ONE-TO-ONE MARKETING AUTOMATION 2/3

As we have seen in our previous article, predictive approaches enable marketers to attribute scoring (or response prediction) to each campaign’s prospect. A means to dispense with generic segmentation logic to adaptive targeting for each campaign; and push the right offer at the right time to the right person. Several stakeholders position themselves on this field and offer solutions enabling precise targeting.

The emailing logic and the limits of predictive marketing

Despite the proven benefits of dispensing with generic segments and moving onto a one-to-one logic, we must not forget the main benefits of pre-defined segments and to make broadcasting logistic management simpler from an operational point of view. Indeed, email marketing is subject to strong constraints; some specific to that channel.

The first concern for marketers is to best manage the marketing pressure applied to the prospect base. If they broadcast few emails, they are under the profitability threshold of the base. But if they broadcast too many, then the risk is to wear out prospects leading to unsubscribes or complaints to MSPs (Mail Service Providers).

This has 2 adverse effects. Unsubscribes and complaints take the prospect out of the base. Known as attrition, this phenomenon is costly in the long run since it requires an increase in the scale of the acquisition strategy to renew the base. The second effect, and certainly the most concerning, is that a high volume of complaints will trigger MSP blockage. This could eventually result in blacklisting which will prohibit you from reaching the addresses of these MSPs. We can imagine the loss this would cause. Emailing professionals know too well this fine limit between the risk of pressuring the prospect base and maximising returns on a campaign.

Managing the marketing pressure

The logic of defined segments enables us to handle these aspects. Besides, the good practice of segmentation tends to include a “reactive” attribute (or not reactive) to the segment in order to quantify its tolerance to the marketing pressure. Thus, when broadcasting a campaign to a prospect segment, marketers know they must not solicit this particular segment for a certain period of time.

It is therefore possible to split broadcasts within segments in order to ensure that we meet the audience objectives of each campaign (from a qualitative and quantitative point of view), and still maintaining the optimum marketing pressure. On paper, it seems simple, but professionals know how difficult this can be. They only have their intuition and their experience to decide how to organise their campaigns. It is manageable but it quickly becomes a full time job for the marketer who then spends most of his time on campaigns logistics; far from creativity and innovation which are the core of his trade.

We can see that the introduction of predictive marketing might increase the complexity of logistics, even if it increases the real potential of the base. With predictive marketing, fixed segments no longer exist through time. To get the most out of predictive marketing, we must break the segmentation logic. But if we break this logic, it becomes impossible to manage the marketing pressure. How to reconcile the two without falling into a segmentation logic disguised as a predictive approach? Is it even possible?

Moving from predictive to prescriptive marketing

In one word: yes. But it requires to call for a branch of mathematics that is far from predictive approaches. If we need algorithms to predict the effects of an action (i.e. broadcasting such email to this prospect has x% of chances to generate an open/click), we need other algorithms to prescribe the actions to run (i.e. what to broadcast to whom and when). Therefore, predictive marketing isn’t the end purpose of the mutation that occurs under the term “marketing automation”. It is the foundation that will support prescriptive marketing.

Mathematics behind prescriptive marketing carry the sweet name of “Operational Research”, initially created for military purposes while planning WWII’s D-day. As we guess, it has something to do with major logistical issues, that the emergence of computers could approach using algorithms capable of rationalising the process. Since then, the intervention’s spectrum of the discipline has made it through to the civil domain, mainly in heavy industries where planning and scheduling problematics are numerous (scheduling when and who, planning what and how…) and more recently, the marketing processes.

Operational Research: algorithms at the service of marketing

Operational Research consists in modelling the decision issues we are faced with and to conceive algorithms to immediately resolve and optimise them. It is the step that follows predictive modelling in the automated decision process.

The strength of that approach is to model the problem in its globality; mathematically describing the decision to make, the KPI to optimise and the constraints to respect. For instance, to implement a retail chain, the decision would be to determine the number of points of sale to open. The KPI would be the coverage of the global network. The constraint would be not to blow the budget and avoid cannibalisation between shops.

Prescriptive marketing applied to email campaigns

As we have seen, predictive marketing tends to consider each individual as a separate segment. We end up with thousands or millions of segments to which we need to broadcast communications amongst a given list of campaigns. Each campaign has its own objectives (a minimum of emails to broadcast, over a certain period); some with various remuneration methods for third party campaigns.

Moreover, MSPs have their own blockage policies. Thus, using predictive marketing, we can predict what will be the probability for each individual to respond to a specific campaign. We can also predict his/her propensity to complaint and unsubscribe if the marketing pressure is too strong or if we broadcast mismatching offers. How to daily decide what campaign to broadcast? to which individuals? Taking into account specifics and still guaranteeing that our objectives for each campaign are attained; and that the global efficiency of the broadcasting planning is maximised?

This is when prescriptive marketing kicks in. Operational Research algorithms will utilise predictions to make the best decision possible while simultaneously respecting all the constraints (those can be on each campaigns objectives, MSP policies, maximal marketing budget…). To model the planning efficiency, each action must be tied to a cost and a gain. For instance, cost can be the unsubscribing potential of the target as well as the cost of the email itself. The gain is the remuneration we get from a click or a transformation. The algorithm will find, amongst the numerous broadcasting planning, the one that maximises the global ROI taking all campaigns into account.

The benefit of this approach is triple, and addresses all issues left open by predictive marketing:

- The logistical complexity is left to the algorithms that guaranty all constraints are answered
- The algorithms make the best of the power given by predictive marketing
- The planning is automated

What to expect at the output of the process? A global planning, where each individual receives completely personalised emails, respecting marketing pressure’s tolerance, and where each campaign fulfils the objective without violating the constraints imposed by MSPs. Moreover, we know in advance the gains that planning will generate since they have been maximised by the algorithms. The first experiments show impressive gains compared to only predictive marketing approaches, where 30% looks like a minimum.

To conclude

It is less probable that predictive marketing will reach its full potential without the need for prescriptive marketing; and this will happen to the detriment of human intervention on campaign management. Campaign management will be delegated to Artificial intelligence modules conceived by Operational Research. Gains in terms of ROI deserve to consider this point, especially since these algorithms are much more an opportunity and a support for marketing than a threat. They are within the scope of assisting the decision making process; no more, no less. They will achieve better and faster what human beings cannot do themselves.

Therefore, we must not forget that the quality of the message, the visual, the global marketing strategy will always be the key to success. The role of the algorithm will be to best exploit the marketing creativity, to sublime it while freeing it from the logistical limits it is confronted with. In other terms, it will enable marketing to concentrate the focus on its core business, while algorithms will take care of the headaches.

Moving from predictive to prescriptive marketing is therefore possible for emailing, but what about the other channels? After all, tomorrow’s marketing automation seems to be multi-channel and brands will want to preserve their clients on all supports. The fact is that technological expansion, also called programmatic, makes it even more possible. This will be the topic of our third and last article: “Moving from predictive to prescriptive marketing: tomorrow’s marketing automation”.

First published on the 12th of January 2016
This article is the second part of a series of 3 articles
Read Part 3

Cédric Hervet, Operational Research Director at SOCIO (NP6 Group)

Part1: Predictive marketing within emailing is the milestone to one-to-one marketing