John Snyder is CEO of Grapeshot - a real-time contextual targeting specialist. Here he discusses the Grapeshot proposition, how it can help both publishers and advertisers - and why context and relevance are as critical as behavioural in delivering performance in the display channel.
Can you give some overview on the Grapeshot offering in the European market?
Grapeshot supplies real-time technology to publishers and agencies to segment online advertising inventory and place ads in relevant contexts for improved performance.
Grapeshot is fully compatible with the European market as our code is multilingual. Grapeshot’s algorithms use patterns and probability weights for words, so automatically understands words grouped together as concepts, without the need for a language-specific dictionary or set of semantic rules. We already have operations in French, Spanish, Italian, Dutch, Swedish and German, and only see our deployments in Europe progressing further…..
You are sometimes put in the category with semantic targeters like Crystal Semantics and Peer39. How do you differentiate from these vendors?
Contextual can sometimes be seen as such a broad term that too many companies with contextual capabilities are grouped together as one. Grapeshot offers a pure contextual targeting approach that determines a category by analyzing all of the words on the page. The fact that do the categorization in real time is something that places us at the top end of the spectrum. We can create individual channels, categorise pages and serve campaigns in real time without cookies. This makes Grapeshot unique.
Probably our most powerful differentiator is the fact that our technology analyses all the words on the page and only assigns a category as the page opens in real time. Instead of categorizing a page into a ‘hard category’ when it is first seen, our software transforms the profile of words we collect for each URL into a category in real-time. This has the advantage that new categories can be built and made live in minutes, helping advertisers find specific words and phrases – for example “iPhone 4S launch” - as a new display advertising segment. Grapeshot permits each advertising campaign to segment the inventory according to it’s own campaign needs and context, rather than simply using a standard set of generic categories like “mobile”.
How does the Grapeshot work? Do you categorize the URL alone? Can you give some insight into your real-time categorization?
Grapeshot code is fundamentally a search technology. When a URL is passed via an adtag or pixel, Grapeshot crawls the URL and pulls out all the words inside the page (not the words in the URL itself!). It then uses probability maths to discern the significant words, and weights each of them with relative importance. This we call the profile of the page – much like Google or any other search system would analyse pages.
In real-time, as the page is loaded in the browser by a consumer Grapeshot uses all the words in the URL’s profile to search for the most relevant categories defined by publisher teams (sell side) or agency teams (buy-side). The search and match is done within 6 milliseconds, and the most relevant channel categories are sent through to the adserver or exchange platform. The real-time assignment of category segments means it is very easy to introduce new segments – so for example ‘iPhone 4S launch’ can compete with ‘mobile’ with the advantage that some categories can be throttled up higher than others – so impression volumes can be elastically controlled to fit sell-side or buy-side demands.
Finally, the actual analysis of words on the page allows us to create hygiene channels that block segment categories being assigned to non-brand-friendly pages or unsuitable content. Category assignments are not arbitrary, nor declared on the basis that content is on a particular website: instead each page is categorised based on the real evidence of several words actually existing inside the page. There is full transparency and a complete audit trail as to why any page genuinely meets the criteria of the advertiser’s segment.
There has been a lot of talk about the effectiveness of behavioural targeting versus contextual. A recent Google study showed that contextual targeted ads were getting a much higher CTR across GDN. How important is context in the real-time space?
Never underestimate the power of an advert in the right place. The more relevant an ad is to the audience the more likely they are to engage in the brand and take away a positive brand experience.
Behavioural targeting dominates our advertising market today, as it has been too easy to drop cookies and track users as they view, for example, five ‘auto for sale’ pages in three days, and thus become de-facto members of the “auto-intending-to-buy” cookie pool. But if one of those “auto” cookies is viewing the latest news on Libya on the AOL portal, chasing them with an auto ad might be too out of context in that instance.
Cyberstalking a user after they visit a certain merchant website also has its detractions, not least as the industry tries to navigate the politics of opt-in and opt-out behavioural strategies.
We think context is king – it makes the ad relevant to the audience if they are already engaged in related content. This means combining the “auto” segment cookies with the content of the URL which they are reading there and then. Therefore at Grapeshot we can combine behavioural and contextual techniques into one unique blend. Also, when users are specifically opted out, we can still rely on contextual intelligence alone to impact the ad’s relevance.
Google has published its own figures for GDN, and the startling verdict is that whilst ‘traditional contextual’ based on the topic of the page (for example ‘mobile’ or ‘auto’) and classic behavioural user-retargeting methods score better than Run of Network; it is keyword contextual targeting that beats the lot. This is using keywords to target specific interests and context – for example the specifics of “iPhone 4S launch”, rather than only the general ‘mobile’ topic. Google has dominated the Search Advertising space with AdWords and AdSense, and we see Display Advertising going through the same technology revolution.
We are therefore very pleased at Grapeshot that we can use keyword context methods, in real-time, on platforms like Admeld, Rubicon and Appnexus.
Real-time bidding is about harnessing intelligence about the words inside the page of the URL, and the cookie id of the user loading that page, to make the most effective bids. Whilst many exchange participants predominantly use just cookie ids to determine relevance to bid; Grapeshot’s focus on keywords and page content broadens the volume of impressions one can bid upon, and uses contextual segmentation in real-time in a marketplace so biased today towards the user rather than the page context as well.
Why are traders buying exclusively on behavioural? Should buyers have a mixed strategy for executing real-time buys – with context and behavioural given equal importance?
Behavioural and contextual combined is in our view the best blend.
However many DSPs offer just a cookie-centric model where users are pooled into valuable cookie lists. When a user in the cookie pool is seen again on the exchange, the user is re-targeted with an old message or assumed to still be interested in the same category they were browsing days before. Based on arbitrary segmentation of the internet into topics like finance, fashion, auto, music sites, for example, the quality of the definitions that educate the cookie pool business rules can also be unclear. Still though, the cookie pool is the dominant single way in which traders almost universally buy on exchanges. The overlay of third-party data can improve the quality of the data for each cookie user, but the cookie centric model persists today, perhaps as a legacy of RightMedia and the first exchanges.
Page context can be equally helpful for building up cookie intelligence. Rather than retarget a user you have seen before, you can analyse the context of each page a user reads, and use keyword evidence of what the user has actual interests in. Creating new keyword segments such as “iPhone 4S launch” helps to quickly identify those users who have read three pages about this context in the last thirty minutes – thereby providing a more real-time definition of both the segment and the actual habits of an unknown cookie id. This prospect cookie building process is vital to augment the existing cookie pools already known to exchange members.
Grapeshot is now integrated with the AppNexus platform? What are your initial impressions of the AppNexus App Market – and how will it enable Grapeshot to deliver more value to buyers?
The AppNexus App market was revealed on the 10th Nov in Manhattan – it is a platform that allows technology companies to benefit from the worldwide base of console users. A company like Grapeshot can insert its application inside the AppNexus bidder interface, making it easy for bidders to create custom contextual segments, like ‘iPhone 4S launch” on the fly.
Grapeshot is indeed integrated on AppNexus. The demand for European inventory to be contextually segmented has been huge. Buyers are attracted to quality inventory from Microsoft and other tenant publishers, as well as the aggregation of Admeld and Rubicon bids.
Our ability to find custom contextual segments across Rubicon, Admeld and a large number of publishers using keyword contextual targeting helps bidders buy inventory, even when they do not have any intelligence on the cookie id offered in the same bid. So instead of paying high prices for a small proportion of the inventory where many bidders compete using (often shared) knowledge about the same cookie id; the bid landscape is extended to those undervalued bids where the cookie-only bidders have simply had to pack-up their bags and leave the arena.
Effective buying is about price and relevance. To bid in the majority of bids where a cookie id falls outside the known cookie pool has a competitive price value, especially when you are one of the few bidders equipped with keyword contextual insight of the page URL.
Our vision at Grapeshot is to offer different contextual lenses to each bidder. Therefore a ‘iPhone 4S launch” segment defined by one bidder is infact proprietary to that one bidder. No-one else sees that segment on AppNexus – only the one customer who created it. This means bidders do not need to use a generic set of topic segments like “mobile” which everyone else can unfortunately also use: instead they have a proprietary edge and can craft contextual segments to meet their own specific advertising campaign needs.
Is there any potential conflict of interest working for multiple agencies?
Two years ago Grapeshot started serving publishers on the buy-side with proprietary contextual segmentation tools. We have regional newspaper networks Johnston Press and Trinity Mirror both using our product, in a market where they both compete for advertising revenues. However each sales team is responding to specific advertiser needs, so that the contextual segments they create are unique and differentiated. One may craft an “X-factor” segment to catch a celebrity-friendly audience, whilst the other creates a “back to school” campaign for local retailers.
Therefore on AppNexus we see the opportunity for our publishers to surface their unique segments via SSP platforms like Rubicon and Admeld, with whom Grapeshot is also integrated.
But we also see the huge opportunity for agencies to craft custom segments using keyword context targeting methods that carve up the pool of aggregate inventory into elastic segments defined by multiple keywords that reflect their specific advertiser’s brief. Some industry veterans have described our Grapeshot capability as “AdWords for Display”: certainly our use of keywords and real-time categorisation gives us the flexibility for each agency to behave differently in the marketplace – so there is not conflict between publishers and agency trading desks, nor between agencies, as each is defining their own suites of content segments.
For Grapeshot, the same URL can by definition be interpreted differently by buyers seeking to fulfill different advertiser briefs. For example, a lifestyle article about the best gadgets to buy for Christmas could be identified as a ‘Christmas’ impression to one advertiser and a ‘gadgets’ impression to another. Again the flexibility of the Grapeshot categorization method delivers unique and proprietary results for our customers.
How does Grapeshot work on the sell-side? How would a publisher use Grapeshot?
Our client publisher base is quite extensive and is growing every month. Daily Mail, Johnston Press, Trinity Mirror Group, Reuters and Ad2One to name a few. Grapeshot is the largest contextual technology provider in the UK. We offer publishers the ability to understand their content outside the confines of their manual editorial categorizations. Grapeshot technology profiles the page and can assign it in real time a category according to the inventory that the advertising market is currently demanding. It will categorize unsegmented news and give value to previously low value inventory. Therefore publishers can find more inventory for popular channels like “technology” and “fashion” within their general news and user generated content.
Grapeshot supplies a console to ad operations teams to create new channel segments that either reflect existing premium channels, or a custom-made for a specific campaign. After creating a new channel in a matter of minutes, Grapeshot shows the operator the actual articles in which the ad would appear – giving confidence to both the ad-ops team, the sales person and the advertiser client.
As pages are loaded by consumers, Grapeshot ranks the channel categories which describe the content, giving the campaign manager the ability to deliver many different contextually relevant advertising campaigns to the same article.
Brand safety is also a top priority for all our publishers. Publishers need to ensure that their advertising partners’ brands, besides being associated with relevant content, is also shielded from potentially harmful material. Grapeshot offers this brand protection service as part of its keyword contextual targeting function.