20 Aug The complicated art of Customer communication
The common marketing mantra is that we should offer the Right Product to the Right Customer at the Right Time.
So to do this we developed traditional Product based marketing in which there are three main steps; decide on the product, find the leads and communicate.
The product part was straightforward, but we all used statistical methods such as clustering, scoring and segmentation to find appropriate leads. These were then further tuned using statistical methods to determine response probability, etc.
This approach gave us right product and right customer but at the wrong time, and as history has shown us, this wasn’t a very effective way of marketing. Typical response rates for banking campaigns average around 1-3%, which in turn led to customer dissatisfaction and opt-out.
Over time, marketing vendors realised that the timing portion of the mantra was important and so marketing in recent years has changed to a more activity based approach in which we monitor a Customer’s activity to find leads and based on this we determine what product offer to communicate.
This was great. We had addressed the timing issue but now had the problem of determining what was the best thing to offer. This problem is further compounded by decisioning in real-time and the implementation across multiple channels.
If the problem is “I need to respond to my Customer in real-time”, and given that a Bank may have hundreds of thousands of ‘interactions’ per second then the only answer is automation. Which means we need to find a way to determine what communication to deliver in an automated way.
Enter the statisticians, who came up with approaches to decide on the Next Best Offer (NBO) or more recently Next Best Action (NBA) in real-time. The results of this statistical analysis are then used to formulate the appropriate communication.
So, in summary the use of statistics has moved from finding leads (where it wasn’t very successful) to determining the appropriate communication. And it really is the only way – if you absolutely have to respond in real-time and into unmanned channels.
UPDATE: (see my updated comments below)
The benefits are that communication is cheap, but the downside is that communications are only as accurate as your statistical approach.
The numbers in the figure above come from a couple of sources; presentations made by a couple of suppliers and from some information I was given about a major UK bank project.
The suppliers stated things like a 30% increase in sales or a 150% increase in accuracy of targeting. These sound like really worthwhile advantages until you realise that the UK bank NBA project cost £50m and that a 30% increase in sales means from 3% to 4%. More than this, it is estimated that payback based on these extra sales to be about 7 years. Hmmm.
But what if the communication can be done in hours rather than seconds? Or if the possible number of communications is high and we don’t know which one to use? Or if the stats are inaccurate in determining an appropriate offer?
Let’s look at a simple scenario.
Peter has €100k deposited into his account.
We can detect this, and we can determine that this is significant (for Peter). So what do we offer him?
Well this isn’t so easy. After all there can be as many as 17 different reasons why Peter received this money. For example: a windfall, Inheritance, insurance pay out, Bonus, tax refund, sale of assets, sale of a house, consolidation of assets, etc.
Similarly, there are many things that Peter may intend to do with this money, including; start a business, invest, buy a house, give as a wedding present to his daughter, put in a college fund, buy a Ferrari, etc.
And because there are many different things Peter could do, there can be many different offers from high interest savings, a mortgage, a pension, shares, a business loan, etc.
So which one would you offer?
Well statistics would say that you have to offer high interest savings. Statistically, it is your best bet, but this may not be the right answer. The accuracy may be <30% and once you have made this automated offer you can’t make another.
The problem is that this approach is treating you like a segment or group (people like Peter bought this) rather than as an individual, and as Sherlock Homes, says in the Sign of Four, “While the individual man is an insoluble puzzle, in the aggregate he becomes a mathematical certainty. You can, for example, never foretell what any one man will do, but you can say with precision what an average number will be up to. Individuals vary, but percentages remain constant. So says the statistician.”
In this case I suggest the answer is to use a human. Talk to the Customer and determine the real need. Sure it costs more, but the accuracy jumps from 10-35% up to 90%. Most importantly Customer acceptance (10x) and satisfaction (+10%) also increase.
So really the choice is, do I go for the cheap and inaccurate or the expensive and accurate? This is really the decision you have to make.
Now think about what your offer might be if the Customer has a new job, is getting married, moving house or having a child. As you can imagine there could be many possible offers – or even combinations of offers.
So I suggest there are two questions you need to ask when planning your communications
• Does this absolutely have to be in real-time?
• Is there a clear one-to-one link between the activity and an offer?
If the answer is Yes then you can automate your communication but if it isn’t, then you need to consider using a different approach (like calling the customer) or relying upon your statistical methods getting it more right than wrong.
Marketing may well be the Right Product to the Right Customer at the Right Time, but now we can determine the Right time, I believe it could be better phrased as “Detect the Change, Determine the Need, Answer the Need”.
Some of you will think that I am criticising Statisticians unfairly. This isn’t true. I have met and worked with many statisticians and I can say that 62.9% of them were nice guys – give or take a 3% sampling error.
Since I originally wrote this article the very recent (2017) rise in the availability of AI and machine-learning chatbots means that it will (soon) become possible to communicate in an automated way (without the use of humans) but with better accuracy than that offered by ‘statistical inference’. So it will be possible to hand over the Large Deposit Lead to a chatbot and ask them to discuss this with the customer to determine and address the appropriate need.
I will leave you with a German Joke.
Hans and Johann go off into the woods to do some hunting.
Hans comes into a clearing and sees a rabbit on the other side. Crouching down he gets his rifle out, aims and fires but he misses to the left.
Sheiße. So he takes a breath, aims and fires again but this time he misses to the right.
Hans then puts his rifle back into its case at which Johann says, “Why are you doing that?”
To which Hans replies, “Statistically ze rabbit is dead.”
PS. No Statisticians were harmed in the making of this article.