Predicting the future value of your customers

Did you know that by the time you on-board a customer you will be able to accurately predict their future value?


From which it becomes clear that the decisions you make at the customer recruitment stage will determine that key metric of future customer value.

These are typically decisions about the recruitment channel and tactic you use, the types of customer you are targeting, and the nature of the proposition you are making to them.

Now you should also take into consideration the fact that a large number of customers that businesses recruit will yield negative value after the costs of recruitment have been deducted, whilst others will be strongly positive.

So, what as a marketer, should you do about this? First stop focusing on cost per acquisition; it’s the wrong metric to be guided by.

Next, we suggest that you should take a multivariate approach so that you consider all the factors together that define customer groups in order to focus on the ones that will bring you value.

We have recently been helping a substantial life insurance broker do this and the result has been transformational; for instance, they can now balance factors like the risk of attrition against the amount of monthly premium paid. They have found that a substantial number of their marketing tactics are yielding negative customer value and have had to be abandoned.

If you would like to discuss how to target your customer recruitment for longer term value then please contact us.

If you would like to read a case study about how we did it then please read our post:

 


UniFida logo

UniFida is the trading name of Marketing Planning Services Ltd, a London based technology and data science company set up in 2014. Our overall aim is to help organisations build more customer value at less marketing cost.

Our technology focus has been to develop UniFida. Our data science business comes both from existing users of UniFida, and from clients looking to us to solve their more complex data related marketing questions.

Marketing is changing at an explosive speed, and our ambition is to help our clients stay empowered and ahead in this challenging environment.


Solving the challenge of allocating recruitment budgets

Targeting Longer Term Customer Value

Using data science to target multi-channel marketing recruitment spend towards longer term customer value

This post will explain why targeting your marketing spend towards longer term customer value is not a one-dimensional problem. For instance, a group that may look bad by one criteria like retention may within it have sub groups that are of high value, and others that are the reverse.

Background

This case study has been derived from working with a substantial life insurance broker who has been using multiple channels to recruit customers in the UK.

We were asked to look at how best to target customer value as judged by the commission returned in the first 24 months from recruitment.

To give some insight into the kinds of issues we were presented with, lapsing early (i.e. in the first 24 months since inception) was one of the main causes of loss of customer value.

Lapsing early was found more amongst people recruited via outbound direct marketing channels than through more self-driven channels such as inbound to web. It was also correlated with the sum assured and the monthly premium. However direct recruiting is cheaper and higher premiums bring more commission.

The relationship between lapsing early and age was non-linear, with both younger and older people lapsing more than those in the middle.

As well as different recruitment channels having different costs per acquisition they also attract different age groups and sums assured.

Approach

So with so many factors at work how did we set about finding the best marketing tactics to deliver the highest return on recruitment marketing spend?

  1. We discovered that lapses in the first six months are very reliable predictors of the propensity to lapse within 24 months for every customer group (we got an R² of 0.99 when fitting our predicted curve to the observed data); this meant that we did not have to wait for 24months worth of history before making a prediction of lapse rates.
  2. The customer groups we were using were defined by a combination of age range, sum assured, and channel.
  3. We used a combination of known lapse experience with predicted to build the overall expectation not only of overall lapse rates but also when lapses would occur.
  4. The net customer value metric we used was the commission income up to 24 months after probability of lapse minus the cost of recruitment from the channel used.
  5. Hence for each customer group we could predict the net customer value; we also knew the historic number of policies being sold in each group and hence we could calculate their overall value.
  6. The net customer value by customer group ranged from -£1X to +£5X. This enabled our client to focus their marketing on areas where the net contribution was positive, customer numbers were substantial, and the required age groups could be targeted.

In conclusion

The focus for this client’s marketing turned away from broad brush channels like daytime TV towards direct marketing approaches. This switch made a very substantial difference to returns but with two caveats. First many direct marketing channels have a finite capacity, and second all recruitment channels suffer from saturation; in other words, the greater your spend in them, the worse the return.

However we strongly recommend that when looking at the effectiveness of different marketing tactics that you take a multivariate approach; that is allow the full effect of all factors on net customer value to be taken into account.


UniFida logo

UniFida is the trading name of Marketing Planning Services Ltd, a London based technology and data science company set up in 2014. Our overall aim is to help organisations build more customer value at less marketing cost.

Our technology focus has been to develop UniFida. Our data science business comes both from existing users of UniFida, and from clients looking to us to solve their more complex data related marketing questions.

Marketing is changing at an explosive speed, and our ambition is to help our clients stay empowered and ahead in this challenging environment.


Customer Value Analysis

Furniture Retailer Explores Customer Value


The Situation

An upmarket UK furniture retailer uses customer value analysis to determine whom to recruit and how best to cross-sell.

The Client’s Business Goals

Initially the client, who had historically mainly monitored store and product performance, just wanted to understand the longer term value of customers.

We proposed extending the brief to looking at the characteristics of high and low value customers, and also the typical customer journey in terms of second and subsequent purchases so as to inform crm activity.

Our Solution

  • Step one was to assemble customer level records from very detailed transactional history, and in particular to join together multiple transactions into one sale when they were closely linked in time
  • Step two was then to count longer term value by recruitment time period (the data going back almost ten years)
  • We then matched the customer data to external demographic and lifestyle data in order to see if there were any characteristics that differentiated high from low value customers, or whether that was influenced by the store they used to make their first order
  • Next we looked the time between first and any subsequent orders
  • And finally we looked at patterns in the sequence of purchases to see, for example for all purchasers of product category A, what were their most likely subsequent purchase categories

Key Benefits

  • An initial and surprising finding was that it was not possible to differentiate high from low value customers by their demographics or lifestyle; however there was a major scale difference between high and low value customers at first order, which was not challenged by subsequent orders. This enabled the client to focus from the start on cross selling and retaining those with initial high value purchases
  • Our next discovery was that nearly all second orders came in the first year since first order, except for a small number of people buying replacement covers for sofas and chairs after a few years. This implied that follow up CRM activity needed to be initiated immediately after first order, and that it was not productive after the first year had elapsed.
  • The third key benefit was that the analysis showed clearly what to cross-sell; we were able to find very distinct patterns in the types of product purchased at second order compared to first order.

UniFida logo

UniFida is the trading name of Marketing Planning Services Ltd, a London based technology and data science company set up in 2014. Our overall aim is to help organisations build more customer value at less marketing cost.

Our technology focus has been to develop UniFida. Our data science business comes both from existing users of UniFida, and from clients looking to us to solve their more complex data related marketing questions.

Marketing is changing at an explosive speed, and our ambition is to help our clients stay empowered and ahead in this challenging environment.


What can possibly go wrong when I allocate my marketing budget according to campaign ROI?

Well, depending on exactly how you do that, quite a lot actually!

Let me give three examples:

1. If you approve any campaign expected to provide an ROI > X you may be approving campaigns for which there may be alternatives that yield better X+++ returns. These may be different types of campaign with the same objective, or for instance the same campaign but just run at a different time of year.

2. It’s well known that all channels can suffer from over-use or saturation, at which point their ROI will start to drop. Just think of the irritation caused by a TV advert repeating too often. But if you cannot tell what the cumulative ROI of all your campaigns in a channel is, how are you going to be able to measure this effect, and know when to reduce spend in one channel or increase it in another?

3. Campaigns don’t just happen in isolation; they happen within the context of all the other campaigns that are happening at the same time, each of which may have an impact on a potential customer. Hence you spend on DRTV will impact your returns from direct mail or door-drops. Understanding the impact of your main channels on other channels can become critical.

So how to resolve these factors, and get to the point where your marketing budget allocation takes account of them?

We have developed a process, and related technology, that is designed to help you get to this point. The way it works in detail varies with each client, because no two clients share the same marketing campaign mix, but the outcome will be that you have a marketing budget that should get you the best possible return.

Contact us for more information and a no-obligation chat.


UniFida logo

UniFida is the trading name of Marketing Planning Services Ltd, a London based technology and data science company set up in 2014. Our overall aim is to help organisations build more customer value at less marketing cost.

Our technology focus has been to develop UniFida. Our data science business comes both from existing users of UniFida, and from clients looking to us to solve their more complex data related marketing questions.

Marketing is changing at an explosive speed, and our ambition is to help our clients stay empowered and ahead in this challenging environment.


If you can lend us your customers’ data, we can tell you the truth about them!

Increasing Customer Knowledge

Estimates differ, but apart from the largest companies who all have their own data scientists, and the smaller SMEs who are not into customer analysis, around 60% of those in the middle do not have the customer knowledge they would like to have to run their businesses.

This does not mean that you don’t have answers to some of your queries, but that you probably don’t always have the ability to ask detailed questions about your customers’ behaviour and get correct answers back.

Every organisation has their own very focused areas where they would like to know more about their customers, and we would like to hear what yours are.

To give you a flavour of some of the questions our data scientists have been asked recently to investigate:

  • how do the different channels, and means, by which we recruit customer impact their longer-term customer value?
  • once customers have made their first purchase what is their next most likely purchase, and what is the likely interval between first and second purchase?
  • what proportion of my customers am I retaining year on year, and what appears to be the factors causing better and worse retention?
  • can I score up my customers with their likelihood to respond to my next campaign?
  • what is the relationship between people browsing my website on multiple occasions, and purchasing?

In reality we never know exactly what we are going to be asked to look into next!

But we want to remove the barriers between your needs for customer knowledge, and our capability to deliver it.

The simplest way is for us to meet up. We can discuss a brief together, take it away, and come back to you with a fixed cost proposal. If it’s too high you can say no, and all you have lost is the time for the briefing meeting. If it’s ok you will get the customer knowledge you need at a reasonable price, delivered on time.

Members of our team have been delivering customer knowledge to clients for over 20 years, so we have built up some experience in the area.

And by using us you avoid the problems and costs of recruitment, when you may not need someone full-time to do the work for you.

Interested? Please email our Head of Data Science Anthony Antoniou to arrange an appointment.

Email: anthony@unifida.co.uk

Do you throw away good customers like households waste good food?

Official estimates for UK household food waste in 2015 come to 7.3m tonnes. Of this the avoidable element, if not wasted, would have saved the average household £480 pa.

Yet we have discovered that marketers who allow themselves to be governed by the use of RFM (recency, frequency, monetary value) as their criteria for selecting which customers to engage with are allowing themselves comparable customer wastage.

RFM has been recognized for many years as a great way to select customers for marketing communications, but it has a significant flaw!

People who still have a relationship with your company get ignored if they are not continually buying from you.

We routinely find customers on the bottom rungs of the RFM ladder that can still be profitably communicated with if you take a different approach.

In a recent case of a home shopping company we found customers who could have provided an extra £1m of net demand being excluded by RFM from receiving catalogues.

To find the valuable individual customers in your older RFM segments we build a multi-variate purchase value propensity model. This takes account of a wider range of variables than RFM, such as product categories, and will combine data from individual transactions in many more complex ways than RFM can do.

We then build a matrix that overlay RFM segments with propensity bands.

Propensity matrix diagram

In the top right hand corner of this matrix we have people who would have been forgotten by RFM but who in fact still have considerable demand value according to the propensity model.

What would you give to know how to stop recruiting unprofitable customers?

In many of our engagements with clients we discover that a high proportion of customers recruited are unprofitable, and for every one of these the company suffers not only their wasted sales and marketing expense, but also a downstream negative return.

For instance, life insurers may be concerned about claims risk, but often a greater cause of lost revenue is customer attrition. If they can recruit customers who stick, and pay higher premiums, this can transform their longer term profitability.

So why not move from a focus on recruitment volume, to one of recruitment value? It may mean lower customer volumes, but it will ensure much better overall profitability.

We want to share with you our five-step approach for stopping recruitment spend going on unprofitable customers;

1. Identify the factors that drive individual customer profitability.
These are going to vary from industry to industry, but they often boil down to purchase value, product margin, and product tenure. In simple terms if the customer buys from the bottom of your range an item that is expensive to deliver, and never repurchases, they are not going to be a prime recruitment target.

2. Analyse the recruitment process features that have an impact on these profitability drivers.
For instance different channels, and tactics within the channel, are going to drive very different types of customer. For one client we compared the longevity of customers recruited through their TV advertising and website (inbound traffic) with those recruited through outbound activities like telemarketing. The inbound traffic was very much stickier once recruited. But also, the types of customers you recruit will have a very strong impact; we like to look at differences in customer contribution by factors such as age, affluence, gender, geography etc.

3. Combining recruitment process features into a contribution calculation.
Once we understand those aspects of your recruitment process that are both controllable, and which have a significant impact on downstream customer contribution, we can then look at how in combination they can be used to identify and predict the future value of recruits.
For a recent client, we found that we had around 1000 combinations of factors that we could identify, that each had a significant number of recruits, and for which we could compute the average longer-term contribution. For instance a combination of factors could be age-band combined with recruitment channel, and particular product features.
For each combination, we could then analyse the average customer contribution provided. This provides the lens through which we can predict longer term customer contribution across all current recruitment activities.

4. Building a heat-map of where to spend the recruitment budget
By combining these individual customer segments into higher level groupings, we can depict go and no-go areas for targeting the recruitment budget.
This approach to targeting needs to be run in combination with an analysis of the cost to recruit by the different recruitment process features. The ideal is to find combinations of features with a low cost to recruit and a high customer value, but this may often yield very few available options. A pragmatic approach is to look at the ratio between downstream customer value and recruitment cost. Using the lens of recruitment process features this becomes very achievable.

5. Predicting forward value from your more recent recruitment activities
A final use of this analytical approach is to model the predicted downstream value of customer recruits in a particular time-period. To do this we allocate recruits in the period by their recruitment process features to a predicted contribution group.
The good news about this last step is that one can validate it on historic data.

We usually manage to get to a high level of accuracy of prediction of the drivers of customer contribution; our latest example was better than plus or minus 2%!