Artificial Intelligence and Machine Learning are hot topics everywhere right now, in every industry, including marketing. Everyone is looking for that edge to help get ahead of their competition – what we at W2O call an unfair advantage – and looking at AI as the ‘silver bullet’ to help provide that something extra. But does it really work that way? Will AI really give us all the answers to keep us one step ahead?

W2O has one of the oldest data science groups in the industry. With over 25% of our staff in Analytics, Data Sciences and Development, W2O have been applying these techniques to how we generate better insights for years. I thought I’d sit down with two of our top experts to find out just how modern techniques improve efficiency and effectiveness for marketing and communications insights. Dr. Helene Brashear and Dr. Yash Gad are two Senior Data Scientists with W2O and have been developing solutions for applying AI and ML in our agency environment. Here is what they had to say about where we are now, and what the future holds for agencies:

Madelyn – In an agency setting, what is AI, and what do we use it to do?

Yash Gad (YG) – For me, anything we can do to speed up work, make us do things more consistently and more efficiently, that’s the job of any good software design. We use it to tag certain data in certain ways and cluster data for a start.

Helene Brashear (HB) – We look for opportunities to decrease time for work or do something we haven’t been able to do before, like handle extremely large volumes of data.

YG – We could throw hundreds and hundreds of people hours at the same problems, but it wouldn’t make sense – we can’t spend the Analyst hours doing that; we need them to spend those hours doing smarter things.

HB – We can now explore data in new ways – looking for the computational models underlying the unstructured data and matching that to our mental models. It’s like the analogy of mapping a neighborhood – if you map of the streets, you can predict the paths that go from point A to B most logically.

Madelyn – OK, this is interesting theory, and great background, but how does applying these models at W2O offer us and our clients an advantage? What is different about these techniques we can’t get to in another way?

YG – Larger and larger scale data sets that would take years if we tried to process them using manual or old techniques. But the real difference is that we can get to an answer we couldn’t get to otherwise; through other manual means. Using these models can reveal unique perspectives we wouldn’t see through brute force of scaling machines.

HB – It’s also a team effort between Data Science, Analytics, and the Strategy teams to come up with the right inputs to get to the question and input design. We need to know what we are looking for to design the models and inputs the right way. When you are teaching a model to look for patterns or recognize paths, you must know what to teach it. This approach requires a blend of domain expertise across the whole creative and agency set, plus analytical work, and machine learning.

Madelyn – Broadly speaking, how do we apply AI/ML to our processes and data sets? Do we see unique benefit for healthcare clients here? How about technology clients?

HB – There are unique challenges to Structured v unstructured data – Healthcare has lots of structured data (like Provider ID numbers, taxonomies, prescription fill data), which require a different approach, and connecting it all are still a challenge. The highly structured data is nice for data structures, but requires more domain expertise to model relevant medical context. Nearly everyone but healthcare has unstructured data, and that presents its own challenges of modeling and interpretation; but we share and learn from all approaches.

Madelyn – What do you think is next or most exciting for our industry or team in this space?

YG – New industries and areas are becoming fans of AI and ML and are becoming familiar with the space, so reading out what the models are giving us in human terms is an increasing challenge – giving out new translations in business-speak.

HB – Distilling knowledge from machine learning models is a big challenge – training a model isn’t good enough. We spend lots of time validated and stress testing models for real world performance. Sometimes you may need to prove that your model isn’t doing something illegal, like in finance or other regulated industries. We must understand what biases our models have so that we understand the outcomes it produces and how we may be impacted by them. Seeing Data Sciences and AI as a collaborative tool and not as a magic box. It’s something that helps people amplify their own talents – the inputs make the outputs, just like a good creative brief makes creative better, working with your DS team makes your data analysis tools better.

One of the challenges of machine learning is that results can be mathematically correct, but not interpretable or meaningful. Because of this you spend a lot of your time strategically crafting your research questions and curating your data.