How to Reduce Marketing Bloat with Basic Data Science Techniques

Spend Less on Marketing without Losing Customers

Momentum is an incredibly powerful factor in marketing strategy. Companies draw on years of experience to refine presentation, process, and content in ways that draw in new customers and keep existing clients happy.

But momentum also leads to bloat. As organizations propel themselves forward, momentum grows with additional tactics that engage customers in new ways. Like a snowball growing as it rolls down a hillside, communications touch points pile on. Eventually, marketing momentum takes a life of its own and, from the customer’s point of view, they see an avalanche of messaging that makes them more likely to ignore it all.

The #1 reason people cite for unsubscribing from email lists: ‘I get too many emails.’ The #2 reason: ‘The emails are not relevant to me.’
— December 2016 MarketingSherpa "Customer Satisfaction Research Study” via eMarketer.

Momentum not only leads to bloat in volume, but also bloat in purpose. Momentum favors tactics that work for multiple client segments over targeted messaging. Even though every marketer knows that focus is critical in theory, it becomes difficult to achieve in practice.

Marketers who can personalize web experiences see a 20 percent uplift in sales... but most don’t know how to do it.
— Econsultancy/Monetate

Solving for the momentum of an avalanche isn’t easy. How do you pull back tactics without risking the success that you’ve worked so hard to create? And how do you match tactics to audience segments when the generalized legacy methodology appears to work well enough?

There are answers, using established machine learning methodologies, and the biggest brands in the world do it every day. For smaller organizations, the task may seem more daunting. Is it worth the effort? The answer is easy: if you could save 50 percent on marketing costs but achieve better results, would you do it?


A Two Step Approach to Reducing Marketing Bloat

STEP 1: Eliminate Waste with Randomized Controlled Experiments

Sometimes it’s necessary to put away some of what you know about your customers in order to make efficient decisions. Commonly associated with clinical trials, a randomized controlled experiment can be equally effective in marketing. Similar to an A/B test, a randomized controlled experiment makes it possible to test the impact of an existing marketing tactic against doing nothing at all. The key is controlling for factors both groups are exposed to and selecting the test groups randomly (rather than based on segmentation). Randomization allows you to account for selection bias and, if carried off correctly, the experiment can teach you where you are spending money unnecessarily.


How to Run a Randomized Controlled Marketing Experiment

  1. Choose a Variable to Test: Which tactic might be redundant? Or very expensive?

  2. Choose Relevant Audience: Do you want to test one segment or multiple?

  3. Finalize Hypothesis: What theory are you trying to test? For example, “Eliminating our customer newsletter will have no impact on retention.” Or “This new and improved tactic will increase conversion 10%.”

  4. Account for Disruptive Factors: What factors could spoil the experiment? For example, overlapping tactics that could trigger a false positive.

  5. Finalize Experimental Process: Ensure that causal relationships will be as clear as possible with a clear protocol and results tracking.

  6. Pilot Study: Conduct a small experiment to minimize risk with a randomly divided test group and control group.

For a great overview on randomized controlled experiments, see Amy Gallow’s excellent piece in Harvard Business Review.

STEP 2: Use Predictive Analytics to Maximize Yield

One of the best uses of predictive analytics is to decide who NOT to market to. In other words, you can use predictive analytics to create a smaller, less expensive campaigns, with higher conversion rates. Using a basic predictive method such as binary logistic regression or more advanced techniques like random forest decision trees, you can create a predictive model that uses historical data to determine who in your customer base is most likely to respond to a given marketing tactic in the future.

By scoring the likelihood of a positive response, you can focus your marketing campaign on the highest yield targets alone. For example, you could decide to only target the top 20% scored portion of your list and still achieve 80% of the best possible result if you targeted everyone. This approach is particularly effective in high cost tactics like direct mail, but is just as relevant in email marketing where the dangers of email overload are continuously mounting.

Reducing marketing bloat without losing customers isn’t hard — it just takes a deliberate experimental process that allows for for iterative improvement. And, of course, a desire to challenge the status quo.