Data gives advertisers the power to fine-tune their campaigns and deliver strong results. Industry experts shared their best advice about advertising data, including methods for collecting it, analyzing it, and putting it to use.
From Karla Crawford-Kerr, VP of Marketing, Hawthorne Advertising.
Don’t just collect data, analyze it continuously.
In advertising data science, it’s not enough to report on the past and present. Data without context only states the obvious—what has already happened. Businesses have spent the past few years accumulating massive amounts of data, with 60 percent-73 percent of that data to go unused, according to Forrester. Big and small data collection is clearly not enough. Data-driven decisions require analysis and insight to deliver value and lead to meaningful change.
To get the most out of advertising data science, marketers need to look ahead and focus on how data can help them make better decisions. Marketers must be open-minded and process-oriented in their Advertising Approach Analysis. This means testing, collecting data, evaluating, testing again, and analyzing to really learn from the data and make tweaks and game changing decisions. This process is a continuous circle in advertising. It never stops. To drive results, it’s important to focus on data from start to finish.
Here are seven steps that advertisers can take to optimize their use of data science.
- Understand Audience and Audience Targeting. This means looking at how and where the target audience is consuming media. It’s critical to be medium agnostic. Do not overlook options like CTV based on perceived cost, as it may be an effective extension of linear or digital.
- A/B Test. Test everything—messaging, content, and creative.
- Test the results. Look at the results of campaigns for attribution and impact. See what worked and what didn’t.
- Make data-driven decisions. Look at the short and long tail of conversions (first touch vs. last touch).
- Trust your instincts. As a marketer it is important to know when to trust instinct vs. getting bogged down with the data. This could mean calling out false positives or evaluation of results that focus too narrowly on the short term.
- More testing. Hone your message and creative.
- Optimize spend and investment.
- Start again. Understand your audience and audience targeting.
Director of Marketing Technology, Titan 47
There are answers for your cross-system attribution questions.
Today’s marketers have a gold mine of data sitting right in front of them that can be turned into insights with even the most basic of technical skills. A mix of Google Tag Manager, marketing channel APIs, and an object-oriented programming language, like Python, can allow you to accomplish even the most complex of tasks—including cross-systems attribution modeling.
Google Analytics has strong attribution capabilities out of the box, however, what if your main KPIs are simple transactions held in Google Analytics? This is a cross-systems attribution problem. You can use Google Tag Manager to create custom dimensions for the Google Analytics client id as well as your product or lead user ids (if applicable). Using Python and the Google Analytics API, you can extract multiple segments at scale into a data warehouse, such as Google BigQuery. From there, data integration can be performed across your Google Analytics data and your database or CRM. Now, you’ll have pre-conversion user data combined with post-conversion customer data for attribution modeling.
Languages like Python are also great for deploying different statistical-based attribution models once your data is integrated. The sky’s the limit when marketing is combined with technical skills and today’s wealth of knowledge online helps marketers jump right in.
Andrew Van Benschoten
Senior Manager, Data Science, Ovative Group
You can base most of your business decisions on simple models.
The data science field has many flashy technologies such as neural networks and reinforcement learning, but the majority of your business questions can be answered with more straightforward approaches.
Linear regression can get you ~70% of the output from a more complex model, and in 1/10th of the time. And in many cases, the added accuracy won’t change your business decision. I’m going to invest in the marketing channel with the highest-ranked ROAS regardless of whether that number is 5 or 5.1. These simpler approaches also have the added benefit of easier interpretability. Your clients may not understand pooling layers or stacking vs. ensembling, but they can understand that a one-dollar change in channel spend leads to a three dollar change in overall revenue.
Throughout this process you might discover that you need a more complex approach to accomplish your business objectives, and that’s ok! Starting simple provides a perfect framework for conducting auxiliary tasks such as exploring the nuances of your data, thinking through appropriate KPIs, and the like. These critical components are often overlooked if you have to spend your time setting up your GPU configuration or debugging a beta-version ML library. At the end of the day, your goal is not to use the latest data science buzzword, but rather to uncover new insights that drive your client to act.
Owner, Reedy River Marketing, LLC
Data science can’t account for all the intangibles, so figure out the story behind it.
I think the biggest misconceptions about advertising data science are that the datasets have to be enormous, the algorithms have to be complex, and an advertiser’s background has to be extremely technical to make it work. The truth of the matter is that ADS scales based on the data you have to work with, and most importantly, the client’s objective. A lot of the platforms commonly used for digital marketing for a smaller and medium-sized business are already incorporating the machine-learning aspect on the backend. Smart-bidding capabilities, automated rules, and a variety of different conversion techniques are already taking advantage of the machine-learning aspect concerning user behavior and desired outcome.
The next step is to decide the best way to utilize these types of tests in conjunction with more manual data science techniques. There are only so many inputs we typically give to a platform to execute our campaigns. We simply can’t completely relay the intangibles and the human experience fully into the machine. Even if we could, it wouldn’t think, analyze, or optimize exactly as we do. That is why we need to find the best convergence of both, tested together, to ultimately achieve the client’s goals with all aspects of ADS working together as seamlessly as possible.
Finally, it is also extremely important not to get lost within the data. We have more data than we could ever fully digest in most cases. It is too easy to get caught up in treating the analysis like starting a 5,000 piece puzzle versus a beautiful painting that is just missing the last few strokes to complete it. The data is telling us an amazing story, but far too often we simply don’t listen. At least not actively.
Search Manager, GroupM Ireland
Capture the data you need to understand the full ecosystem.
This is one of the biggest challenges and is often overlooked: measuring the level of contribution of a channel in the mix to achieve the right trade-offs by efficiently connecting multiple data points. Let’s see how to bring the pieces of the puzzle together and unlock opportunities.
Data visualization tools are the way to go, as they allow real-time reporting reflecting the maturity of tracking an entire KPI in real-time with a complete vision: based on data-driven models, algorithms and machine learning.
Attribution & Contribution
There is a real awareness on how to use attribution and contribution tools; it is now the norm to value the upstream of the funnel and not only value last-click models. It is demonstrating the necessity of synergies in the wider marketing ecosystems.
Bring That Puzzle Together
Between paid search and TV, the connection is intuitive: the behavior of a TV viewer is predictable and measurable, and the channel cross-synchronization will maximize the impact of TV. But how to prove the efficiencies driven on foot traffic, this is the holy grail that agencies can offer thanks to operational maturity and omnichannel approaches.
Start using the data that the consumer actively shares with your business—think questionnaires, polls to maximize data capturing capabilities instead of fighting the growing chimeras of a zero-cookie internet.
The Way Through The Forest
Clustering and trainable algorithms can assist to find the true way in your forest of data. This can unlock unimaginable patterns and campaign automation with unprecedented efficiencies.
Reach For The Stars
To conclude, don’t forget to feed the top of the funnel to fill the bottom and start maximizing on the cross-channel synergies by taking advantage of symbiotic ecosystems.