Customer analytics – a recap:
As we’ve seen in previous posts, there are five important elements to the process of marketing/customer analytics:
1. Start with the data: we need to ensure we have the right data, but we should make sure this is individual-level data – data held for each customer. Without it we can’t do effective customer analytics. So a big question for the company is: how to build the right infrastructure to collect that level of data, with that level of granularity?
2. Exploring the data: we shouldn’t start with big complex models, but look always to start simply with a basic exploration of the data. Even when we go on to use complex mathematical or statistical models, we still need to be able to understand the underlying data, to be able to sense check our explorations and predictions against that data which allows us confidence to assert, yes the underlying data does suggest/show that our modelled effects are real or reasonable.
3. Predicting the future: a huge benefit of data analytics is our ability to build predictive models which enable us to make predictions about the future, and so to help the firm try to deal with the inherent uncertainty in the marketplace and the competitive environment in which it’s operating. But prediction should be done ‘one customer at a time’, i.e. at the individual customer level. If we want to know how many people are going to buy a certain product across a particular geographic territory (i.e. looking at the aggregate for the marketplace), that’s not going to be driven at customer-level but can be predicted using higher-level data. But if we want to look at what each specific individual customer is going to do (or might do) next, given what they did before – or given what another very similar customer just did – then customer-level data will allow us to build models to predict that.
4. Optimising the future: to obtain the best results for the firm in terms of sales and profits, we are going to want to make tweaks and changes as we go along – such as pricing, discounting, campaign mailing (or emails), offers made, products sold (or bundled), and so on. Data analytics will help us to hone in on those changes we calculate will have the best likelihood of success and/or make the most significant impact in revenues or profits.
5. Decisions flowing into actions: we should ensure that the analytical process supports the firm in taking the right decisions, at the right time, and in the right way. There’s no point in the company expending a lot of time and effort (and possibly even money too) going through the previous four stages if the company is not then going to take the necessary business actions off the back of them.
So we start with the question or business problem that requires solving, and our end goal is to take some business action which attempts to rectify that problem. The elements of data analytics (collection, exploration, prediction, optimisation) are there to support the end goal of action-taking in order to achieve the firm’s required strategic, tactical or operational outcomes.
The 21st century, data-driven company:
In the modern business world, everyone at the highest levels in the business (c-suite, board of directors) needs to be able to understand customer analytics because they need to know how the company can make a profit from each and every one of their customers (meaning each of their customers individually, often referred to as the single customer view). Firms need to move their focus away from mass-marketing, mass-advertising, mass-selling – these techniques are increasingly less effective as society becomes more fragmented and as the firm’s potential customers become more stratified/tribalised and their attention becomes more diffused, making them harder to reach and even harder to engage.
Although the idea of moving to a single customer focus may seem like a tough ask, it shouldn’t be difficult given the technology which firms can have available to them today. That’s especially true given the massively reduced cost of that technology, making it readily and easily available to even the leanest and most boot-strapped of start-ups.
It’s all a matter of understanding the way business has moved, being willing to respond to that change, and above all being willing and ready to grasp the opportunity – necessity, even – to become the data-driven, customer-centric, 21st century business. Now is not the time to be technophobic, or tech-agnostic. Now is the time to dive in and embrace the new data paradigm.
Getting to optimal:
One distinction of the 21st century, data-driven company is that it uses technology to measure customers at the individual level, and to capture all relevant descriptive data about each one of those customers: where they came from, what they do (and don’t do), where they go to, what they say to their friends about the firm and its products, and so on. Optimally, this data collection should be conducted in real-time, or as close to it as possible, and across multiple platforms. Importantly, all that data from multiple sources/platforms needs to be brought together in one place which is easily searchable and viewable: the 21st century firm will be adept at solving the problem of data fusion.
But an even more important distinction is that it takes a different view of profitability too. Where the old-school view is one of products as profit-centres, the 21st century company sees the customer is the key profit-centre. Actually, the optimal company will take the integrated view, where both products and customers are seen as profit centres, but even here the ‘customer as profit-centre’ view takes precedence since, without customers, there will be no sales and no profits (no matter how wonderful the products are).
Contribution is the key profit measure – it flexes with sales activity level, and measures all direct profits generated from sales, before accounting for the firm’s fixed costs/overheads. In its most simple form, contribution can be thought of as turnover less costs (specifically, direct/variable costs of production, selling and fulfilment), but the modern company should understand how contribution is generated as a result of all the actions and behaviours of each and every single customer. In this way, the firm will be better placed to influence customers’ actions and behaviours in order to maximise contribution (and thus net operating profit).
The Contribution Cube:
To simplify things we can view contribution as a 3-dimensional cube, with customers on one plane (x), products on a second plane (y), and some other factor (usually location/geographical region) on the third plane (z). The easiest way to visualise this is to think of a Rubik’s Cube, with it’s 3x3x3 cubic structure. The factor of interest (in this case contribution, but could be anything ranging from simple counts of some descriptive data point – say, website visits – through to more complex data such as revenues or net profits) can be built up as layers in each of the three dimensions of the cube. The cube itself can also move along a timeline, which we can think of as the fourth dimension of time, t.
For example, for some specified period of time (week, month, year, etc.) total aggregate contribution will consist of the sum of all profits generated by each and every customer (x). The horizontal x-axis can be seen as consisting of multiple layers, with each one of these layers representing one customer who exists in one geographic location (z) and who may have purchased any number of different products (y). But total aggregate contribution will likewise consist of the sum of all profits generated by each and every product (y). The vertical y-axis consists of multiple layers, each representing one product. Products may be sold in many geographic locations (z) and may be purchased by any number of different customers (x). In this sense, contribution is an x-y table consisting of rows and columns representing how the purchases of various products by various customers makes up the aggregate total contribution. This can be extended further by exploded out backwards by some variable such as geographical territory, as well as extending out across the 4th dimension of time.
To optimise contribution (direct profits) the company should be looking at both the product and the customer to see how each individual one can be optimised such that, taken all together, total aggregate contributions will be maximised.
In terms of products, the company should be looking at how it can maximise product margin for each product – either by reducing costs of production, or by optimising the selling price, or perhaps even by reviewing which products are most successful/profitable in order to optimise the overall product range. Loss-making products (other than those created specifically as loss-leaders) should be earmarked and lessons learned so as to avoid loss-making products is future.
In terms of customers, the company should be looking at how it can maximise contribution by customer – either by maximising the volume sold to them (getting them to buy more stuff) or the amount sold to them (getting them to buy more expensive stuff), or by minimising the marketing costs to recruit/retain that customer or the costs to service that customer. Customers should also be reviewed in total to establish which groups/types of customers are most engaged/profitable in order to optimise the overall customer base. Loss-making customers (especially those which remain loss-making even after a specified lead-time: after year 0, after year 1, etc.) should be sacked and lessons learned to avoid recruiting similar loss-making customers in future.
Past into present:
Marketing and customer analytics has come a long way since the 1950’s. Back then consumer-facing firms sold through physical stores and the only customer data able to be collected was purchasing data (perhaps augmented by some demographic and survey data). Purchasing data was broken out by individual retail site or shop, allowing site-level decisions to be made about stock-holding levels, selling prices, seasonal and in-store discounts, and so on. Site-level data is still relevant today, since many business decisions will be made at this level. But the modern firm should be looking for far more granularity of data – all the way down to individual customer level wherever possible – and this should be the goal for data collection today. Then site-level analytics will still be possible, allowing for the required business decisions to be taken at site-level, but with the added advantage that customer-level data can be further analysed and leveraged.
Even in subsequent decades when direct response marketing really kicked in (late 1960’s and 1970’s), the data collected was not at individual customer level, although it did get much closer to it. Here the level of granularity is by campaign or by mail piece sent. An individual customer is tracked as part of a larger cohort, a group of like customers who are all acquired at the same time through the same means, or as part of a population, a group of customers who all receive the same mailing or promotion at the same time. Analysis of customers is performed on the wider group, and inferences made about individuals in the group – or others similar to them – by applying an average calculated for the whole group. (The assumption is that all individuals in that group all behave exactly the same, and exactly in line with the calculated average, which is fine up to a point, but averages can also be misleading so should be used with some caveats.)
This campaign-level data continues to be of relevance today, especially for those consumer-facing companies employing direct response marketing techniques (either on- or off-line, or more likely a combination of the two). It does have some degree of customer-level granularity since all mailings are tracked based on mail or media codes, so it is possible to see which mailings an individual customer received and responded to. From this data, mailings selections for future campaigns or mailings can be made which aim to enhance response rates based on better targeting of campaigns to customers based on their past behaviours/responses. It also allows a much wider range of company-specific factors to be tracked to see how individuals’ purchasing behaviours – and hence responses and sales – are affected by, for example:
- prices and discounts
- what kind of promotional or incentive offers were made
- the headlines or sales copy used
- frequency and timings of:
- campaign mailings (how often sent, which day received, whether reminder mailings sent or not)
- other forms of direct-response advertising (press adverts or flyer inserts in others’ publications)
- which kinds of mailings work best for different customers (flyers, mini-catalogues, full-size catalogues, postcards, sales letters)
- how products were promoted and presented in catalogues (half-page, single page, double-page spread)
- aesthetic elements of mailings (colours used, text fonts or sizes)
One of the great benefits of direct marketing methodologies is how easily it lends itself to split (A/B) testing of different elements of the marketing mix (for example, testing the price or the offer made). Here a mailing is split into two populations, one the control population and the other the test population. As long all test subjects are homogeneous yet randomly assigned across both populations, with the populations both being of sufficient size, and the test is conducted such that only one variable in the mailing changes across the two populations (with all other elements of the mailing staying the same), statistical analysis can be run on the results to establish how successful the changed element was in improving response rates.
In today’s marketplace, much of what was done in the past through a printed medium (such as mailshots and catalogue mailings) can now be done easily and cheaply online with emails and/or digital advertising (pay-per-click, banner ads). And with the added advantage of the automation of digital data collection directly into an order history database. All the variables likely to be tested under a paper-based direct-response business will still be relevant: how many emails to send and how frequently, which days and times are best to send them, what email subject headers work best, what sales copy works best. Emails can be tested to optimise for open rates and click-through rates; website landing pages can be optimised for visits, stickiness, conversion rates (specifically on landing pages linked from online ads or email copy), and so on.
Despite the stated positives, there are some deficiencies with these direct-response data sets. They allow the firm to know what a customer is spending on their products and how the customer is responding to their mailings (allowing the firm to optimise for those variables within the company’s control). But they cannot give the firm insights into the customer’s external purchases (what they’re spending on competitors’ same-category products, or on other different-category products). This external purchase data is crucial in being able to work out the customer lifetime value (CLV) of any given customer.
With the introduction of electronic point of sale (EPOS) barcode scanners in most retail sites in the 1980’s, along with the increasing prevalence of payment by debit card – both of which allow for the capture of significant amounts of self-identified consumer purchasing data (either through store loyalty cards or through debit cards) – the opportunity for individualised purchase-level data collection exploded. Use of EPOS data collection opens the firm up to a whole host of optimisation possibilities: special individualised offers to drive up-sells (encouraging purchase of better quality versions of a product the customer already buys) and cross-sells (encouraging purchase of a product which similar customers already buy), as well as tracking individual customers’ buying habits over a time.
With the advent of EPOS, it’s also possible for firms who aren’t able to collect this data (perhaps they’re a mail order catalogue retailer) to cross-reference their own customer data sets with these much larger consumer purchase and payment processing data sets. In this way even firms which don’t have EPOS data collection are able to unlock further customer insights.
We can see from the technological changes and advancements which have taken place over the last 6-7 decades that even the older kinds of data sets remain relevant and useful. Indeed, as new technological capabilities come on stream, so our ability to unlock more insights from older data types and data sets also advances. The moral of the story here is that all these old school data sets from the past are still relevant – don’t discard them, keep collecting them, and take them with you into the brave new world of real-time data analytics. It is the technology which will change, not the data itself – we should seek to augment our older data collections as we go forward, rather than replacing them outright.
The digital data age:
The internet era changed the face of marketing and customer analytics, even despite the fact that the vast majority of all consumer retail sales still happen through traditional channels not online (figures are currently around 7% in USA and 10% in UK; source). There are many advantages of data collection via digital means over traditional means. Firstly, with the right skills and technological tools, the data collection process itself can be easily automated. Second, the range of data which can be collected is huge, and still growing. In particular, firms can collect more data on individuals’ behaviours external to the firm itself, e.g. which competitors and alternate products they’re reviewing. Third, there’s far greater opportunity to target adverts and offers very specifically to just one individual, based on their previous behaviours, or based on some other very tightly-defined context (location, past purchase history). This latter technique may cookie tracking (based on the individual IP address) or other advanced digital advertising features, or it may employ firm-led initiatives such as loyalty programs or use of vouchers/promotion codes.
We saw above how site-level purchasing data and customer database data, when combined with (or cross-referenced to) wider consumer purchasing data has given firms greater insight into their customers. The same process is being applied with web data too. By fusing web/digital data with all the other layers of data the firm has or can access, even greater customer insights can be uncovered. While the stats above show the low overall levels of online purchasing compared to traditional purchasing, many customers are using the internet to browse and find products, and to research them in depth to help make a purchasing decision, before going to a traditional store to make the final purchase. Web data can be collected and analysed on customers’ web visits and reviews then cross-referenced to the eventual sale. A more broad analysis of product popularity can also be undertaken – how many people are searching for and researching about the product online as compared to how many are actually being sold in store, or vice versa. This gives a sense of the conversion level for that product – are customers researching and reviewing the product but ultimately not buying it? If not, the firm has a problem which needs to be addressed.
Applications of digital data sets in customer analytics:
Applications of customer analytics will always be led by the problem the firm wishes to address. Here are a few examples of the kinds of problems which can come up, and how digital data collection and analysis can help solve them. Wherever possible, firms should be seeking to maximise the data they can collect from these digital sources.
1. Advertising attribution: a key problem in online marketing is one of attribution, that is, which sale get attributed to which marketing or advertising activity. This is important since the firm wants to track how effective it’s various methods of marketing and advertising are, usually measured as return on investment (ROI). Current practice is to attribute the sale to the last click that was made by the customer just before the sale was logged.
In reality, the customer may have seen a number of ads over a period of a few days, all of which may have contributed to awareness of the product or offer. Or perhaps the customer had seen a banner ad recently over the last week which pushed awareness of the brand higher in the customer’s mind and it was this that triggered the eventual sale. So it’s really the entire path or sequence of customer behaviours and actions (and exposures to the company’s adverts and other promotional matter) which affects the sale – the problem is how to cost all of these to that eventual sale.
Digital data collection allows for full path data to be collected by individual customer, meaning the firm can get a much clearer picture of how effective all marketing efforts have been in achieving the sale – and so allowing more accurate attribution of sales and calculation of marketing ROI’s.
2. Channel optimisation: as customers become multi-channel in how they browse, interact with and buy from a company, a key issue becomes how does the firm optimise the customer’s purchase through the most valuable/profitable channel in order to maximise it’s overall profits. This might be driven either by price differentials between different channels (the firm might be able to charge a higher price for the same content on one channel compared to another) or by cost differentials (advertising via one channel may be more expensive than through another channel). A further aspect is, how many channels should the firm advertise and offer its products through, so as to maximise profit (i.e. so as not to cannibalise sales from a more profitable channel to a less profitable channel).
Although counter-intuitive, the evidence collected by marketing research shows that, by adding another channel such as mobile, some customers will switch to the new channel and be lost from the old channel but – over the longer-term – there is no overall cannibalisation of sales and in fact sales may even go up since new customers come in. Offering a new alternative channel through which customers can interact with and purchase from the firm is a positive move.
One of the great benefits of customer analytics in the new digital area is that data can be collected and analysed across multiple channels simultaneously. In this way the firm can better assess the optimally profitable mix of channels by which is should conduct its business.
3. Discounting (or word-of-mouth) attribution: where the firm sends a discount voucher to a customer to incentivise sales, it expects a certain uplift in sales as a result of that discount being taken up. But often the discount is framed such that the offer can be shared, and the customer is likely to share it among friends and family. As a result, the uplift in sales will occur not just against the individual customer but also against members of their wider social network too. If only the customer’s sales are attributed to that discount offer, the value to the firm of making the discount offer will be understated. The sales generated from the customer’s friends should also be allocated to the discount offer too.
Remember that a discount offer is a cost of marketing since it reduces the profit margin achieved on a given level of sales, but it is offered in the expectation that it will increase sales volumes (by converting those customers who are more price-sensitive). The firm wants to state accurately the value to the firm of the discount offer; to do so it needs to correctly allocate all increases in sales which came about due to that offer. With access to large social media / social networking data sets (and associated data from those sites such as feeds and shares), the firm can more easily track which friends the customer shared the offer with which subsequently resulted in additional sales. Marketing research has shown that, for any given discount offer, around 2/3rds of the sales uplift with that offer came directly from the original customer themselves while the remaining 1/3rd came from the friends and network they shared the offer with.
This is not just a good way of measuring the effects on sales and profits of word-of-mouth, sharing and recommendations, but it also helps to analyse individual customer value to the firm at a deeper level. The customer who is not just a regular purchaser from the firm, but who also regularly shares and encourages others to purchase, is even more valuable to the firm than that customer who is just a regular purchaser. Digital data analysis allows a firm to get the richer and deeper insights into their customers to allow for this level of segmentation by customer value.
4. Predicting purchasing: this is the seminal problem for all companies – how to predict who is most likely to buy so as to best target limited marketing or advertising resources to those most to buy. With no customer-level data the firm can never make predictions about who will buy (other than at the broadest levels based on population averages). With each additional layer of data that becomes available and which can be added into a customer overview, the more likely the firm is to be able to predict who will buy based on who saw which adverts. Online data only will be less useful than data about both online and television advert views. If you add in to this data on social media exposure (specifically, how many friends like the product), the greater the accuracy the firm can predict who will buy after viewing their ads. There is a meme bandied around – often by accountants trying to be smart – that Facebook likes don’t equal sales. Yet marketing researchers have shown that Facebook likes are a predictor of whether or not an ad-viewer will become a buyer, where that ad-viewer has also been exposed to likes.
The future is now:
Technology continues to advance and opens up more and more avenues of the kinds of data companies can collect about their customers e.g. mobile data (customers on the move), path data (what you did and where you did it), eye tracking (what you looked at and in what order). More data creates opportunities to develop more insights and generate more revenue- and profit-enhancing actions. This is the leading edge of data analytics and it is happening now. Here are just a few examples:
- Intentions data (what customers intended to buy, as compared to what they actually bought, to get a sense of how much purchasing is planned vs. how much is spontaneous/unplanned – purchase intentions can be assessed through surveys for offline stores or search-engine keywords/search terms for online stores)
- Path data (where did customers go when in the shop or on the website, what did they browse, what did they put in their baskets – in offline stores this can be tracked using RFID/radio frequency identification tags attached to shopping trolleys, more advanced technology for in-store tracking uses the customer’s mobile phone, if they have location tracking turned on, giving stores access via the phone companies of both the path data and the individual that data relates to – online website path tracking one just one website can be built in or accessed via Google Analytics services, or for wider internet path tracking cookies can be used to track individuals’ movements across the web)
- Field of vision data (where the customer is looking, how the eye is tracking across a shelf or around a web page, how many times a product is viewed/enters field of vision, and how this compares with which products are considered for purchase, and actually purchased – marketing research has collected data in-store using eye-cams on voluntary research subjects)
Real-world examples:
Advanced analytical techniques in use today exploit the large and varied new data sets now available – and leverage the ability to better fuse or integrate those different data sets from disperate sources – in order to drive marketing efforts which are both more pro-active and predictive, and better targeted or contextualised.
1. Smartphone targeting: advanced brick-and-mortar retailers can reach consumers who are inside their stores using geo-spatial location data available from phone companies for smart-phone users whose wifi is enabled (that’s mosst of us), coupled with the phone’s fixed IP address which is an individualised identifier. When the phone user enters a store, their phone picks up the store’s wifi signal, so the retailer knows they’re in-store right now. If that consumer has recently visited the retailer’s website, the two data sets can be linked so the retailer knows they’re a potential customer, and often what products they’re interested in too. The retailer can then target them with a real-time in-store discount offer by text (using technology to take advantage of the marketing adage of ‘right product, right place, right time’).
2. Content design: by using descriptive meta-tags any content being served to an audience online (be it written word, audio, video, images), a digital retailer or service provider can easily track the audience’s consumption of that content by data mining and analysing the meta-tags of past content consumed. This can be done both in aggregate across their whole userbase and right down to individual customer level. In this way a profile can be built up of the most popular (and therefore most commercially viable) content to produce/offer in future. The same method can also be empoyed to better segment the audience for more effective sales targeting in future, as well as for specific recommendations targeted to individual customers.
3. Churn modelling: this is a significant problem for all membership-, loyalty- or subscription-based marketing. How can you predict who will give up their membershipor subscription in the next period? As firms move towards some for of subscription model (at least in part) in order to help secure some recurring revenue, and so drive up customer lifetime value, churn becomes an increasingly significant risk metric. If large numbers of customers drop out of the subscription program and new customers are not recruited to replace them, the company’s customer base and revenue will begin to deteriorate, which would soon become unsustainable.
Marketing research has shown that social network data is a strong predictor of likelihood to remain in a subscription program or to drop out (to churn). Firms can scrape social media web data to gather more customer insight and intelligence and build that into their predictive churn models. Social web data can be processed using natural language techniques, which turn the text (words, likes, etc.) into numerical data which can be analysed.
4. Customer lifetime value monitoring: any company wishing to increase its revenue will need to (a) identify its best customers, i.e. the most profitable over the long-term; and (b) increase the stickiness or engagement and longevity of those customers. There are a whole host of innovative things a company can do to help drive customer engagement and longevity – both in terms of things it can do internally on its own, and in partnership with other companies (a growing area of opportunity). But this is all driven off having accurate and up-to-date lifetime value data for each and every one of its customers.
5. Customer ROI maximisation: while on the one hand a firm will want to engage its best customers, it will also want to try to change the behaviours or its less-good customers to make them more engaged and more profitable. Consistent use of customer-level data allows a company to track every behaviour and action of everyone interacting with the firm, and the profitability or ROI (return on investment) of every customer. It is then able to apply judicious use of selected offers to try to encourage customers from one group to another. For example, to encourage the less-good customers (the almost-boughts, the cart-abandoners, the one-purchase-wonders) from the less-engaged customer group to the more-engaged group, or to drive those in the neutrally-engaged, middle-ground customer group into the heavy-buyer group.
Each different customer grouping requires different treatment, different marketing, different offers – but the goal should to maximise the ROI of every individual customer, every time. It can do this by trying to move customers along a customer journey line from A (new contact/customer) to Z (absolute legend/advocate). Of course many/most customers will drop off at some point along the journey; predictive anaytics can help to predict who will drop off at which point. At each point on the path the firm’s goal should be to move that customer on by just one more notch – but always in the most profitable, cost-effective way possible (maximising ROI).
6. Customer service optimisation: for firms with telephone ordering or customer service helplines, it is possible to link the inbound telephone number with the customer account data and the customer ROI/LTV, meaning that whoever is calling into the firm, the call handler knows even before they answer the call who’s calling and how valuable a customer they are (key data will be length of custom, LTV, how many products purchased annually/over lifetime, average monthly/annual order value). This allows for call optimisation both for customer service (speed of call, issue-resolution) and order-placement (reduced friction, aids up-selling/cross-selling).
A further innovation is the inclusion of voice-monitoring software which (by processing natural language and intonation) detects sentiment/emotion levels within the caller and returns relevant key points about that customer (churn risk, etc.) along with suitable script for the call-handler to use to try to obtain the best outcome for the firm