Every interaction between a brand, and a customer is a great insight. Collecting data from different sources i.e. online click, previous purchase, interaction with call centre or a store visit, helps to build great customer profile. When you marry this profile with algorithms, you end up having quite accurate prediction on how to target customer. Back in the ’90s, most interactions were face-to-face or by mail. Later on, online became a game changer in terms of marketing influence. Nowadays, with mobile devices, geolocation turned on on them, and social media networks, you can target customers in real-time, wherever they are, in the most personalised way by listening to what they do and what they say. There is a “but” – connecting all of the channels is quite difficult, but it definitely helps to understand and predict customer’s future behaviour.
What is the reason to use predictive intelligence/analytics? Answer to this question is very simple – it helps you to see who is most likely to buy, and who will buy more or more often.
I can see three solutions here:
Somehow powerful targeting is the one based on RFM model (Recency – How recently did the customer purchase? Frequency – How often do they purchase? Monetary Value – How much do they spend?). This modeling is based purely on customer behaviour from the past. It may work better in some sectors and worse in others i.e. this may work brilliantly in travel sector or for mobile networks, because people tend to book holidays/take new phone contracts regularly in the same intervals, while in consumer electronics customers tend to buy impulsively.
RFM method is descriptive (describing characteristics of a population), and does not forecast behaviour – it assumes that customers will continue behaving the same – it does not take into account the impact of lifestage or lifecycle transitions on the likelihood of response.
When used as the primary targeting method, it may lead to over emailing to the most attractive RFM segments and to neglect of other segments that would be profitable if developed properly. Remember, history can’t always predict future, especially when it involves people.
This method answers two questions – What happened? And why did it happen?
Predictive modelling might work better. It’s more accurate than RFM and it’s based on post-purchase cross-sell. Timing is very important here, so robust testing plan has to be developed.
You never know when it is the best time to cross-sell more products to your customers. In the travel sector timings may be stretched more compared with the consumer electronics sector. People tend to book holidays well in advance, so you may have 3, 6, 12 months, or even more, to send them email communications selling travel insurance, car park, hotel, excursions etc.
In consumer electronics you don’t have that luxury. Post-purchase cross-sell needs to happen shortly after the purchase i.e. if your customer hasn’t bought a TV stand or HDMI cable together with a TV you will need to try to sell it to this customer as soon as possible. There is a chance that customer may own a TV stand. If not, the customer is going to buy it on the same day or across next few days. At this point you may also exclude this customer from receiving any promotional emails. Test your timing against two control groups – one to receive all emails and one not receiving anything.
Using this method you may answer the following question – What will happen?
And the last but not least – use your real-time data. It falls under predictive modelling, but it is based on present, rather than past behaviours.
Every ESPs provide dynamic content and triggered deployment functionality, but the way that these features are used depend on the data structure and availability. In the past dynamic content was mainly driven by profile data or preferences. Nowadays, you can do much more using API calls and passing data to your ESP triggering emails in real time. If you don’t have SCV, you may fake it with multiple API calls passing data to your ESP where it will be stored. It isn’t easy, but it’s doable. There are several email campaigns that you can deploy instantly, caused by certain behaviours i.e. Welcome email, online browsing or abandoned basket. I touched on this when Dave Chaffey interviewed me in the article on email best practice.
Example of use of real-time data: it tend to happen that customer’s loved one has birthday, so he/she will buy, let’s say, digital camera as a gift, or customer’s washing machine broke down, and so it’s a distress purchase. In these two examples, online browsing data and timely deployed email my end customer journey with a purchase, before that customer goes to your competitor.
The biggest challenge
You may say: There is too much information. And you are absolutely right. For that reason there is a need to separate actionable items from the mass of data. Your goal is to aggregate all of the data, so you can predict and affect customers future actions to be able to understand their complex journeys.
There are many tool out there that will help you with your predictive intelligence endeavours, but I am going to mention three that I had a chance to get to know a bit more recently. Well, these are not little, but definitely powerful. Have a look at the links below:
These three tools will do similar things in similar way. Deciding which one to pick will be based on your personal preference, but all three are really good.
All of this is massive data mining exercise, and, when done correctly, will return the investment. So keep on trying and you will succeed. Good luck!