When bankcard first arrived in October 1974, the choice to apply for one was relatively simple. Would you like to buy something now and pay for it later plus interest, of keep on paying for things with cash or cheque.
Today, there are many hundred credit cards with the CBA alone offering 9 types, each with different rates, fees and charges, terms, repayment options, rewards and benefits. Understanding how credit cards are chosen is no longer a one-dimensional problem.
The CBA The Strategy team had identified some gaps in their offering and were considering introducing yet another credit card intro an already crowded market. It was however getting complicated.
Was there room in the market? What if the credit card was unattractive and flopped? Worse still, what if it was so attractive that it cannibalised all the other cards and the bank lost money?
What was needed was a model so the team could prod and poke it, invent product scenarios and see what would happen to their market share, and bottom line, before launch.
The Size of the Problem
This was to be the largest choice model ever developed by SurveyEngine. 15,000 participants were recruited to provide the large amount of response data needed for modelling.
Over 80 factors were tested in parallel covering a vast 1018 unique combinations of viable credits cards, one for every grain of sand on the planet.
While the scale of the experiment was ambitious, it was manageable. The same could not be said of a looming problem with the client.
Client Problems – and Selling Fruit
To understand the problem,consider a fruit stand selling apples.
A fruit seller sells premium red apples for $2 and ordinary green apples for $1. He observes that people are buying twice as many green apples as red. Perhaps they prefer the green to red, or maybe they prefer red but don’t want to pay twice as much, he muses.
In fact he cannot infer anything at all, there is not enough data nor will there ever be from sales alone. For a definitive answer he needs to change the prices, even for just one day. If then twice as many green are still being sold, it is solely the colour that is making people choose. In order to settle the question he needs to test the implausible, even if he would never consider selling premium red apples for $1.
Precisely the same situation came up with the credit cards experiment. The experiment included some hypothetical combinations which were indeed implausible – platinum cards with no annual fees, student cards with concierge services and so on.
While there was some resistance to including “implausible combinations” since they would never be offered, the purpose of an ‘agnostic’ scientific approach was clarified and the project proceeded successfully.
For commercial reasons its not possible to reveal any specific results here, sufficed to say, the model directly informed retail credit card strategy for at least the following year.
The model let the bank …
- identify and abandon non-performing benefits. A large number of card benefits were tested including frequent flyer rewards, chip security, concierge services etc. The model was able to directly value, in dollar terms, what each benefit was worth to a customer. Or to put another way, what the corresponding compensation in dollars would need to be for each benefit if it was removed. This provided a simple value filter to the bank which put a razor to all the benefits programmes where the customer’s estimate of value was less than the cost to implement it.
- validate the viability of new cards. A number of new card concepts were included in the model. The model was able to show which cards and feature set were both viable and would take overall share from competitors while not cannibalising the bank offers.
- increase profits $8,000,000 p.a. per card. The model showed that a specific set of changes to interest-rate, annual fee and interest free days would improve profit without reducing market share. These changes were implemented and independently estimated to increase the profit per by approximately $8,000,000 AUD per year per card.