Ad agency Doe-Anderson has been around for over 100 years. Three years ago, the agency launched an innovation unit to focus on adtech as a way to stay competitive and identify new revenue streams. We spoke to Lee Dorsey, SVP of Innovation at Doe-Anderson, about those efforts.
Tell us about the innovation unit.
Dorsey: We’re building out an ecosystem and tapping into an existing one within Louisville. We’re doing this by finding startups with some of the skill sets that we don’t have internally, and then partnering those with a lot of our clients’ challenges. It’s all about finding the sweet spot between the two.
Do you have an example?
Dorsey: We sat down with Enable.ai in the infancy of the company’s development and collaborated on a social media performance and insight process using artificial intelligence (AI).
We can take any of the social content that our clients have and then break it up into sub-variables. So if it’s an image of a cat and a dog, then we would isolate that there’s a cat or a dog; or if it’s blue or green; or if it uses these words or emojis. Isolating all of that into singular variables means we can analyze how our specific creative decisions drive performance.
The other half of that puzzle is that we’re using natural language processing (NLP) to break that data into topical analysis, sentiment analysis and then any kind of business-indicating language. The theory is just because you get 100 comments doesn’t mean the content is great. Ninety-eight percent of those comments could be fluff or could be negative. The way that most companies currently measure and evaluate social doesn’t add that level of nuance.
What is Doe-Anderson learning from this method?
Dorsey: We realized that a lot of rich conversation indicates some linkage between what’s happening online and what you want to have happen offline — such as the user making a purchase or engaging with that brand somewhere. So we’ve built and automated all of that, thereby allowing us to quickly turn around insights and create entirely new metrics for evaluating the success of our program.
Do you have an example?
Dorsey: We did a pilot project with Texas Roadhouse during the pandemic. The restaurant chain had to shut down for in-person dining. It was a big challenge, because they really hadn’t done a lot of carry-out dining. They needed to make sure they could stay open.
What we realized by monitoring the conversation on their social media pages is that the role that they play for consumers is not just feeding them, it’s a full entertainment experience. We realized that we needed to quickly pivot our strategy to replicate the in-dining experience at home.
So we did a lot of different things, including an augmented reality cowboy hat. We launched a Spotify playlist with jams on the jukebox that you could typically listen to while actually dining in one of the restaurants. We created a specific community for parents, offering various activities such as coloring sheets or a crossword — the kinds of things you might customarily give to a child at the table [when you’re there] to entertain them. We also released some of the cocktail recipes so the parents could enjoy that while their kids are entertained.
Walk us through the process. Company X comes to you now, after COVID. How do you help?
Dorsey: Our clients already own their data; however, they can’t make sense of it because it’s unstructured. What we do is ingest that data, flip it into something that’s more meaningful and then provide the consultative and strategic services that actually make it impactful.
We prioritize the things we want to measure, and then train the AI model to recognize them so we can push it across historical data sets. Then, moving forward, the model automatically generates what those tags are so that we can analyze it faster.
We do the same thing at the comment level. We would review historical conversations to understand what things already exist. We would have conversations with you to understand your business objectives and prioritize any kind of business-building conversation within those variables. First, we train the models to generate X. This provides us with a good starting data set from which we can build reports for the client or we can use it ourselves to inform the decisions that we make.
The results have been fantastic. It helps us get a glimpse behind the curtain on some of the algorithms that Facebook, Instagram or Twitter are using. We’re able to get much better reach, along with much better engagement, by looking at this. It enables us to constantly evolve our strategy to reflect whatever is happening on the back end there. I think this is the way social media strategy should be run.