Makia Yagodich, VP of Analytics at Cannella, shares insights of how
Cannella is using emerging technologies for response forecasting
Q. As a media company, how have you responded to the data
and analytics challenges and opportunities?
A. As the industry and our business continue to evolve,
advanced data and analytics are no longer a ‘nice to have’
feature but have instead become table stakes for media
agencies. Understanding the sources of all consumer
responses across multiple touch-points is critical to properly
optimizing media schedules, valuing media channel
contributions, and maximizing ROI.
Cannella understands this shift and has invested heavily in
building out a data and analytics department. Five years
ago having a data scientist on our staff was a luxury, now
it’s a necessity. We’re not only looking to find more efficient
solutions to today’s existing media challenges, but proactively
addressing those that we anticipate arising in the future.
Q. Looking at all the developments in analytics over the few
years, in your role as VP of Analytics, what has been the most
significant change you’ve seen in media management?
A. In TV marketing, advanced analytics has emerged as one
of the most influential tools in driving performance evaluation.
Cutting edge methodologies like neural networks and machine
learning can be applied to the problems of old to offer insights
and business intelligence much more efficiently now. These
tools are now with reach.
Q. How are marketers adapting their use of response
mechanisms to aid in attribution?
A. Historically, direct response, or performance based
television, relied heavily on deterministic attribution methods
such as the 1-800 number, unique URLs or promo codes. It’s
hard to beat deterministic attribution and we will always look
to implement these tools where it makes sense. Additionally,
with the dominance of mobile as the preferred device to
inquire (60% of web traffic comes through mobile) we’re
seeing more marketers embrace SMS/text messaging within
their CTA for offline media.
For spot level attribution, our approach is to first capture as
much deterministic response as possible. Obtaining as much
certainty on the total mix of response through deterministic
methods is always beneficial, so we encourage our clients
to integrate as many deterministic methods as feasible.
Then, using probabilistic methods, we model the remaining
responses that can’t be attributed to a source on a 1: 1 level.
If we’re working with a marketer that sells product through
multiple sales channels such as third party marketplaces and
retail, we need to look at the marketing investment at a higher
level and identify patterns that can be statistically validated.
In these situations we need to incorporate a media mix model
approach to evaluate media channel contribution.
Q. What is Cannella doing then, specifically, to try and address
this complex question?
A. At Cannella, we’re applying machine and deep learning
techniques to the problems of attribution and response
forecasting. Neural networking and machine learning can run
thousands of algorithms at a time, which previously couldn’t
have been accomplished. It has been a game changer.
We’re now deploying automated scripts that loop through
the variable input and model selection process. What once
took data scientists months to tune can now be done in a
few hours. We provide a scalable, probabilistic approach to
attributing responses at the spot level.
We’re now also running models to predict response as part of
the tuning process. We use a customized version of a neural
network that can provide high levels of accuracy. The neural
network on our cloud platform allows us to analyze hundreds
of variables over several thousand iterations to find the best
solution. We’re adapting these processes to more accurately
predict the responses per airing for proactive media
management and schedule optimization.
Q. How are these models used in the day-to-day media
A. We embed these analytics into our planning and strategic
process. This provides insight and understanding to the
expected response rate over a two week window and allows for
recommendations on optimal dayparts, stations, and creatives
to help bolster performance. By leveraging our internet based
reporting and visualization tools, we enable our buyers to drive
maximum value for our clients with increased planning and
Q. So what do you see as coming up next in the continuous
evolution of the relationship between engagement, response
A. I believe these relationships will continue to evolve at a
rapid pace. I see three key areas of development:
1. Speed: faster attribution and eventually real time
2. Impressions: more granular data at the household level
including more robust retail and 3rd party viewing data
3. Engagement: better targeting of households including
identification of in-market users
Contact Cannella today to put these resources
to work for you.
Caitlin Haire — Vice President, Business Development
Trista Torrez — Vice President, Business Development
Los Angeles, CA
New York City, NY