“e-Push or e-Pull? Laggards and First-Movers in European On-Line Banking

Authors

  • Jacques Bughin

    Corresponding author
    1. Partner principal at McKinsey & Company and holds a Ph.D. in economics and operations research. His e-commerce research concentrates on profitability assessment of business models on-line as well as marketing effectiveness for Internet companies. He has published about 50 articles in leading international journals such as Electronic Markets, Journal of Industrial Economics, the McKinsey Quarterly, Management Science, and European Economic Review.
      Address: McKinsey & Company, Av. Louise 480 b22, 1050 Brussels, Belgium. Tel. +32 2 645 42 11  Fax. +32 646 45 48.
    Search for more papers by this author

Address: McKinsey & Company, Av. Louise 480 b22, 1050 Brussels, Belgium. Tel. +32 2 645 42 11  Fax. +32 646 45 48.

Abstract

While a strategic imperative for financial institutions, major differences currently exist among banks' abilities to convert their customers to using the Internet. In addition, not a lot is known as to what drives those differences, although one can easily hypothesize that on-line banking adoption must be due to a combination of demand (“pull”) and firm (“push”) factors. This paper presents a statistical appraisal of the determinants of on-line customer penetration for a cross-sectional sample of the major incumbent banks in Western Europe for the combination of the years 1988 to 2000. “Pull” factors play a large role in explaining customer conversion to Internet banking, yet bank-specific factors (or “push” factors) are not marginal. Among others, banks with traditionally high cost-effectiveness and that already offer wide private ATM coverage for their customers are also the ones which have already started to migrate a larger proportion of their customer base on-line. Interestingly, among all “push” factors analyzed, cost-effectiveness emerges as the largest leverage effect on customer conversion. Finally, a cluster analysis supplements the regression results along the axes of “pull” and “push”, and identifies a set of early movers and laggards among the banks in our sample. Generally speaking, those first-movers already exhibit stronger off-line profitability than laggards, which may indicate that the on-line marketplace may reproduce performance off-line (i.e., currently successful banks have nothing to fear from aggressive start-ups).

Introduction

The rapid rate of adoption of the Internet is well known. Today, it has millions of users worldwide after just a few years of commercial existence (Nua Internet Surveys, 2000). It has also grown significantly economically, generating more than 500 billions of US revenue by end of 1999, (Internet Indicators, 1999). Using the weights applied by the Bureau of Economic Analysis, the value added by this revenue in the US is about $260 billion US, or already 2.7 % of the American Gross Domestic Product (Landefeld and Fraumeri, 2000).

With this pace of adoption, the Internet will inevitably make some industries become much more literate electronically than in the past, in particular the industry of banking and other financial services, which is well-suited for on-line: as Field and Hoffner (1998) indeed acknowledge, financial services have the advantage that no physical goods are involved, as any transaction can be successfully described by a contract electronically.2

In Europe alone, the number of banks moving to the Internet doubled in less than six months to be at more than 2400 web sites by end of 2000, while seventy percent of those sites already offer advanced services, such as on-line transactions, money transfer, bill payments or stock trading (eMarketer, 2000).3 Furthermore, on-line banking allows banks to develop new services such as personalized financial information menus, e-mail and real-time brokerage, which will reinforce the value proposition, and therefore, the customer retention of banks that provide these bundles of services (Bakos and Brynjolfsson, 2000). Finally, Internet may provide cost-savings for banks. Many analysts, including the lead research analyst Mary Meeker from Morgan Stanley Dean Witter (1999), have recommended banks offer on-line services in order to benefit from the low cost delivery channel offered by the Internet as opposed to the branch network. Compared to ATM's or the recent VRU (voice recognition unit) and PC-banking, the Internet has the breakthrough advantage that the use of a browser eliminates most of the network and software costs attached to other electronic banking innovations, with the consequence that the unit cost can be only a small percentage of traditional costs (Clemons and Hitt, 2000; McQuivey, 1998; OECD, 1999).

Despite all this, the adoption of on-line banking worldwide, and in particular in Europe, still exhibits wide variance. At one extreme, Finnish banks are announcing customer conversion rates of greater than 20% to on-line, while, at the other extreme one of the largest Italian bank had almost no customers on-line by early 2000. True, Nordic banks are well-known to have the highest penetration of Internet usage worldwide, at least higher than Italy, but disparities among banks within the same country are also noticeable. France is another example: The leading universal bank has been able to date to convert more than 11 percent of its customer base in 2000 to the Internet, while other minor players such as the Credit Commercial de France have only managed to convert a mere 2 percent in the same period.

Given those disparities, and the importance of on-line banking as a potentially crucial electronic market for banks (Chen & Hitt, 2000), the question addressed in this article is to quantify and sort out the various drivers of on-line banking, specifically what mix of “push” and “pull” factors drives usage. Relying on a unique cross-sectional sample of incumbent banks in Western Europe for the year 2000 and using logistic regression techniques to identify the relative contribution of “push” and “pull” factors, the contributions are twofold:

  • First, significant “pull” effects are statistically visible in Europe to explain on-line banking differences (e.g., on-line banking is systematically linked to a country's level of Internet penetration as a measure of a customer readiness to transact on-line), yet “push factors also prevail: Banks with low cost-ratios and/or dense ATMs coverage have managed to convert more of their established customer base on-line than other banks, even after controlling for readiness to transact on-line.
  • Second, an investigation into clustering banks along the “pull” and “push” axes identifies five statistically separate clusters, one of which clearly includes the best performing banks in managing the conversion of their customers on-line in the best electronic-ready markets of Europe. This set of clear winners also contrasts with a set of laggards among European banks, whose poor on-line performance correlates with their poor global return performance. Otherwise stated, the “digital divide” is only a symptom of market performance distribution at a more global level.

Data and Hypotheses

In this section descriptive statistics on on-line European banking are presented. Those data will be used later on for a more robust statistical analysis of the drivers of on-line banking conversion.

Hypotheses

Three major hypotheses are being tested in this article. First, on-line banking penetration should be linked to consumer readiness to transact on-line, itself depending on tenure on-line, and on some key-demographics population segments, such as the mostly time-starved, highly-educated individual (e.g., Bellman et al., 1999; GVU, 1999; Lohse et al., 2000).

Second, from a strategy perspective, banks pursuing a global cost-effectiveness strategy to differentiate themselves versus the competition will likely be the ones which have been pushing the on-line channel faster than others. Additionally, one would expect that the organizational structure for customer acquisition is more flexible, possibly separate from the established brand to replicate a “pure attacker” mentality (Hagel & Singer, 1999).

Finally, we are interested in the question of the “digital divide” among banks: Are banks which are leaders at converting their customers to online banking “capitalizing” on chance or is there something truly distinctive about these banks compared to laggards? The question is important for industry dynamics as it is often mentioned that the Internet may favor industry concentration in the long-term (Bailey & Bakos, 1997; OECD, 1998).

Data

In order to test the hypotheses above, a sample of Internet banking penetration for 65 western European banks (i= 1.65) has been compiled from Investment Banking reports by Merrill Lynch and J.P. Morgan for the years 1999 and 2000. The sample covers ten Western European countries (j= 1.10) from France to UK4 and comprises all well-known banks.5

The sample collected includes statistics of the bank's total client base (henceforth, symbol= Cli,j), on-line conversion, i.e., percentage of on-line bank registered users divided by total client base (CONVi,j) as well as the Internet banking organizational structure (ORGi,j= 1, if structure is standalone).

The dataset has been supplemented with country Internet penetration and country e-commerce usage data from IDC (Internetj and ECOMMj)6, as well as with traditional bank statistics directly downloadable from IBCA, including

  • (a):indicators of size, i.e., balance sheet assets (ASSETSi,j) and full-time employees (EMPLi,j);
  • (b):indicators of profitability, i.e., return on average assets (ROAAi,j) and return on average equity (ROAEi,j)7;
  • (c):indicators of cost-effectiveness, i.e., cost-income ratio(C/Ii,j); cost-assets ratio(COAi,j= C of C/I divided by ASSETS);
  • (d):indicators of non-brick banking, i.e., ATM coverage per customer (ATMi,j= number of privately owned ATM's. divided by CL).

All the indicators above can be seen as “push” factors. We also have used the data to build other possible indicators of “pull,” such as the “wealth” of a bank franchise, as measured by ASSETS per customer. In general, it is well-known that the first cohort to move on-line has been the most wealthy sociodemographics, as in the same way, they are also likely to be the first to use on-line investing possibilities (Bakos et al., 2000).8

Those data are readily obtainable from the source mentioned and are harmonized in their definition so as to provide meaningful comparative analyses. Descriptive statistics on the rough data are provided in Table 1, with the following highlights:

Table 1.  Sample features. Notes: STDV=standard deviation; MIN/MAX= minimum/maximum in the sample For Sources: ML= Merrill Lynch; JPM =JPMorgan; MMXI =MediaMetrix/Jupiter Communications * Billion Euro; **:basis points; ***: percentage; ****thousandsThumbnail image of
  • First, average on-line customer conversion for 1999–2000 among the European banks is still relatively low, averaging less than 6% of total bank customers. With respect to the average on-line population, this represents about 17% of Internet users, i.e., less than one on-line user among five had converted to on-line banking by the turn of the century in Western Europe.
  • Second, on-line banking penetration is about 40% of e-commerce usage, i.e., 60% of Internet users transacting on-line have not yet converted to on-line banking. This confirms GVU (1999), and Lohse et al. (2000), whose data illustrate that security is still a large issue for some e-commerce users; while on-line visitors will have no issue buying low value goods such as books and CDs, they are still reluctant to perform on-line banking, where more sensitive information may be exchanged on the public web infrastructure.
  • Third, the data exhibit strong variance with large ranges between the minimum and the maximum observed values. In particular, on-line banking conversion, CONV, has a standard deviation which is larger than its mean. In other words, assuming a normal distribution, the top 5% European banks may have already converted close to 20% of their customer base by the average of 1999–2000, while some other banks would virtually have none of their customers converted to banking on-line. Henc, this paper endeavors to identify factors that have driven this superior conversion ability among some banks.
  • Finally, other statistics are worth mentioning: There is an average of one ATM per 2800 customers, while front-end branch employees (representing about 35% of total employees) have a span of 500 clients. Given that average branch size in Europe is known to be in the range of five employees, this means that the ATM coverage is now as large as branch density in our sample. With this extensive density coverage, banks with strong policy of ATM development might also be the ones to pioneer on-line push to their customers.

Explaining On-Line Conversion Difference

The empirical analysis intends to explain inter-bank variance of on-line conversion, through multivariate regression. Before this, a quick check on the partial correlation between all variables of Table 1 is necessary, as it is likely that many indicators are correlated with each other, and we may wish to reduce the multicollinearity dimension of the analysis.

The multivariate model

A logistic model for the multivariate analysis is used because of the well-known fact that the logistic S-curve replicates the consumer adoption curve (Bass et al., 1994).

The generic equation estimated via ordinary least squares is of the form:

image

where:

  • X, Z are respectively vectors of bank-specific regressors and country-specific Internet penetration indicators;
  • a,b',c', are coefficients to be estimated,
  • u is a disturbance term assumed to be normal and i.i.d.

The vector (X) may theoretically include all variables, as such or combined with each other, described in Table 1, i.e., ORGi,j, BROKERi,j, EMPLi,j, Cli,j, ASSETSi,j, ROAEi,j, ROAAi,j, C/Ii,j, COAi,j, and ATMi,j. The vector Z is composed of WEALTH, INTERNET and ECOMM.

Partial correlation analysis

Before estimating the multi-variate model through regression techniques, Table 2 illustrates the set of partial correlations that are statistically significant at the risk 5% threshold, among all the possible regressors.9Table 2 demonstrates that some variables are indeed very much correlated, with evident clusters. For example, CL, EMPL and ASSETS are all indicators of bank size. ROAE, C/I, and ROAA are all correlated indicators of a bank profitability. Also, cost per assets, COA, and cost income, C/I, are correlated indicators of cost-effectiveness, while Internet penetration and e-commerce usage, finally, are clear indicators of a country “e-literacy”.

Table 2.  Correlation matrix. Note. Only significant correlations at 5% are shown.Thumbnail image of

In order to minimize so-called “collinearity”, but while preserving the most information in the data, a factorial analysis via traditional VariMax rotation has been performed on each group of correlated variables. The new “regressors” constructed are then built using the first principal factor (henceforth, F). In practice, each factor explains more than 50% of the variance among the indicators built10; those factors are labeled F1 for Size; F2 for profitability; F3 for cost-effectiveness. F4 stands for on-line readiness.

“Push” or “pull”? The multivariate regression results

The regression results, after correction for heteroskedasticity, are summarized in Table 3.

Table 3.  Logistic regression of online banking. Note. Only statistically significant coefficient reported (α <10%). Cross-effects among RHS variables excluded, but not significant at traditional risk level.Thumbnail image of

For illustration, we have used two empirical variants of equation (1) below, with (1′) using the first principal component of the factors, while the version (1) picks the highest partial correlate as variable in the multi-variate regression.

For instance, concerning the size “cluster,” EMPL has been chosen rather than ASSET given the largest correlation observed with CONV in the rough data (EMPL partial correlation with CONV is negative, and =−0.31, versus -0.12 for ASSETS, suggesting in passing that smaller banks can be more prone to convert customers to on-line). Finally, only statistically significant coefficients are shown in Table 3, using a risk threshold of ten percent. From the adjusted R-square of the equation, especially version (1′), the model has a sufficiently good fit for interpretation analysis. In addition, and as discussed hereafter, the coefficients have the signs expected from our early hypotheses.

“Pull” wins-“Push” is relevant

But first, we assess the relative weights of “push” and “pull” in the evolution of on-line banking. Hence, using the regression results, Figure 1 splits the contribution to variance of “push” and “pull” factors for equations (1) and (1′), after cancellation of multicollinearity.

Figure 1.

Mix of “push” and “pull” factors in European on-line banking conversion.

In both cases, about two-third of the variance is explained by the “pull”, but this also means that other factors linked to bank strategies are also relevant (one third of the variance contribution explained), and must be analysed more deeply.

On-line readiness

Figure 2 ranks the elasticities at the mean, based upon equation (1) and (1′). The elasticity of on-line banking to Internet penetration is not only positive (as expected), but the elasticity estimate is also the highest in absolute value among all the regressors. Furthermore, the elasticity is higher than one at the mean, suggesting increasing on-line banking when Internet usage will become widespread in Europe (and not only in Scandinavia alone for instance).

Figure 2.

Elasticities for on-line banking conversion.

To illustrate this point, Figure 3 shows the elasticity function derived from the logistic regression (1), and plotted against country Internet penetration. From the curve, it is seen that the on-line bank elasticity becomes higher than one when a country achieves a penetration of about 30 percent of quarterly Internet usage, e.g., countries such as Germany, UK, or Norway which have already bypassed this Internet threshold will start disproportionately to increase on-line banking usage. Currently, countries such as Denmark or Norway exhibit elasticities as high as 2.0, in contrast to, say, Portugal with elasticity being less than 0.4, 5 times lower than the Scandinavian countries. If Internet penetration continues the same pattern of doubling every year, this means that Portugal is at least two years beyond the development of Scandinavian countries, an often quoted metric in Europe when comparing the North and the South (e.g. OECD, 1998).

Figure 3.

The Internet elasticity function of online banking.

Bank-specific factors

An analysis of bank-specific, “push” factors also has yeilded its own interesting points.

First, bank size exerts a significantly negative impact on bank “push” strategy to convert on-line customers. Otherwise stated, smaller banks are more active in converting their customer base to on-line channels. The elasticity at the mean is, however, small at about 0.2 in absolute value.

Second, banks with lower cost structure and which have already established a large electronic channel base (as measured by private ATM density) tend to convert more customers to the on-line channel. This means that banks with low cost strategy recurrently push towards lower cost channels, such as the on-line channels.

Third, and of noticeable importance, the “cost-effectiveness” measure, rather than the “convenience” measure (ATM density) has the largest elasticity in absolute value, confirming the aggressiveness of cost effective banks to adopt the Internet where it is argued that Internet transactions can be made at less than 10% of traditional channel costs (Morgan Stanley Dean Witter, 1999).

Fourth, organizational structure of on-line bank activities never appeared in this sample of banks to be a discriminative factor. Organizational structure possibly plays a stronger role for customer acquisition rather than customer conversion, as emphasized in companion work (Bughin, 2001).

Finally, and in all cases, the elasticities for bank-specific factors are also much lower in absolute value than for the effect of a country Internet readiness. However, the sum of those elasticities can be above one, in particular for the hypothetical bank which combines lower cost structure, larger ATM coverage and smaller size than average. How many banks share those features for on-line banking penetration is a question discussed in the next section.

Clustering of On-Line Banking

Our logistic regression analysis identified three factors, cost effectiveness, customer convenience, and on-line readiness as significant influences on the proportion of a bank's customers who bank on-line.

Five clusters

In parallel to regression analysis, we performed a cluster analysis to identify where individual European banks are located in the space defined by those factors. The analysis was run using standard clustering methods using the SAS package. After a tolerance level of 10 percent, five clusters were identified for a total explained variance of 42%; however, customer convenience never appears as a strong cluster component, and has thus been ommitted. The clusters in terms of cost-effectiveness and Internet readiness are described in Figure 4, with examples of banks belonging to each of them.

Figure 4.

Cluster analysis of banks.

The first and largest cluster (48% of the total of clusters) is the one describing banks centering their performance at the mean of cost effectiveness and on-line readiness. Another small but relevant cluster is the one composed of banks that are pulled by customer readiness to use on-line; this cluster (10% of the banks) comprises exclusively Scandinavian banks. On the opposite side, a cluster of banks (18%) has a strong on-line penetration because of pushing for on-line; this includes mostly French or German banks. Finally, two clusters define the extremes of high and low efficiency and Internet penetration. The first, a small nucleus (7%) comprising the best performing banks overall, combining strong pull/push factors for on-line conversion, includes Scandinavian banks such as SvenskaHandelsbank. The second, of the same size (7%), includes banks with relatively poorer levels of “pull” and “push” factors.

Digital divide?

Membership of the clusters identified was also correlated with other factors that are independent of the two main drivers of on-line banking conversion: Internet penetration (“pull”) and cost efficiency (“push”). As illustrated in Figure 5, membership in the clusters with the highest and lowest performance on the pull and push dimensions is also correlated with profitability measures such as bank cost-income (C/I) ratio and return on equity (ROAE).

Figure 5.

Correlates of clusters.

The fact that the highest performing cluster is positively correlated with overall financial performance (and vice versa) means that this digital divide in on-line banking is also the partial consequence of off-line performance. This topic of course requires more than these data alone, but points already to an interesting new field of research.

Conclusion

This paper is, to our knowledge, the first to assess the drivers of on-line banking customer conversion for European banks. While the conversion is still low, on-line customer conversion is shown to be linked to “pull” factors, but also significantly to “push” bank-specific factors. Larger banks and/or banks with higher cost base seem as well to be lagging their peers in terms of on-line adoption, while banks with an established electronic coverage (as measured by their private ATM density) have also managed to convert more customers to the Internet. The fact that banks evolve around significant clusters, with the most poorly performing clusters also being linked with globally weak financial bank performanc, may mean that on-line will reinforce the gap between high performing and laggard financial institutions. This calls, however, for further research in the future in order to confirm the results reported above.

As a first priority, however, the existing sample should be extended to a larger amount of banks, both in Europe and the US, potentially with a time-series dimension to reflect upon customer speed of adoption. Those data are scarce at this stage, but results reported here suggest that major difference in customer adoption to on-line is a critical research area for the years to come.

Footnotes

  • 1

    This paper is based on a research program launched by the Digital Economy Lab of McKinsey & Company. The author thanks his DEL colleagues (Michael Zeisser, Byron Auguste, Driek Desmet, Marc Singer and Jurgen Wunram) for their support in developing this article. The article has also been significantly improved based on comments by a referee.

  • 2

    Software also is another product category with the advantage of being automatically downloaded on-line upon the settlement of a transaction.

  • 3

    For another proof, a joint AOL/McKinsey proprietary study conducted in 1999 had shown that more than 30% of customers may want to leave their traditional financial institutions as the result of the convenience offered by the Internet.

  • 4

    The 10 countries are illustrated in Figure 3.

  • 5

    The sample among others include banks such as BNP-Paribas, Deutsche Bank, ABN-Amro, or Credit Suisse or Halifax.

  • 6

    The on-line penetration is based on quarterly users, while e-commerce users are defined as users who have bought at least once per quarter.

  • 7

    The average assets and average equity measure reflect the average assets/equity holdings during the year.

  • 8

    The referee suggested this point.

  • 9

    WEALTH is not shown in the correlation matrix as it never appeared as significantly correlated with any other variable.

  • 10

    Analyses are available from the author upon request. The lower factor weight is for profitability where the first factor explains 51% of the variance.

Ancillary