Last month, Grapeshot partnered with Vice Media to help brands maximize ad effectiveness with a customized brand safety solution. Grapeshot’s Live Context Marketing Engine is a unique search and targeting platform that deploys machine learning to unlock value from data. The Audience Retargeting and Keyword Analytics platform for advertisers currently integrates with AppNexus, TradeDesk, AOL, and MediaMath, delivering simple, easy to understand datasets across 140 billion consumer touch points per day, in real-time using programmatic decision systems. We spoke to Grapeshot CEO John Snyder, to know more about search marketing analytics, programmatic, audience targeting and growing use of machine learning in advertising technology.
MTS: What is the inspiration behind conceiving Grapeshot, a search marketing technology platform?
John Snyder (John): Search and keywords are an everyday part of our lives. People search with intent. However, what we read, choose to watch on video, the media we consume or products we buy also have keywords behind them. So our idea is to use all the words we consume as individuals and consumer cohorts or personas, to make the digital insight for marketers richer and relevant.
MTS: How deep is Grapeshot into programmatic? Who are your customers and partners?
John: Grapeshot is used in 148 countries by 50,000+ brands for targeting and insights on DSP platforms like TradeDesk, MediaMath, AppNexus, Turn, and AOL. Publishers such as The Daily Mail, Vice Media and The Economist also use Grapeshot to package premium value inventory, based on all the words inside each and every page impression. The exciting dimension of programmatic is we unpack all the keywords inside the page, of some 3 million URLs sent to Grapeshot servers per second, with the ability of buyers to select their custom choice of keywords to meet specific brand targeting (or brand safety avoidance) needs.
MTS: What is the difference between programmatic solutions for DSP and SSP providers. Does this segmentation apply to your business?
John: SSP providers like Teads or AOL use Grapeshot to package premium inventory based on the context of the page in which an ad unit is being sold. Buyers on DSPs use the same Grapeshot technology to choose what inventory to buy, often using their own audience targeting data alongside Grapeshot’s understanding of the ‘live audience’ context of the ad impression. ‘Audience In Context’ delivers much greater relevance and performance for the buyer, than audience re-targeting alone. Without context, there is less relevance.
MTS: What are your expectations of SSPs and programmatic Ad Exchanges to justify your innovation?
John: Buying ads in brand-safe contexts which avoid toxic environments for brands are just as important as targeting live audiences based on what they are actually reading in that moment: whether the words on the page which the consumer is loading in their browser at that very moment or the mobile app they are using. Buyers and sellers are paying a small CPM fee for customized brand safety. Our business is growing at over 100% year on year, which tells me we are fulfilling a very clear need for precision targeting and brand protection.
MTS: Grapeshot offers Audience Segmentation as well as Predictive Analytics as a product. How did the company manage to bring the two together?
John: Context defines what the audience is doing right now. Just like you would want to watch the Super Bowl live, rather than wait days later to catch it on video; so a ‘live audience’ is an important set of consumers for any brand to reach. Audience segmentation can now be done for any choice of words – so the buyer defines their own sets of choice words and phrases directly resonating with the brand’s values, instead of audience segments defined by third-parties without the control and transparency to use choice words. A live audience is reachable instantly. However, social conversations are constantly alive with the flux of consumer interests. By trending the words out of these conversations, and listening to breaking news, we can run predictive algorithms to predict which words and phrases will become important over the next 12 to 72 hours, and use these to target live audiences on SSPs and DSPs. Grapeshot Predicts has quickly become a new use case for advertisers, keen to be in the conversation of the moment. As Usain Bolt won the 100 meters Olympics race, Nike ads were targeted to those live audiences reading the breaking story online on programmatic DSPs and got Nike’s brand message across to a live audience in a targeted, relevant way.
Recommended Read: Interview with Malcolm Cox, Chief Marketing Officer at Grapeshot
MTS: How does Grapeshot WordRankTM , which offers consumer insights based on adaptive AI/ML, help marketers create meaningful campaigns?
John: Often insight is a report with detailed spreadsheets and the need for data scientists to crunch a very large array of numbers. You have no doubt seen 10,000 lines of a Google Adwords report with rows and rows of keyword data. We think this is too static and cumbersome. Instead, Grapeshot’s WordRank technology actually adapts the weights of words in real-time – which means as the Usain Bolt story breaks at a given hour of a given day, the weights change dynamically.
This, in turn, gives us the power to trend and predict, and isolate the most significant words at that moment in time. So instead of long lists and detailed reports, we give the marketer an instant story. The billions of bits which make up ad campaign impressions can be distilled into an easy-to-read trending group of significant words – which in turn can buy ads, personalize a web experience, or switch out different creatives.
MTS: What are ‘Brand Moment’ and ‘User Experience’? How do you use programmatic to deliver both?
John: In the digital world, a consumer interacts with pages, ads, and content of all media types, each creating a digital touchpoint that we can log. Grapeshot turns those touchpoints into words, at the speed and volume necessary for processing a DSP programmatic bid-stream. Extremely low latency. These touchpoints are the building blocks for impacting a User Experience.
While the ad slot might be targeted with keywords, the real transition is to use logs of accumulated touchpoints to decide what sort of experience to deliver to the consumer at that next point in time. The context of a page, as they see an ad, is a simple momentary opportunity. Combining a near-term history of moments educates software systems as to what creative type, landing page, or personalized website experience to deliver after the ad slot is filled with an ad. The ad experience and the post-ad experience.
Brands are relying more and more on Data Management Platforms (DMPs) to capture these consumer touchpoints and use Grapeshot’s signal to color the variety of custom audience segments housed in the DMP, based on the keywords these consumers read. Contextual Audience data is more powerful to impact the User Experience than just age, gender, and familiar demographics. Brands want to know if their audience is interested in music festivals or beer festivals. The subtle difference is important when designing the optimal User Experience.
The Brand Moment is amplified when the content of the message resonates with the context of the page impression. The Brand starts to be part of the moment and not an interruption for the consumer. This is why Brands need to use programmatic platforms to rediscover the value of context, at scale. Retargeting audiences without due care and obsession for context leads to brands placed on offensive terrorist content as much as misplaced contexts which dilute the Brand Moment which creative teams strive for.
MTS: What should consumers expect when a Search Technology platform promises to deliver real-time analysis?
John: We all know how fast a Google search can be. Grapeshot is running search-based matching all the time. Just as what consumers read, changes all the time, so our underlying algorithms use the content of programmatic impressions to search out which marketers’ brands’ needs (targeting and brand safety criteria), and which branded content, actually match. It means any new marketing content or targeting strategy can immediately be matched to the changing pulse of the bid-stream and the internet.
The internet content in the bid-stream is searching out the most relevant marketer contexts. Reports can also be published in real-time, not just the activation of a programmatic media buy itself.
MTS: What are the factors that make Grapeshot’s programmatic work effectively with your partners and customers?
John: We have spent a lot of time on the principles of control and transparency. Programmatic offers a large volume of potential impressions in which to place a brand. But, Programmatic can also seemingly be a large black box of available audience and content that can be unwittingly dangerous to any one brand. We make sure customers can see all the keyword sets that sit behind any one of our segment definitions, and can see what inventory these words match too. They genuinely see the words and impressions that they are targeting or avoiding.
Likewise, we show what WordRank words and segments get assigned to any URL that the customer inspects. It means buyers and sellers have a real currency to trade with, evidenced by the words we collect for each ad impression’s context, the words, and contexts with which any customer engages with or those phrases the brand custodian can use to craft their own custom definitions of segments.
MTS: How do you see search technology changing/evolving further as the programmatic market grows big?
John: Search is almost always seen as the search box in which consumers type an average of 1.7 words. However, when search gets implicit, rather than explicit – the words on the page, or behind the video clip (i.e. metadata) start to be the search itself. For Grapeshot, these are not single words (like a bullet) but all the words on the page that fire out to discover other assets that match the very same sorts of words.
The overlap, in a fuzzy probabilistic calculation, is underpinned by weights behind words. We can now programmatically connect types of content with other content, using the power of WordRank algorithms that adjust their weights, or significance of words, in real-time too. This means the ability to connect marketers brand contexts with consumer contexts, at speed, and at scale.
MTS: What should marketers do to make their AI-assisted efforts more productive?
John: We have adaptive algorithms that change the weight of words, as I have said. AI more generally uses adaptive learning techniques to constantly adjust, refine and learn. Yet many AI systems turn an input sequence of data into a derivative output – sometimes without the simplicity and clarity of explaining why or how. They just do. I see marketers needing to answer the question why, rather than take AI outputs at face value, so keywords are an understandable dimension for not simply making connections, but explaining also why certain connections were made. A transparency to how AI learns if you like.
MTS: How can marketers leverage machine-learning capabilities to churn relevant content based on algorithmic search technology?
John: Algorithmic search technology is using many words and weights to find matches. It is not the documents in a Google search results list that are ranked, but the very words themselves. It means the query of 400+ words can also have the query terms themselves ranked in real-time. This gets exciting as a marketer’s branded content can itself now become the query, and the words within can also be adjusted for relative significance as a set of query terms – to discover the other assets it should link to or audience segments it should be distributed to.
Machine learning can use evidence of consumption or engagement to adjust the weights and hence refine the experience that one consumer or a cohort of consumers have with that one piece of branded content or creative. Marketer’s content is no longer the destination of a search, it initiates the search. This means the content finds other relevant content and audiences in an efficient and scaled way.
MTS: According to you, what are the biggest challenges within programmatic in 2017 and beyond?
John: The irony is we have buyers and sellers in an ad marketplace with a huge volume of impressions to trade. There is less friction to participate on account of automated processes, but often buyers and sellers still trade only with the people they already know. Specific agency teams working with certain publishers and websites. Yet, ‘sellers’ always need more ‘buyers’; and, buyers would like to explore a range of sellers for more premium programmatic buying. Direct deals have not scaled as much as predicted for 2017, so I see deal discovery as one of the necessary components.
At a keyword level, we see what buyers want to buy, and also the inventory and live audience each publisher also has to offer (in both RTB and PMP private marketplaces). Having coined the phrase ‘Tinder for Programmatic’, I am keen to see how new technologies like ours can auto-discover genuine overlaps between buyer needs and seller availability – defined by WordRank keywords of course!
MTS: How do you see newer AI-based and predictive technologies impacting marketing campaigns using search marketing platforms?
John: Search marketing is already a developed space in terms of SEO and SEM. People know the power of keywords and how to compete against their peers and the competition. Yet some of the CPA costs on display and mobile programmatic platforms have not been properly benchmarked against traditional SEM search metrics. What people are reading at the moment compared to searching with intent are indeed different, but consumers research before they buy, and they read more than the Google search result itself!
Once search teams embrace the power of keyword targeting on DSPs, AI will be a very useful way to optimize across these media planning environments, so that search itself becomes more multichannel than a Google search strategy. Search marketing can bridge more into the programmatic display and become more integrated with what consumers are reading and researching at a moment in time – not just searching – which can be an equally valuable way to reach consumers when classic SEM is so saturated and mature.
MTS: Location-based analytics and people-based marketing are critical to the adoption of programmatic. How do you see stack analytics boosting mobile targeting and retargeting in 2017?
John: Location is very key. We are creatures of habit, and the context of wanting to be entertained whilst on the subway, compared to busy at work, or coming up for a lunch break, or relaxing back at home in the evening – are very different locations and timeframes. This context is absolutely vital with the context of the content that I am actually engaged with.
Mobile is the ‘king of location’, but it is also such a private personal medium – so the marketer needs a ‘less is more’ mantra. Frequency capping and controlling the excesses of audience retargeting are vital to getting a non-interruptive, highly contextual experience which the consumer is ready to enjoy. Retarget with care, key off context and engage well with the consumer would be my advice. Adblockers have to be kept at bay!
MTS: What startups within AI/ML and programmatic are you watching/keen on right now?
John: AppNexus is not a start-up but I love the open platform of their technology stack where smaller companies can program the bidding logic. It means you can use AI routines to decide what price, and with targeting variables, you can target ad impressions. So programmatic is now programmable at the bid level meaning a diverse range of different campaign line items can be shrunk into programmable logic fuelled by AI.
John: We all need spreadsheets to get into detail, but the tools of data scientists normally engage with reams of spreadsheet data that would overwhelm most of us. Powerpoint is for the story, a simple message with 3 bullets. Marketers need the top-level story – not only to communicate with colleagues and partners in an easy way but also to show what the focus is; what actually matters. We strive to condense out of billions of bits, the summary WordRank sets of trending words and phrases – the topical clusters of insight to avoid the very need to reach for Excel or Tableau and going for the deep-dive.
Real-time insight needs to be digestible and actionable. If it is locked inside large data sets that need the data scientist then you have not given the marketer simple to use, focused insights that capture the essence of the story. The data scientist will also introduce a time delay to extract that insight in a manual way, again removing the edge of having live insight in the moment. Live Contextual Insights need to real-time and simple. Use the data scientist as a team member for the necessary detail, but take command of the evolving story and focus on the actions would be my piece.
MTS: Finally, how do you see Programmatic impacting Brand Safety and Audience enhancement in the future?
John: I met with the CEOs of Integral, Moat and DoubleVerify, a couple of years back, and the fear of fraud, non-viewable ads, and non-human traffic was rife in the adtech industry. The call for Transparency and Control from top brands like P&G resonate with the efficient use of marketing dollars and reaching real consumers who actually saw the ad; not bots. However, the real value of these fraud and safety filters coupled with isolating just those ads actually viewed is.
However, the real value of these fraud and safety filters coupled with isolating just those ads actually viewed makes sure that all the machine learning we apply to programmatic logs is applied to just the human consumers who saw ads; not the bots and the non-viewed ads.
The subset of pure impressions is the very foundation for building audience segments based on the contexts that consumers actually read. If you build upon impressions initiated by robots or not actually viewed, then you have introduced noise into your DMP, into your insight engine. So it is best to cleanse the signal, not to just impact the efficiency of media buying, but also the purity of the insights you garner from the digital trace of programmatic advertising itself.
MTS: Tag the one Martech/AdTech investor/CEO who you would like to see featured at our TechByte Series:
John: I would like to see Jonah Goodhart from MOAT featured on Tech Bytes.
Impressive to sell an analytics company for $850m when in the programmatic universe they neither decision prebid, nor fully blocked ads post-buy. They report with analytics of engagement. And, apart from separating out the fraud, the non-viewed impressions, they focus on engagement of that consumer within the single ad impression, looking at hover rate, dwell time, and the like.
I should mention Moat can also report on Grapeshot contexts and Nielsen demographics – but for me, I would like to hear why Analytics matters so much more than just efficient buying.
MTS: Thank you, John, for speaking to us. We look forward to having you very soon again at MTS for more insights on MarTech.
Stay tuned for more on business insights on video ad tech, programmatic and header-bidding technology market.
To participate in our Tech Bytes program, email us at firstname.lastname@example.org