branch gets only a 7% importance score, which might seem low. To understand this, we have to look at the levels that describe the feature. All three of those levels are quite convenient for a customer. The only question is which is most convenient. At the aggregate level, there isn’t much difference, with a location close to home being somewhat more convenient. It is possible there are segments of respondents for whom the utility scores may vary. But it is also true that more variation among the levels would have made this feature more important. For example, if the levels had been defined as being within walking distance, a five-minute drive and a twenty-minute drive, the importance of this feature would likely be very different. For a smaller bank that is trying to understand the importance of location, this type of level specification may be more appropriate. For a larger bank that is trying to identify the ideal location (given that it is going to be convenient), the levels used in this study may be preferable.

In terms of convenience, there is a preference for banks that are open over the weekend and that have ATMs in many places. Regional and community banks seem to have a slight preference, at least in this sample. Excellent customer service is preferred, but merely good customer service seems acceptable. In terms of prior relationship, we did not specify that relationship as being positive and hence a recommendation seems more valuable. While all of these factors have importance, the overwhelming importance of the checking account features cannot be denied. Monthly fees are a definite no-no, while free checking is a big plus. Amongst competing banks that are offering free checking, many of these other factors may come into play, but if free checking is not offered the competition can quickly become lopsided.

What Else?

Conjoint analysis results can also be used to run simulations to understand how attractive different bank profiles would be to these customers. If we were to create, say, three competing bank profiles, it is quite straightforward to identify the proportion of respondents who would choose each bank. Of course, inclusion of brand names would make the exercise much more interesting.

Given that utilities are available at the individual level, it would be possible to conduct segmentation analysis on these results to identify preference based segments. These segments are likely to be more distinct than those based on traditional importance scales.

In conclusion, studying bank choice by using conjoint analysis can be very fruitful, as long as the conjoint is properly designed and analyzed to meet the objectives of the sponsoring bank.