Traditionally the core of bank margin is in risk management. The core of risk management is data.. thus Banks have been the among the best data businesses (as IBM knows). Banks “learn” about their customers through bank interaction: payroll, card transactions, lending. This has helped banks make better risk decisions (both credit and fraud/identity). Within the bank data cycle the traditional use of data is for an internal benefit: risk and cross sale of the bank’s products and services (not that of consumers or merchants). However the “virtuous cycle of banking data” is very different from that enjoyed by Amazon and Google, both in the scale and type of data and consumer facing use.
Capital One led expansion of bank data use in the early 90s in order to “construct a more integrated and scientific approach to marketing bank cards” (see history of Capital One). COF thus expanded the internal use of data to external marketing, and their tremendous success led others to follow suit. The next milestone for the banking industry is to use data for the benefit of the consumer: to take part in creating great customer experiences…this is not an easy task.
Banks NEED a new data business… Why? As data economies of scale diminish, risk management and identity are no longer sustainable competitive advantages (see Changing Economics of Payments). In 1980, banks had the BEST consumer behavior data (compared to other businesses). Today banks are still placed well to make consumer credit risk decisions, but have lost their consumer behavior data advantage to entities that touch the consumer more often (mobile handsets, mobile operators, google, social networks, …etc).
Beyond the behavior data deficiency there is a larger challenge for banks: partnership. Banks have always been instrumental in enabling commerce. Enablement requires the ability to partner within a trusted network among millions of participants (see Collaboration and the Sharing Economy). The value of data is at its intersection. Google, Facebook and Amazon have focused on “value orchestration”. Google will take a loss on the payment, ship goods to your house for free, all to gain the advertising spend (2000bps vs the 200 bps of payments). Banks, retailers, mobile operators all want to have a role in expanding commerce. As I outlined in Transformation of Commercial Networks: Unlocking $2T in Value, today data is flowing to the entities best able to create value (and great consumer experiences).
Facebook in particular has done a fabulous job here, with the Facebook Exchange and the recast Facebook Audience Network creating a model for other data companies to play. Banks have certainly learned that the power of a network (V/MA) can drive collaborative investment where they all win. The challenge for any single bank going it alone is that their data is not unique enough, nor do they scale (see Data Leakage).
Last week we saw Amex launch Amex Advance with Acxiom. Amex is 3-5 years ahead of most financial instutions in data … and Acxiom is a clear leader in providing Financial Services Institution (FSI) data services coupled with deep understanding of privacy and data security.
Acxiom has managed Citi and Amex’s marketing data for over 20 years, and last year won Citi’s Innovation Partner Award.
Amex has great data– they have both the consumer agreement and the merchant agreement, and thus can deliver services based upon their rights or permissions granted by either (as both issuer and acquirer). The challenge for Amex has been that that their consumer footprint is narrow and very focused on affluent and their transaction footprint is skewed toward travel and entertainment. Thus, few marketers will waste their time working to secure access to a data set that represents less than 10% of their marketing needs. Amex and Acxiom created an innovative solution to this problem by allowing external parties to build models from Amex and other 1st party data sets (example demographics, or merchant onboarded purchase data). There is a great detailed discussion of Amex Advance in this LiveRamp blog by Marc Ginsberg – VP and GM of Amex Advance.
To be clear, most payment based “targeting” today is based upon statistical models. As described by this AdAge Article “ Amex data models don’t allow for direct targeting of its card holders”. This modelling approach allows their data to play far beyond the scope of their footprint. For example, if you (as a merchant) could onboard your POS data with Amex and look for common characteristics/behaviors of your loyal customers (example over 40 with high travel). By combining with other behavioral data sets (like TV viewing or web site visits) you can more finely identify target segments. Axciom and Amex have done a very good job leveraging their ownership advantage to enable smaller data sets to scale to larger use in a way that allows others to participate.
The Challenges of Modeling
While modeling expands coverage of a narrow data set and may protect against individual consumer “lists” being sent, they are very much based on individual consumer behavior. The ability to identify an individual is the core threshold for consumer privacy, not whether PII (name, address, SSN, …) is involved. Individuals can be identified by their behavior (what they do).. and their attributes. For example I have no need to exchange PII or anonymized identifiers if I can precisely describe 40-50 characteristics of a person.
My last company 41st Parameter took this approach in the fraud space, as we recognized that identifying 41 parameters of a computing device could uniquely identify it among billions of devices. The same is true for consumer information. For example, I could be identified by a description of my behavior and demographics: Over 50, travels > 10 mo, Lives in NC, plays golf, eats too much McDonald’s, ATT phone, banks with Citi, buys Great Grains cereal, drives a Nissan, favorite website of weatherchannel, uses twitter, …
Additionally, most statisticians would agree that once a tuned model is created to identify attributes of a consumer toward a given objective (or state) there is no longer an need for the data (upon which the model was derived). In other words, effective models degrade the value of the underlying data, by enabling correlated behaviors to serve as a proxy for the target behavior.
For example, consumers that buys sun tan lotion and spends time on tripadvisor.com are likely to be planning a trip to someplace sunny.
Few consumers may have issues with this type of analysis, but what if correlations could be established to pregnancy, cancer, divorce, or infidelity? Allowing payment data to be co-mingled with TV, web behavior and SKU level behavior presents risks in use and has the potential do degrade the value of payment data overall.
Controlled sharing of anonymized IDs is actually more protective of consumers, and more relevant to marketers. If someone “steals” anonymized IDs they must know the context in which they are to be used (travel). Data owners are similarly in control of the construction of IDs and WHO to share it with (based upon the a unique anonymized ID that only the counterparty can translate). Acxiom’s LiveRamp and Commerce Signals are focused on this approach with top retailers.
For FSIs looking to chase this Amex success, there are 3 primary challenges
- Business/Economic. In areas where Consumer behavior doesn’t change often, once models are built (and tuned), the underlying data is no longer needed. Thus who owns the “models”, particularly when they are based upon collaborative data sets?
- Consumer privacy responsibilities are not fulfilled only by modeling. Is there a possibility that your data will play with other data in an uncontrolled fashion? In this famous Forbes article on the Father who found out his Daughter was pregnant through a a flyer featuring smiling infants, clothes and cribs after her purchases at Target of things like those Target found were purchased by pregnant customers. Modelling vs “lists” is a distinction without a difference, in both cases individual consumer information is being used to create a marketing “action”.
- Control of Use. Knowledge and control over use of data are key for privacy, compliance and creating sustainable value. In 4 party payment schemes (ie V/MA), payment data can have multiple owners.. each with both rights and responsibilities. Each responsible party of data sharing should seek to understand where the data they send out is going, what other data it is being combined with, and how the data is being aggregated, used, and sold in other combinations and models. Models can unintentionally leak consumer insights into the market without FSI control.
There are a number of firms that collect data from banks and other sources, find ways to match bank data with other data, and then sell very detailed insights from these models. No reader of this blog—which I write for “payment and data geeks”—would believe these mix-and-match data firms sell individual data like name, address, etc. These firms wouldn’t sell it—if for no other reason that they couldn’t resell it over and over if they lost their control over the data. These firms understand the value of banks data and insights in their models very well (and profitably).
Just as an example of one of these firms, here are a few quotes made by the heads of a firm at an Investor Risk Day presentation:
“One of the coolest things about this vast amount of…data… . So all these bank’s data that we are getting— they were anonymous, and yet we have found a very cool way to link them all up and down a consumer’s wallet.”
At the same meeting in a Question and Answer session involving the business heads:
Q: “I’m sorry if I’m missing something basic, but given that your data is depersonalized, how do you manage to derive wallet share?”
A: “Oh, it’s the coolest thing. And it’s not going to make you feel any better about the NSA.”
“The coolest ones are when you are actually seeing transactions going from one account to settling to another one. … These are the father paying off the daughter’s bills—those kinds of linkages allow us to go from an individual to a household, et cetera, et cetera.”
The takeaway: if FSIs don’t know and control who has access to their data and multiple uses that firms can make of it, other firms can take great pride in taking the data given to them and linking it with other sources and tracking individual behavior—like father/daughter interactions—in a way they compare to the NSA, but “cool.”
These models don’t give PII, such as name and address of the particular father and daughter. But proprietary insights based on data points “were anonymous”—but presumably are no more—comes in part from data banks give them. Banks need to know where their data is going, what use is being made, who is getting detailed insights based in part on their data, and whether—if they think that all this is OK—they want a share of the profits and a say over how their data is used and the customers. At least one major bank—JPMC—has decided it wants more control and protection for the data on its customer, and has take steps to restrict access and control.
Take Away – Data Democratization/Data Collaboration
There are 2 likely scenarios for financial institutions (and investors) to assess.
- Growing data intensity. Google/FB/Amazon win it all
- The Democratization of Data. A world where smaller data sets collaborate to enable value. My chief scientist tells me that 80-90% of all our behavior is local. Imagine a scheme that can bring together consumers (directly) and these local data sets effectively. (see Small Wins)
Members of payment networks all want to create a data business, one that is sustainable and allows financial participants to guide the use of data toward value propositions, and incent participation (consumer, merchant, 3rd parties). Logically, control of use is not something existing data providers are keen to embrace as their existing models would need ownership defined and “permission” granted. The corollary here is that no one will have a data business in a world of rampant leakage, or consumer data backlash.
The technology of sharing data is straightforward, the economic model and network to incent economic participation is not. The beauty of Visa and Mastercard is that millions of businesses invest billions of dollars to make it work. The top issue (IMHO) is that operating rules are hard to change (because of network size). My company is of course looking to play a role in helping enable control of use for all data owners, our model focuses on enabling bi-lateral rules and standards by which controlled data (signals) sharing can occur.
Top 5 Bank
… our data is valuable but is must play in 1000s of different models… What we should have learned is lessons from our past successes. We are the Bank that created Visa, and even though I compete with [others], a common network allowed us to invest in a way where we all win. Visa pointed us to you and after some Due Diligence we believe Commerce Signals has the right model, we want to work with you.. “
In my next blog I’ll describe what I believe financial institution collaboration looks like in a new data platform where control of use matters.
Previous Post on this topic: Payments Data and Google Attribution