One thing we have learned over the last few years is that we’re being asked to achieve more, with less. We’re facing increasing competition globally for products and services; more niche companies emerging online that provide everything the consumer wants at a lower cost base; consumers are expecting us to provide relevant offers to them, no matter how deep our relationship; new products or services being released by our own organisations; internal competition for share of our customer and prospect wallet; and often with tightening budgets.
With each of these factors affecting our marketing planning we often find ourselves struggling to reconcile the effect of each of these levers, requirements or constraints across our campaign planning. We create priorities depending on who shouts loudest or has the biggest budget, or what is subjectively perceived to be the next best offer.
Marketers are seeking innovative ways to acquire, retain and grow customers. The situation is exacerbated by a growing number of customer touch-points, more granular audience segments and more product and offer permutations. The temptation is to send more messages to more customers in the hope of achieving better marketing results. This is a particular challenge for companies that have a large number of competing communications they want to send to the same or overlapping sets of customers.
This situation raises a number of business challenges:
Contact optimisation is a mathematical approach for identifying the best combination of messages for each customer from among competing options, while complying with business objectives and constraints, with the goal of targeting the right customer, with the most attractive offer, at the optimal moment, using the most strategic channel.
This, however, is a complex problem to solve. Customers x channels x offers x time options creates potentially billions of variable combinations. To add to the complexity, marketing organisations are also constrained by business goals and restrictions, such as revenue targets, contact policies and budget caps.
Under these pressures, organisations will use a variety of methods to get the best results, all of which are perfectly valid, but will often result in significantly different rates of success.
To demonstrate the pros/cons of some of these methods, I’m going to use a fictional organisation, the Purple Supplies Company (PSC). PSC doesn’t have many customers (10) or many campaigns (3), but I hope it’ll demonstrate the point.
PSC’s data scientists have scored each customer on their likelihood (out of a maximum of 100) to take up the offer presented in each campaign as part of their daily data preparation activities. They’ve chosen this simple propensity score for consistency, but could have chosen any metric of success, such as potential revenue.
We’ll assess each campaign’s potential effectiveness against each “prioritisation/optimisation” method used and compare them using a per campaign and overall score.
We also have some very simple business constraints:
1. One customer can receive one campaign only
2. One campaign has a minimum of 3 and maximum of 4 recipients.
These are applied to ensure we don’t over market to an individual or leave any campaigns without enough recipients.
This approach is effectively a random based selection on the first campaign to run that day. It may use elements of customer scoring but is most likely to choose the first data records it hits in the database are output. The resulting Campaign scores are arbitrary and is simply based on the first 3 or 4 customers available in each campaign and depending on the order they are run in the day will create significant variations. Thankfully this approach is increasingly rare, but we do still see it on occasion.
In this case we prioritise our Campaigns and select the best customer(s) for each Campaign in order. The target recipient for that campaign is the highest scoring customer within the campaign. Whilst this approach is likely to produce good results for the higher priority Campaigns, it does result in poorer results for the lower priority communications as they will typically be left with the lower scored customers to choose from. i.e. the higher the campaign ranking the larger pool of customers they get to cherry pick. In this instance Campaigns are taken in the order A > B > C. In the example below, Campaign C can only be sent to customers 1, 5 and 8.
This switches the emphasis on to the customer first. In this approach, each customer is evaluated row by row, evaluating each Campaign in order to select the highest score campaign for that individual. This can mean however that the higher up in the data the customer is, the more likely that they will get a Campaign offer they are likely to respond to as they have wider pool of campaigns to select from. As you can see in the data set below customers 9 and 10 must receive Campaign C as they would otherwise breach the minimum number of recipients per Campaign. The further down in the customer list an individual appears, the fewer campaign options they are left with.
This approach evaluates all of scores, rules and constraints and reduce the bias of Customer or Campaign. Optimisation algorithms apply all the logic, across all the data at the same time and are designed to identify the best combination of all the factors in order to provide the maximum output score. It can be difficult to determine which factors have greatest influence, and the introduction of one rule can significantly impact the output results, but always with the same goal.
Even using this simplistic model, this approach results in an 8% uplift on the best performing alternative method and 12.5% uplift on the worst performing method. Convert this to the financial measures of your organisation it could result in many tens of thousands of pounds/dollars in incremental revenue.
Whilst it’s easy to see and evaluate the impact on 3 Campaigns and 10 Customers, it’s much harder to do this with millions of customers, tens of campaigns, all creating a significant number of permutations within a day or week.
There are however some organisational challenges to implementing a more optimised approach.
Many marketing organisations we have worked with are swayed by the demands of product managers, their requirements push the highest scoring customers to the top of their campaigns, however this is likely to have a negative impact on other campaigns and the business overall. Changing the way product managers are recognised and rewarded is key to ensuring that the organisation benefits. It is important to set targets, but not at the cost of other business units.
Other organisations struggle to score their data at a meaningful level, whilst the most optimal approach is to score every offer, against every customer, this can create a significant burden on the data scientists. Starting at a high product/category level and working down through the product hierarchies will result in improving scores and optimisation over time.
It is tempting to try to create the perfect optimisation model for the organisation on day one, however this activity can take many months, if not years, to evolve and will never be completely “perfect”. Most successful optimisation projects start with a limited scope to prove the concept and build confidence in the solution, followed by incremental deployment, adding new rules, campaigns and scoring as the business learns and develops.
Investing time and resources, addressing these challenges and embedding marketing optimisation into the marketer’s business process will always lead to overall campaign effectiveness.
Contact us to find out how Purple Square can help you with your marketing optimisation initiatives.
After the surprise announcement on 4th April 2019 that IBM are selling their remaining Marketing &am...
Invest for success The bulk of enterprise software implementation effort (and budget!) is typically ...