Ilya Vedrashko is fascinated by digital interactions and what they tell us about people. Though his work heading a team of consumer-intelligence analysts at Hill Holliday involves what some think of as social listening, he prefers to call what he does "computational anthropology."
"In this line of work, we analyze traces of people's online activities to understand their relationship with the world around them, including, but not limited to, brands. In one project we did in mid-2000s, we compiled and analyzed hundreds of pictures people had posted on Flickr to understand how people do laundry, for example," he said.
The standard approach to social listening -- observing posts in social media about specific brands -- is flawed, said Mr. Vedrashko, SVP, director-consumer intelligence at Hill Holliday.
"The automated sentiment-analysis solutions that we've considered for our work have certain shortcomings that right now prevent the solutions from being very useful to us, although they may be perfectly adequate in other contexts," he said.
So, what's wrong with them? For one thing, all posts are not necessary generated by real people. "In samples of tweets we have recently analyzed, for every one tweet posted by a user that we categorized as human we counted two tweets posted by either a bot -- such as an automated account broadcasting news -- or an organization," said Mr. Vedrashko. "When we look at a graph showing a change in sentiment on a particular subject, it is important for us to know who this graph represents. Are they our customers?"
How to categorize posts is another problem. Not only is labeling posts as positive or negative often a subjective decision, it may not be the best way to understand sentiment. "Rather than counting the number of negative posts about a brand in general and then peek into a word cloud trying to decipher the specifics, we prefer to count the number of posts regarding a particular customer-service issue, for example," said Mr. Vedrashko. "So in our reports we produce what we call 'topic maps' and track the change of mentions of a set of particular topics from one period to another."
Mr. Vedrashko leads a group of nine consumer- and business-intelligence staffers from Hill Holliday's Boston office. The team focuses on "cognitive profiling," attempting to understand how people make decisions, then segmenting them based on those determinations. They gather digital and offline data to produce subscription-based reports on brands and help others at the agency, such as creatives, understand their target audiences. He has a BA in business administration and international relations from American University in Bulgaria and did graduate work in the philosophy of virtual culture at Bulgaria's Sofia University. Then, Mr. Vedrashko received his masters of science degree in comparative media studies from MIT in 2006.
His team is in the process of developing a product for gauging social interactions in a more sophisticated way than what's available today. "It's very manual but we're trying to create efficiencies …. I think this is the next step in the social-measurement space," he said.
"There's a lot of money and effort that goes into that space," he continued. "I would like this money to go toward understanding people not brand mentions."
Ad Age: You mentioned cognitive profiling. How do you define that?
Mr. Vedrashko: Cognitive profiling in our group refers to understanding people's cognitive processes, personality aspects and decision-making strategies with regard to a product, service or media programming including advertising. This approach differs from -- and augments -- profiling people by their demographic characteristics or by their "psychographics" -- beliefs and behaviors that transpire from their survey responses.
For example, individuals differ in the extent to which "they believe they can control events that affect them" -- the so-called locus of control (this is a quote from the Wikipedia definition). Understanding this difference would suggest that advertising for a gym or a dieting regimen should vary to appeal to people on the different ends of the scale. This practice is particularly useful in the early stages of ideation, as well as in campaign diagnostics.
Ad Age: You were doing social listening before it was really called that. Describe that early work.
Mr. Vedrashko: My own work in the field started around 1999 with a feature I wrote for a school newspaper about students who spent a lot of time playing MUDs (Multi-User Dimensions or Multi-User Domains, roughly, online text-based "World of Warcraft" types of games). While writing my thesis at MIT, I studied how players incorporated fictional and real brands into their lives in virtual worlds such as "Second Life."
What is now known as "social listening" was first commercialized in the mid-1990s by a company that would monitor web pages and online bulletin boards for negative posts about its clients. I think my first public presentation about why businesses should analyze forum posts was around 2000 or 2001; around that time we built a simple program that would crawl forums looking for posts on a certain subject of interest to our company.
Ad Age: How does data come into play when developing your subscription reports for client brands?
Mr. Vedrashko: Our process is similar to data journalism; we look for stories in the datasets that we buy, aggregate or produce ourselves. The reports are editorially driven; each report is overseen by one or two editors who are intimately familiar with the client's business and are actively looking for ideas to develop and datasets to explore.
Ad Age: What are some ways your work affects campaign creative?
Mr. Vedrashko: Our work plugs into many advertising processes, from demand evaluation to planning to optimization to analytics and diagnostics. Within creative development, we generally provide recommendations for approaches that in our opinion have the highest odds of achieving a desired outcome. We have provided research support for a lot of Hill's work you see out in the wild. We would love to be able to point at a TV spot and say "I did this!" but most of what we do is done too far upstream. A fair amount of our work is preventive; that is, we advise on what to avoid doing. As you can imagine, the things that did not happen are hard to take credit for.
