December 1, 2015
When it comes to monitoring and analyzing media advertising, the future is now. No longer is it good enough for advertisers to know when their ads were broadcast and placed; now, thanks to Veritone Media's Cognitive Media Platform, they can find out how they were delivered or "sold" by the host. Ryan Steelberg, named one of the "50 Most Powerful People in Radio," previously was the co-founder of AdForce, 2CAN Media and dMarc Broadcasting before launching Veritone Media. Here, he discusses the evolution of marketing technology and what's on the horizon for radio advertising.
What were you doing before Veritone Media ... and what made you decide to do this?
We've been in the marketing tech business since 1994. My brother Chad and I co-founded a business called AdForce; it was one of the first and largest digital ad tech platforms. We worked with a many of the most influential websites and networks between 1994-2001, including Yahoo!, AOL, Netscape, Geocities and others. We then took the company public and sold it, after which we started dMarc Broadcasting, providing ad insertion and clearance for radio and TV stations. We sold that business to Google in 2006.
Chad and I have primarily been involved with programmatic SaaS platforms throughout my career, with an acute focus on ad management, delivery, tracking and reporting. An important differentiation between these other ventures and Veritone is we now have the technology to unlock the metadata within audio and video files. This advancement allowed us to develop a technology platform that unlocks insight into both private and public audio and video data files, not just for advertising and media but also for the world in general.
That being said, the world of advertising is changing - dramatically. Consumer time and attention is rapidly shifting from live viewing to on-demand consumption and from long form to short form, with broad-reaching implications for advertisers and broadcasters alike. What we have seen over the years is the exponential growth of data and the trillions of impressions that are now available to advertisers.
It is a far cry from what we've been dealing with and have been servicing for the last 15-plus years. These changes have resulted in more than a few challenges:
- The sheer increase in the number of ad avails people see in the U.S., from a few hundred to 5,000 a day, and the magnitude of clutter that has moved into TV, the Internet and the mobile device is astronomical. Managing this amount of data is challenge number one.
- Challenge number two is that people are becoming more accustomed to ignoring ads, or skipping them altogether. Recent ad blocking technology has become very accessible and, as the consumer base shifts their time from radio and linear TV watching to fragmented media such as Internet video, on demand and mobile viewing, the ad model of today simply can't keep up.
Our idea started with a radio problem. One of our largest agency clients manages radio, representing over 50 clients from Uber, Dollar Shave Club and Draft Kings. They came to us with a simple problem: While they had an "okay" system to verify when traditional over-air spots cleared, they had no visibility into what was actually said when the ad was read, for live reads and organic mentions.
For instance, when Rush Limbaugh or Colin Cowherd do a live spot ad read, advertisers wanted to verify if, and when, the talent actually did the spot without waiting two to three weeks for results. Additionally, it was becoming increasingly important to understand the sentiment or tone of the discussion.
Now, through Veritone Media, we are able to provide our clients with premium media buying services that not only provide near real-time analysis for verification, but also audience insights and live-read sentiment.
What sets us apart from any other technology today is our Cognitive Media Platform, or CMP. We initially developed it to process audio feeds. Today, within minutes after receiving the feed from the station, either over-the-air, satellite, or through an http stream, we are able to generate fast, reliable, high-integrity insight into every aspect of in-program content.
Essentially, the CMP takes media broadcasts and turns them into rich and actionable metadata. We've built a suite of tools to turn this data into intelligence that provides solutions for advertisers. What's more, media groups can now take this data and use it in real-life business executions and operations for ad-compliant verification, as well as content search, discovery and distribution.
How did you design the CMP?
We designed the CMP as a cloud-based open framework. There are literally thousands of cognitive engines (CEs) out there, from lie detection to sentiment extraction to voice and facial recognition. Unfortunately, the failure was that most of them processed data in proprietary silos, making it difficult to integrate and activate them efficiently. It would be as if Google built a search engine that only looked for ads on websites compared to indexing all of the content and objects across the Web. The end result would provide a very shallow and limited search experience.
What we've built is an ecosystem that ingests, analyzes and activates private and public media to provide actionable insights into content like never before. This transformation from data to understanding is achieved by providing accessibility to a variety of cognitive engines that act in concert with one another. The CMP is essentially the conductor, controlling cognitive engines such as transcription, facial recognition, speaker and object identification, tonal analysis and audience measurement. Customers, independently or through a professional services relationship with Veritone, select the engine that will provide the information that matters most to their business to seamlessly display the results in near-real time. This index and metadata can also be consumed via a suite of APIs.
Exactly how does that work?
Veritone's cognitive processes start with the ingestion of media from either file upload, download or pull and push stream. The media is then introduced to the CMP as unstructured data. Let's say the client is Dollar Shave Club. They are running spots on Westwood One affiliates. The software ingests all the affiliate audio and video feeds and, by using a series of patented algorithms, it specifically looks for a combination of key words and phrases to provide an unprecedented view into the media, identifying, classifying and making searchable any mention of the Dollar Shave Club.
We've also built a fingerprinting engine for pre-produced spots, to provide complete 360-degree visibility. This allows clients to include comments and ratings -- a "Rate it, Comment it, Share it" -- and relay that information in report form every day.
Is your software able to catch every mention and live read on radio ... and other media?
We're not at 100% yet, but we are getting very close. Radio is a more controlled environment; there aren't tens of thousands of national hosts, so it's easier to control quality and execution. But when you touch on the huge, unlimited content and advertising opportunities such as you have in YouTube, for instance, the scale is significantly larger. The problem is that to get the same reach on YouTube, you have to make buys that spans hundreds or even thousands of channels. There's no way a human being can source ads on thousands of videos and analyze the attributes and quality of them. We believe we have also solved this problem.
So how does your software analyze the quality of a live read?
It starts with the performance. The personality is supposed to follow a script that has the specs they are required to use, such as saying certain key words and a phone number or URL. It depends on how good the personality is at delivering the information. It may be a bad script template that's saved by a brilliant delivery. There are live reads that have done incredibly well with the audience despite the poor copy. On the flipside -- and we have all heard this on the radio -- is a brilliantly written script but the delivery is terrible because the host doesn't execute, or conveys the sense that he or she isn't truly supporting or endorsing the sponsor.
Our goal today is to bring those differences to the surface and make them immediately actionable, by being proactive and saying, "These specific live reads failed on this program or station, for these reasons, so let's fix it immediately and improve the media buy."
How can software decipher if the personality is being sarcastic or not feeling it?
There are multiple ways; it depends on, specifically, being able to hone in on positive and negative connotations. It could be 100% text-based analysis to see if they're using negative or inflammatory words in a discussion.
Advertisers today want to know why their ads are, or are not, working. It's not just whether the ad cleared; it's also about the mindset or sentiment of the reading. It is also important to determine what the radio or TV host was just talking about. Say they were covering a terrible car accident; that wouldn't necessarily be the best spot to run a new diaper ad and expect it to do well. Rather, a better choice in this instance would be to insert a car insurance or legal representation ad.
And are YouTube channels the next platform for your software to conquer?
After successfully dealing with broadcast media, we're expanding our tech stack to process YouTube video channels at a scale that will allow us to bring a whole new level of insight and intelligence inside these channels. The ultimate goal is to target ads more effectively in thousands of YouTube channels and provide intelligent new buying opportunities. For instance, if an advertiser wishes to do branded product integrations, within YouTube and digital video, we can cost-effectively verify them and report back proven attribution in near-real time.
Our recent agreement with multichannel network, Maker Studios, the global leader in short-form video and the largest influencer network across platforms worldwide, provides our clients with access to 12 billion monthly views and 800 million YouTube subscribers. This access allows advertisers to tap creators from Maker Studios' huge network for native ad campaigns, while providing Maker with analytics and tools that provide unique insights into content and performance.
Partnerships with other multichannel networks include Car Throttle (#1 auto) and The Young Turks (#1 News, Talk, Sports).
So your platform is fully operational?
Yes, we have been using the platform with our advertising clients for the past 18 months, across radio and other audio formats. Additionally, earlier this year, we started licensing our platform to broadcast groups and networks directly, including Westwood One, Greater Media and others. Our goal is to greatly automate the process of producing and delivering live reads and branded integrations into both audio and video platforms. We're planning on offering clients the chance to buy across thousands of audio and video programs, with completely visibility and intelligence.
On the technology front, we have greatly expanded the scale and power of our CMP (Cognitive Media Platform). We have increased our suite of Cognitive Engines, as well as our cloud capacity and processing efficiency. I would be remiss if I didn't mention the strong partners that we have aligned ourselves with, including Microsoft and the Azure team.
How do you price your service offerings, which include access to your technology platform, CMP?
We currently offer two service models, each with its own pricing structure. For our ad media services business (Uber, Regus, Dollar Shave Club, etc.), we charge a commission against billings/spend similar to traditional agencies. But we bundle our proprietary technology platform with this commission plan, providing great value to each of our advertising clients.
Additionally, we offer access and utilization to our CMP platform, via SaaS license, to third-party agencies, broadcasters, networks and others. We establish the pricing against the volume (hours) of audio and video content to be processed, in addition to the number of seats or accounts that are needed for each client. This is a traditional SaaS structure, which provides flexibility and scale for our clients.
Is ad insertion and monitoring on podcasting the next frontier?
Actually, we already do it for podcasting. We leverage our technology for many of our direct podcasts buys (for our direct advertising clients), with spend in the volume of several millions of dollars. It's very effective to analyze and process what the podcast hosts are talking about, which really unlocks more value and avail opportunities. We pass that intelligence on to advertising clients for attribution modeling and ROI analysis. We continue to be very bullish on the podcast marketplace and will continue to invest heavily in this vertical.
Considering how much your technology has advanced in the past few years, where do you see Veritone Media five years from now?
It starts with us continuing to hire the best. We currently are 80+ person shop, but I wish we had 300 people. We're continuing to invest in great sales people, media buyers, software engineers and operation professionals. Outside of operations and staffing, what I'm most excited about is that this technology shouldn't just marginally improve advertising but be used to fundamentally transform legacy audio, video (TV and radio included), while bridging the gap of consumption. Whether it's listening to UCLA football, consuming TV or following a trending topic, I want rich content brought back to me, packaged when I want to consume it. I have a high level of confidence that in five years, this technology and others like ours will radically transform how traditional media is accessed, consumed and monetized.
Check out Veritone's technology by clicking: http://veritonemedia.com/videos