I realized a little while ago that I may have more loyalty cards and memberships than the average person. (And that I more actively prove my loyalty than the average person). But as anybody who has ever signed up to a mailing list or for a store card knows, having a loyalty card doesn’t necessarily guarantee loyalty (unless you think of shopping as a sport). It just means that at one time we were enticed by a “shiny object” or a great consultant that deserves a raise.
There are statistics, statistics and statistics out there on loyalty but here are a few that total the rest:
- On average, loyal customers are worth up to ten times as much as their first purchase – White House Office of Consumer Affairs
- The probability of selling to an existing customer is 60 – 70%. The probability of selling to a new prospect is 5-20% – Marketing Metrics
- It costs six to seven times more to acquire a new customer than retain an existing one – Bain & Company
So yes, at the risk of seeming self-serving, it is worth keeping an avid shopper like myself happy.
In the first of this series I listed the things that would keep most consumers happy. I’ll cover the first two in this article. There are, of course, nuanced complexities across different markets and retail sectors (diamond rings versus ice cream versus wrapping paper), but simply put, customers want a good relationship with their retailers and service providers. If you don’t believe me, refer to the statistics.
Let’s get to the cashier at the end of the aisle – how do we use the data we have about our consumers, products and services and market to be successful? I refer to any entity that transacts as a customer e.g. individual, household, business, etc.
The right product – what product, when and for how long?
We know, even just by observation, that there are products and services that are most appropriate, or in demand, at either certain times of the year, or stage of a person’s life. For example, don’t try to sell me income insurance when I am about to retire, but feel free to sell me froyo all year long.
Understanding an individual customer’s stage of life is important when targeting them directly (more on this in the next section), but to understand “the right product” to have in stock, the trick is to know how demand changes throughout time. This is called demand sensing – understanding the troughs, spikes and plateaus of demand, throughout a year and over years.
We sense demand by using time series forecasting techniques that break demand down to:
- Trend – is demand increasing or decreasing on average?
- Seasonality – does demand generally peak or drop at certain times of the year?
- Known – do we have data to support a spike or trough like promotions, holidays, economics?
- Unknown (because we can’t know everything).
The most common time series forecasting techniques are ESM and ARIMA models. ARIMAs have an added advantage over ESMs of being able to take “known” factors (events, holidays and other information that may impact demand). These types of models are sometimes referred to as causal models because they can measure the impact, duration and complex interactions between trend, seasonality and known factors. For example, sweaters need to be discounted by 80% in summer unless recently worn by Taylor Swift on an album cover released within the last 6 months and/or exclusively sold online for 2 days to countries in the other hemisphere.
These models also allow us to understand the impact of different scenarios e.g. what if we run the promotion for 3 weeks instead of 2, what if we increase prices by 10% because of supplier shortage. Applying advanced forecasting techniques like these helped Nestlé Oceania more than halve their forecast bias, streamline their process and better understand the impact of their promotions.
Once we can sense demand we can then start to shape demand by optimizing across the supply chain. For more on demand forecasting follow The Business Forecasting Deal blog or in this title.
Good service - who to target, how to target and why target?
Ok, this is a given. Well-trained, customer-centric staff is precious gold, but knowing what trained staff should offer and when to make the offer is conflict-free customer service diamond.
Some types of data attributes that are useful for predicting the best treatment to give a customer:
- Transactional – purchases or inbound interactions made by the customer
- Behavioural – responses by the customer to interactions and patterns in transactions
- Geodemographic – statistical characteristics of a customer e.g. gender, region of residence
- Derived 3rd party – data accumulated over various sources providing pre-calculated metrics that can be applied to a customer.
If a customer always shops when they are given an offer (SALE!), they will probably continue that behaviour. However, as we know, people don’t all behave the same way, and trying to get a headline description of our "target customer" isn't easy.
One method to generalize customers is to group them into segments using data attributes. A common technique is Recency, Frequency and Monetary/Value segmentation (RFM). RFM uses data on how recently a customer transacted, how often a customer transacted in a period of time, and how much a customer has spent/cost in a period of time to split customers up across a cross-section of those dimensions e.g. low tenured but high value customers, high value lapsing customers, etc. Keeping customers (like me) that are high R, F and M happy will generally work out well.
If we want to take it to the next step, we need to learn from attributes of customers that have and have not responded to offers and channels in the past, create a picture (model) of how to differentiate them, then extrapolate the model to give us what is likely to occur in the future. This is called propensity modelling, a form of predictive modelling. Using statistically-based techniques rather than business rules, removes the need for us to pre-suppose every detail about every customer. These techniques are used to predict the likelihood of a customer taking up an offer, lapsing from a loyalty program or even the life stage and lifetime value of a customer. For larger data, this is best carried out using machine learning techniques in a data mining framework for automation and validation.
One step further is to optimally allocate an offer amongst competing offers and channels to a customer while accounting for operational constraints (budget, resources) and customer preferences (frequency of contact, other products of interest). New Zealand’s leading coalition loyalty program Fly Buys, uses a combination of these techniques to target customers with the best offers and maximise the return their partners gain from the program.
Finally, the ability to understand what customers are saying in surveys, through the call centre and on social media forums about our products, services and processes adds further power to predictions. Applying text analytics techniques that use a combination of machine learning and linguistic rules to extract sentiment, discover discussion topics and predict outcomes is how organisations like Lenovo decreased the number of call centre requests by 30%-50%. Marrying behaviour with feedback is the fundamental objective of good customer service and the basis of concepts like Net Promoter Score.
“A feeling that I got a good deal” and “Convenience” in II.B. Until then, if you want to go beyond a “shiny object” to a meaningful relationship with your customers, start an Analytics conversation within your organisation by sharing the case studies above and visiting SAS Advanced Analytics to get more information on the boutique techniques described. For those out for a quick shop, visit SAS Visual Analytics.
A Shopaholic’s Guide to Analytics II.A was published on Left of the Date Line.