It's not often chief scientists with multiple engineering degrees work with tools like wood or flashlights, but Kevin Haley makes a point of problem solving in all walks of life, at work and at his home in Austin, Texas. So, when a mischievous raccoon invading the attic was keeping his daughter up at night, Mr. Haley, chief scientist at TV and video ad platform Videology used the bothersome situation as an excuse to devise a contraption to ensure the varmint kept his distance.
"This raccoon was outsmarting me and it was driving me crazy," said Mr. Haley.
So, the next time Rocky attempted to enter the attic, he found himself sliding down a ramp, part of a Rube Goldberg-like setup contrived by Mr. Haley (pictured above). The raccoon proceeded to knock over a mirror aimed at a flashlight which triggered a sensor and in turn set off an alarm, alerting Mr. Haley to the creature's return. He waited for the raccoon to leave, then ran out to secure the entrance so he couldn't get back in.
The anecdote may seem disconnected from the high-tech world of digital ad data, but it's indicative of Mr. Haley's approach to data science. He strives to respond to actual difficulties rather than getting caught up in the numbers, which can result in creating solutions in search of problems.
"One of the things I have seen in my past is that data science and mathematicians can work a bit in isolation and a bit too much in theory," he said. "Let's make sure that what we're doing is solving a real problem…to make sure we're not doing math for the sake of it."
After achieving two engineering degrees from Cornell -- the first in Operations Research and Industrial Engineering followed by a degree in Mechanical Engineering, Mr. Haley delved into the arguably more glamorous world of cosmetics, crunching numbers at P&G to plan production of hundreds of eye-shadow shades for brands including Max Factor and Cover Girl.
These were supply-and-demand and inventory questions that he had to answer; therefore, not a complete departure from the video advertising inventory supply-and-demand questions Mr. Haley deals with today.
"That was where I started getting involved with the process of forecasting," he explained. At Videology he works alongside 12 mathematicians and data scientists. In the seven years he's been with the firm, Mr. Haley has witnessed a shift in how programmatic ad buying is defined.
"Historically, it used to be more thought of as the process of RTB and participating in an auction within an exchange," he said. "The idea of programmatic is kind of being raised a level up…. It's about automation and improved results."
Ad Age: Are there certain things you learned while working for P&G that informed your approach to your current work?
Mr. Haley: When planning the production of eye shadows, it's about the trade-offs of inventory-holding cost, in-store out of stocks and manufacturing efficiencies. These are in constant opposition. The key is to design a system that choreographs that dance and enables the ability to maximize overall yield of the system -- yield to buyers, sellers and builders. Advertising is no different. For example, a good system for managing ad execution needs to enable one to manage the trade-offs between campaign KPIs, reach and cost of the entire system, all demand, all supply and all data at once.
Ad Age: When you were studying engineering, did you ever think you'd end up working in the advertising industry?
Mr. Haley: Absolutely not! In fact, I assumed advertising was dominated by art and that science played little to no role in the industry. The purist engineer in me kind of scoffed at it. It was at P&G where I first saw the power of advertising. We felt we had the best products, but marketing largely drove the success of our sales. I had an engineering classmate and friend from Cornell that transitioned from production planning (same role as mine) to brand management. This was before ad tech took off, but I began to realize that there is science to good advertising, and there was an opportunity to bring more of it to the industry.
Ad Age: Do you think engineering students are aware of the new career opportunities in ad tech?
Mr. Haley: I think current engineering students are becoming more aware of opportunities in ad tech, but they get a biased exposure dominated by the behemoths in the industry that are concentrated in "digital" advertising, which likely comes across as "been there, done that." I don't think students realize that the opportunity exists to evolve an entire industry. Technology has changed so many industries over the past couple of decades. Young people today should realize that they can be part of the evolution of the TV advertising industry.
Ad Age: When evaluating new types of data to use for audience segmentation and targeting, what criteria do you use?
Mr. Haley: I always think of two questions: Does it matter and does it make a difference? Data is expensive to buy and to process. We are very dynamic in the use of data. If it is important to the system, we use it more. If it is not, we use it less. Part of our allocation process is to understand the data needs, their costs and therefore, what new data is needed and what existing data is not providing enough value. In a nutshell, we let the marketplace tell us what is important and let the allocation process evaluate it.
Ad Age: What kind of data are less or more valuable than they get credit for when it comes to your work?
Mr. Haley: Surrogates get too much value. True success gets too little value. Data that is easy and accessed by all, which tends to be the surrogates, fall into the over-valued bucket. The good data, the data that measures or drives actual results, is too often overlooked.
The natural tendency is for measurement to default to the least common denominator. That is usually some surrogate that all partners or platforms involved in a campaign have access to. Therefore, we feel it is important for advertisers and media companies to consolidate ad spend to a single platform so that measurement can be consistent. But that platform needs to be one powered by data and scientists that focus on the right data, not the easy data. Another driver of miss-valued data is when the cost and/or reach are not factored in. It is very easy to get great results to a few people. It is also very easy to get great results at a high cost. Viewing data in a silo provides misleading perceptions of value.