for marketing campaigns is amazingly hard. Much of it comes right down to trial and error, despite the fact that we all know that more targeted strategies would work higher. We just don’t know easy methods to get there. The method often includes launching a campaign, observing it, learning, making adjustments, after which trying again. This trial-and-error approach has real strengths. It encourages movement over paralysis. It allows teams to learn quickly, especially in fast-changing markets. For early-stage growth or limited data environments, it is commonly the one practical option.
I would like to introduce a unique approach. One which is, undoubtedly, tougher, advanced, and sophisticated, but in addition revolutionary and memorable. That is the approach that takes corporations to the subsequent level of information maturity. Let me introduce you to expected value modeling.
Before we start, I would like to preface by saying this approach takes up full chapters in some data science textbooks. Nevertheless, I intend to be as non-technical as possible. I’ll keep the ideas conceptual, while still providing a transparent framework on how this might be achieved. In case you are taken with learning more, I’ll cite useful sources at the tip.
Let’s begin.
What’s Expected Value Modeling?
Expected value is a key analytical framework that permits decision-makers to think about tradeoffs when there are unequal costs and advantages. Consider a scenario where a a machine learning model helps diagnose a patient with cancer. Frameworks and models that only include easy accuracy (either the prediction was right or unsuitable) don’t account for the tradeoffs within the predictions.
On this case, not every “unsuitable prediction” is similar. Not diagnosing a patient with cancer after they have it’s infinitely more costly than diagnosing someone with cancer after they even have it. Each predictions were technically unsuitable, but one cost a life, the opposite didn’t.
Thankfully, our marketing strategies are usually not life-or-death situations. But this principle applies the identical. The choice on who to focus on in a marketing campaign, and who to not, may end in largely different costs for the business.
Expected Value Modeling expands this horizon to account for more possible outcomes, and allows us to measure the price or advantage of each. This framework is deeply depending on business knowledge of material experts to find out the results of every end result. Our goal here is to grasp easy methods to design a technique that statistically optimizes for our goal. For the rest of this text, we can be focused on learning who to focus on in a marketing strategy so we maximize profit.
Start with a Purchase Likelihood Model
A Purchase Likelihood Model is a machine learning model that predicts the probability that a customer will purchase a product. Let’s consider we’re running an ad campaign for an e-commerce business. All and sundry that clicks on the ad creates a row of information. They see the campaign, browse your store, and ultimately decides to buy or not to buy a product. During this process, a mess of information points must be collected. The machine learning model analyses all historical data to acknowledge patterns. It learns what are the aspects that influence the probability of a customer to buy. Then, it applies those patterns to latest customers to predict if they’ll purchase a product.
This model by itself is of maximum value. It tells the business who’re the shoppers most definitely to purchase a product and what elements of the campaign influence purchase likelihood. We will use these insights to tailor our next ad campaign. That is what data-driven decision making looks like.
Implementing Expected Value Modeling
To maneuver forward, it’s important to grasp the concept of a confusion matrix. A confusion matrix is a x table where represents all possible outcomes. For simplicity, I’ll follow a 2 x 2 confusion matrix.
This matrix incorporates the expected outcomes in a single axis and the actual outcomes in the opposite. It provides us with 4 cells, one for every possible end result in a binary classification problem, as is our purchase likelihood model (either a customer purchases a product or doesn’t). This ends in the next possibilities:
- True Positive: we predicted the shopper would purchase, they usually actually did.
- False Positive: we predicted the shopper would purchase, but they didn’t.
- False Negative: we predicted the shopper would NOT purchase, but they did.
- True Negative: we predicted the shopper would NOT purchase, they usually in truth didn’t.
Here’s an illustration:
To implement expected values to every end result we want to have a deep understanding of the business. We want to know the next information:
- Profit per product sold.
- Cost per click.
- Purchase probability per customer.
In the identical example for our e-commerce store, let’s consider the next values:
- Profit per product sold = $50
- Cost per click = $1
- Purchase probability per customer = from our Purchase Likelihood Model
Knowing this information we will determine that the good thing about a customer clicking on our ad campaign and buying a product (True Positive) can be the profit per product ($50) minus the price per click ($1), which equals $49. The fee of a customer clicking on our campaign but not purchasing (False Positive) is just the price incurred for the press, so -$1. The results of not targeting a customer that will not purchase is $0, since no cost was incurred and no revenue was earned. The results of not targeting someone that will purchase can also be $0 for a similar reasons.
I do need to acknowledge the chance costs of not targeting someone that will purchase or the potential of someone purchasing without being targeted. These are more abstract and subjective, although not unattainable to measure. For simplicity, I is not going to consider them on this scenario.
This leaves us with the next confusion matrix:
Cool, we now know the concrete cost or advantage of each end result of our ad campaign. This permits us to grasp the expected value of a targeting a customer by utilizing the next equation (sorry for throwing math at you):
Where the expected value is equal the probability of response (P(buy)) times the value of a response (Profit if buy) plus the probability of a non-response (1 — P(buy)) times the cost of a non-response (Loss if no buy).
If we would like the expected value of targeting a customer to be positive, meaning now we have a profit, then we will rearrange the equation to the next:
Which means that, based on our purchase likelihood model, we should always goal every customer with a purchase order likelihood exceeding 2%.
You don’t must have a level in math or statistics to implement this, but I wanted to indicate how we got there.
We have now our answer: we want to focus on all customers whose purchase probability is above 2%. We will now return to our purchase likelihood model an discover which customer segments fit the factors.
We have now discovered exactly who to focus on, we tailored our campaign to their needs, and deployed a marketing campaign that works. We designed our strategy with all the correct foundations by making true data-driven decisions.
Taking it one step further with Profit Curves
We have now built our framework and designed our marketing campaign in a way that optimizes our ROI. Nevertheless, there are sometimes additional constraints that limits our ability to deploy a campaign, often related to how much budget is allocated and the way many individuals might be targeted. In these scenarios, it is beneficial to know not only the optimal decision, but in addition the expected value across a wide selection of possibilities. In those situations, we will embed expected value calculation into our purchase likelihood model training process.
As an alternative of selecting models purely based on technical performance, we will evaluate them based on expected profit. Or use a combined approach that balances predictive strength and economic impact.
While we’re constructing our model, we will calculate the expected profit across your entire range of those who we will goal, from targeting no one to utterly everyone we will. In consequence, we get a profit curve plot:
Within the y-axis now we have the expected profit for the marketing campaign based on how many individuals we goal. Within the x-axis now we have purchase likelihood threshold. We get increasingly narrow with our campaign as we increase the edge. If we increase it all of the approach to 100%, we won’t goal anyone. If we drop all of the approach to 0%, we will goal everyone.
As in our example before, we see that the utmost expected profit lies after we goal every population with above a 2% purchase likelihood rating. Nevertheless, perhaps now we have a more strict budget, or we would like to develop a separate campaign just for the really high likelihood customers. On this case, we will compare our budget to the curve and discover that targeting customers above a 12% likelihood rating continues to be expected to supply a robust profit on a fraction of the price. Then, we will go to the identical process we did before to design this campaign. We discover who’re these customers, what impacts their purchase likelihood, and proceed to tailor our marketing campaign to their needs.
It starts and ends with business knowledge
We have now seen the chances and value that expected value modeling can provide, but I need to reiterate how necessary it’s to have knowledge of the business to make sure the whole lot works easily. It’s crucial to have a solid understanding of the prices and advantages related to each possible end result. It’s paramount to properly interpret the model results to totally understand what levers might be pulled to affect purchase likelihood.
Even though it is a posh approach, it isn’t my intent to sound discouraging to the reader who’s learning about these techniques for the primary time. Quite the alternative. I’m writing about this to spotlight that such methods are not any longer reserved to large corporations. Small and medium size businesses have access to the identical data collection and modeling tools, opening the door for anyone that wishes to take their business to the subsequent level.
References
Provost, F., and Fawcett, T. . O’Reilly Media.
.
