Our Approach to Marketing Automation

It’s clear that 2017 has been the year where companies start shifting their focus towards efforts on marketing automation and experimentation through the use of machine learning. For AdWords, Google has consistently been pushing improvements to their enhanced CPCs, increasing and decreasing bids dynamically in every auction based on a user’s location, past searches, device model, and more.

For the Google Display Network (GDN), new ad formats are arising that take 5+ creatives and 5+ messages which get mixed and matched into multiple variations according to the site they show on and the user they get shown to. For Facebook, automated budget allocation has become an option, allowing Facebook to take your daily budget and distribute it across campaigns according to which ones have the highest chance of performing well that day.

These are only a few of the examples that have cropped up over the past few months.

While we love that our partners are actively investing in technology to free up our time to focus on more strategic initiatives, we hold a healthy skepticism for all marketing automation developments. At the end of the day, these companies are developing systems that are designed to work for the majority of advertisers. This means that there’s a pretty good chance that not every solution is going to work for every brand we work with. The best way to make sure that marketing automation efforts are working for you is to test, build, and monitor.

Marketing automation should never be seen as an instant solve to your problems or a “set it and forget it” solution. If you’re not testing solutions, you’re probably running with a solution that’s actually hurting your performance. With Google, Facebook and many other vendors now offering campaign experiments designed to dynamically split traffic by 50%, it’s easier than ever to keep testing new automated systems. Fortunately, these experiments mirror your current campaign setup, allowing bids to stay in sync while testing things like ad rotation, delivery methods, bid modifiers and more.

In the event that campaign experiments aren’t an option, we recommend duplicating campaigns and splitting these campaigns by geos that perform relatively similar, allowing you to test on a separated group of people. By performing a causal impact analysis on these campaigns, you can see both the performance difference and the incrementality of the change.

There have been plenty of times where our reps have continued to recommend we use a new automated system or that we don’t use our own technology which powers an unrecommended strategy. However, by testing instead of blindly following recommendations, we’ve been able to drive better results for our clients, prevent potential account issues, and ultimately determine which automated solutions work best.

Part of the problem with marketing automation right now is that (oftentimes) you’re blindly trusting the very same companies that make money from you spending more to create systems that help you spend less. Many vendors rely on telling people to “trust us, it just works,” creating a black box of marketing that makes you feel helpless to change things when things don’t perform exactly how you expected. Additionally, many of these systems are using machine learning, a relatively new field that can work tremendously, but also provides little feedback into how data points were looked at and weighted along the way.

We believe that one of the best ways to break this cycle and make sure marketing automation works for you is to build your own systems or work with your agency to make a custom-built solution. While the level of investment is higher initially, in the long term, you gain the ability to replicate your success across other marketing programs instead of having no clue how you drove great performance.

With your own system, you get to use a mixture of your own data (instead of just using a vendor’s data), so you know exactly how the process operates and you can continue to tweak logic to improve performance. Instead of just turning off an automated process when it doesn’t work, you now have an opportunity to try out new logic or new data points to find a better solution.

As our clients have come to us with problems, we’ve helped develop custom solutions that automate:

The updating of mobile and geo modifiers on a daily basis, leveraging historical insights and forecasting.

The development of a full-fledged PLA structure using only a product feed.

The process of creating and updating ad copy on a global scale, referencing promotional calendars.

If you have a database full of internal data, an analyst to interpret the data, a marketer to create a decisioning process, and a developer to create a script that codifies the process and leverages the data, you can begin taking control of your own marketing automation.

When so many programs of ours rely on some form of automation, and the number of automated elements grows each month, it’s more important than ever to keep a watchful eye on the programs to make sure nothing is going wrong. Automation always requires some tolerance for risk, one that we believe is acceptable in most cases for the underlying benefit of time saved. Knowing this risk, PMG has built an internal alerting system called “Deputy” that actively monitors all of our client’s programs across search, display, and social on an hourly basis.

Deputy looks for irregularities like campaigns that suddenly stop spending, pixels that stop showing conversions, bids that grow abnormally high/low, feeds that have elevated levels of disapproval, or pages that lead to a 404 page or 301 redirects. If an issue is found, teams are immediately notified, helping to drive a quicker resolution. Deputy can also send our teams ads, keywords, or other elements that have been created by automated processes to help them keep an active eye on what’s being placed into the accounts.

As we implement more automated processes and document more things that could potentially go wrong with our campaigns, Deputy will grow to protect our clients from high impact issues, saving hundreds of thousands of dollars in the long run.

When you set up monitoring processes yourself, you can rest assured that the things you believe are errors are always going to get caught. This removes a lot of the worry and the “what-ifs” that marketing automation can bring up.

—

In total, marketing automation is a wonderful thing for the industry, helping businesses drive better marketing results in the background with real-time changes. Over the course of the next five years, this change will drive a radical shift in the skills required for account managers. These individuals must think more strategically, constantly test systems against each other, create logic to develop new systems and leverage new data points and find ways to monitor points of failure. We’re poised and ready to help facilitate this change for our portfolio of brands. Are you?