The optimism of #Purpose & science of #design in forging #digital customer experiences! Opinions my own!

Marketing optimisation is an attractive concept: use predictive modelling and business rules to work out exactly which combinations of contacts, products and offers the direct marketing budget should be spent on. But optimisation tools tend to be costly and have historically been applied to improve the performance of high-volume direct mailing. With postal mail far less prominent today, are these tools still relevant and capable of improving today’s broad mix of offline and online channels?

Helping to plan

“If you can measure the outcome, then all channels can be optimised – online or offline. Typical optimisation solutions are designed to operate in a single channel marketing environment, and in general are able to deliver what is asked of them. However, once you introduce other channels, most solutions are not as qualified.

Goals have changed for today’s enterprise marketing organizations. While once measured by the sheer amount of marketing activity they were generating, these groups are now expected to be profit centers, executing only thoseactivities that will truly maximize the profit returned to the corporation. Marketing departments, in general, are not adequately prepared to function in this profit driven environment. The size of the potential set of offer-customer-channel combinations that they could execute against is mind-boggling. How is a marketer to decide what the optimal subset of combinations is that will maximize profit?

Marketing optimisation takes the principles of target marketing – right customer, right offer, right time – and applies them to all a company’s customers, products and campaigns to find the optimum outbound mix of contacts and offers. Rather than finding the best individual for a predetermined campaign – the conventional method – optimisation tools work out how a particular customer (or perhaps segment) should be approached: via which channels, how often and with which offers. The critical part is that the software does this within specified constraints, be that budget, channel capacity, maximum number of contacts per customer per month and so forth.

Each set of constraints forms a “scenario” and companies typically work through and rerun various scenarios until they have a solution that suits. Optimising solely for profit might suggest tripling the marketing budget to achieve the maximum possible returns while optimising within a fixed budget and perhaps setting a maximum response volume (for example at a call centre) would produce a completely different, smaller set of suggested contacts. On the basis of the results, marketers can then plan out their year’s campaigns and allocate budget roughly as required.

To come up with these figures, complex business rules and multiple propensity models are typically combined by custom software to predict how each customer or segment might respond, and so produce an optimum strategy set within defined constraints. Much other historical data is required, such as how response through a channel varies as outbound volume increases. It tends to be a technique that only companies with a clean, well-populated database and a good store of historical response information can adopt.

Historically, optimisation has been used with direct channels like the telephone and direct mail. With the rise of email and the web, the situation looks somewhat different. Though outbound email is very suitable for optimisation, its low cost delivery takes away much of direct mail’s justification for investment in optimisation, that is, reducing mailing costs while maintaining and increasing response. However there is a strong case for predictive modelling for email to avoid overcontact as well as to select the most attractive offers and creative treatments.

Attributing response or sales uplift is essential to make use of optimisation and doing this for broadcast media such as TV and radio or channels such as field sales has always involved a certain amount of educated guesswork. The recently-arrived social media channels are similar: it’s tough enough to measure ROI on them directly, let alone predict what ROI will be and therefore what budget should be allocated to them in the year ahead. Even in direct channels, it’s often impossible to know which of the seven contacts a customer experienced was the one that made him or her go out and buy the product. Was it the mailing, the emails or the final phone call? It’s increasingly difficult to resolve and spot response, particularly when it’s a high value purchase and we might be basing our optimisation on faulty response data.

Without some accuracy in attribution, cross-channel optimisation starts to become dangerously inaccurate. Accuracy can only come from populating and linking response data from all channels at individual level on a single-customer-view database. Not a trivial task.

The speed and interactivity of web and email also differentiates them from conventional channels and, as the web gives access to unheard-of levels of product information, user-generated reviews, price comparisons and so on, buyers are increasingly taking control of the purchase process themselves rather than simply responding to offers.

The biggest trend now is to react to consumers rather than to predict their behaviour and send outbound messages. The marketing programmes of today understand the mechanisms in the customer journey and set up systems to respond quickly and appropriately rather than organising complex outbound campaigns; something we call Next Best Action or Synergy Marketing.

We can no longer predict accurately months in advance what we are going to do, however that historical data is still essential to build understanding of customer behaviour. We still rely on past data to predict the future but wehave to monitor and optimise as we go along.

Linear programming

There are multiple ways to perform Marketing Optimization and decide on investment allocation. Linear Programming is a well known methodology in the field of operational management. It was invented in 1939 and heavily utilized during the WWII for military logistics.

Linear Programming solves optimization problems under a set of linear constraints. Typical business uses involve work scheduling, inventory routing and workload assignments. In mathematical terms Linear Programming is expressed as:

• Maximize the objective function: c T x=∑ i c i x i
• subject to a set of linear constraints: Ax≤b
• and x i ≥0  for each i

The objective function is a simple linear combination and is therefore ideal for portfolio optimization problems. Variables c i   can represent individual Returns-of-Investments (ROI) and x  would stand for the amount invested in each asset. The set of linear constraints is encoded as rows of the A  matrix.

To put it simply:

• The set of constraints (matrix A  and vector b  ) define faces of a high-dimensional “solid” and we are looking for a point x  inside that solid where the objective function is largest.
• The maximal point can only be at one of the vertices of the solid.
• We start with any vertex and move along the edges in the direction of a growing objective function.
• When we reach a vertex where all edges from that point lead to vertices with lower value of the objective function, the optimal solution was found.

The Simplex Algorithm has a very nice property – we only need to check a finite number of points at the vertices of the solid.

Inbound or outbound?

Optimising multichannel outbound campaigns is certainly possible but its ROI will depend greatly on the mix of different channels concerned: the more predictable the response and income, the more useful optimisation techniques will be. Then there needs to be a good mix of products and services to select from. On top of that, there must be a top quality SCV in place, accurate cost and other data, plus the skills and infrastructure to make use of it.

Outbound volumes have also dropped dramatically, with the last decade’s carpet bombing giving way to smaller renewal and cross-selling campaigns, which also reduces optimisation’s appeal. Likewise, inbound targeting via the web and triggered email or via the call centre will also compete for funds. Every company will have to make its own choice.