Just saw this release by PubMatic. Is this their own admission that RTB obviates the need for the publisher-side optimizers like PubMatic, Rubicon, and AdMeld?
http://www.pubmatic.com/press-release-10-07-2010
From the release: Publishers generated significantly higher revenue from RTB campaigns in comparison to Non-RTB campaigns, across all advertising verticals, by an average of 64%.
Just read this article from Ad.com about their attribution management scheme: http://www.clickz.com/clickz/column/1740414/views-view-tracking
We ran into this attribution 'method' over the course of a campaign where we competed directly against Ad.com in the Q2/Q3 timeframe. The client asked us to provide last-click attrbition reports, and kept telling us that Ad.com was outperforming us by a wide margin. After we got access to the Ad.com reports to compare results, we realized that Ad.com takes full credit for view-through conversions, and the client that we were working with had:
1. no idea that the conversions provided by Ad.com included view-through and click-through conversions. They were under the impression that the reports provided by AOL were all based on last click attribution.
2. 90%+ of the conversions claimed by AOL were view through conversions.
It indicates at least a few problem areas:
a. The digital marketing community overall is still not very sophisticated about attribution management, and it will take a lot of educational effort from the vendors to get the community up to speed. This can be a big hurdle in rolling out more sophisticated attribution models. The only route to wide adoption will be to figure out a tracking & attribution system that can be a simple drop-in, like Google Analytics, and the marketer only has to care about acting on the final reports provided.
b. The agencies / service providers have a lot to lose from removing the obfuscation around display advertising metrics.
There is a lot of discussion around attribution for brand campaigns, but I do not see the same kind of dicussion around attribution management for direct response campaigns. Search escaped the scrutiny because search provided the vast majority of the last clicks. Display is a different beast, and the industry needs to develop models for assigning value to the entire path for DR campaigns. We are still in the early days of making display advertising accountable.
ClearSaleing seems to be developing interesting approaches to improving the attribution for display campaigns, but it sounds complex and needs to be customized to every situation. That does not appear to be a scalable solution for small-to-mid sized advertisers.
A couple of interesting ideas surfaced during a discussion about pricing for high quality inventory on the exchange.
My hypothesis is that since the exchange is doing a higher level audience aggregation than any publisher, it is reducing the value of the audience aggregation done by the publisher. When using advertiser or 3rd party data to aggregate audiences on the exchange, the value of audience aggregtion is provided by the exchange and the data providers. The publisher's value then reduces to providing content that generates high audience engagement -- which means the inventory should really be valued much lower than the typical direct sales pricing. In turn, this translates to schemes like unbranded inventory on the exchange to prevent channel conflict.
I believe that ad inventory that has good placement on high quality content pages is limited. For example, on the Google exchange only about 10%-12% of the inventory is identified as above the fold, while about 40% of the inventory placement is unknown. If we extrapolate from the known above the fold vs. below the fold distribution, perhaps about 20% of all inventory on the exchange is above the fold, and perhaps 25% of that inventory is on high quality publishers. That is a fairly small segment of the total inventory, and it is essential for the exchanges to develop schemes to help sustain appropriate pricing for that inventory.
Reserve price by impression
The simplest option is to put in a reserve price by placement for ad impressions. A branded, above the fold 300x250 placement may have a reserve price of $5 CPM, while the same location, if unbranded, may have a reserve price of $3 CPM. Reserve pricing for impressions is already in place for the exchanges that matter.
Reserve price by audience segment
Another option would be to support reserve price by audience segment. If the publisher is creating content that aggregates certain segments, then it makes sense to place a premium on tapping that audience in the current audience oriented buying environment. The challenge would be in managing the segments since there are no standards for audience segments, and there is unlikely to be except at the high level demographic characteristics. Also the advertiser may value a user for reasons completely unrelated to why the publisher values the user. For example, cafemom.com may value a segment of their audience as moms between 25 yrs and 35 yrs. The advertiser may value some members of that audience because they are interested in luxury travel.
This mismatch should not be a problem. The publisher will set a reserve based on what the publisher believes is the media value of that audience segment. If the publisher believes moms in the 25-35 age group are worth $8 CPM for the CPG companies, then that should be the reserve price set by the publisher. If Carnival Cruise decides that some members of that audience are worth more to them, then they should buy impressions. If not, they should try to find that audience somewhere else.
If there is enough market demand for that audience segment at the reserve price, the publisher's inventory will sell out. If not, a different strategy would have to be developed to get the most revenues out of the excess inventory. For example, the same impression could be put up for multiple auctions with different information about the user. One auction could be for [mom, 25-35], which another could be for [female, 25-35]. There will have to be two levels of auctions to determine the final winning bid – creating a more complex exchange.
Reserve price by audience segment and placement
It would appear that the right approach for the publishers would be to set reserve prices using a combination of placement and audience segment information. This scheme puts the onus on the publishers to provide more data about each auction -- both placement and audience. I find it unusual that Google has to use technology to infer that a placement is above the fold or below the fold when the publisher could easily provide that information -- at least for the publishers aspiring to get a high price for their inventory. This lack of data from the publishers has crceated a slew of ad verification vendors who claim to provide information about the visibility of the placement and brand safety. Information regarding visibility and brand safety should come direct from the publisher. An infrastructure provider like Google should only have to audit this data periodically to make sure that the publisher is compliant with the requirements.
Having to provide the audience data would create work for the publishers. They would have to work with data providers like Targus and Exelate to segment their audience, and provide that data for every auction. There will be technical challenges in moving that additional data through the exchanges handling billions of auctions per day. And there has to be some method of auditing the quality of data provided the publishers. None of this is trivial, and the value has to be significantly higher than the work involved to make it worthwhile.
Will media buying on the exchange move in this direction?, and provide that data for every auction. There may also be technical challenges in adding a bunch of data to the billions of auction requests flowing through the eexchanges.
Over the last few months, I have come to understand the display ecosystem a little better, specifically with an exchange oriented viewpoint. Here are some random thoughts on the issues that need to be resolved in order to make the ecosystem thrive.
Exchanges & quality inventory
The exchanges have a perception problem on their hands. They have been cast as the bottom feeders of the display ecosystem – partly for some perfectly valid reasons, partly because of limitations of the early efforts, and partly by the incumbents who see the threat exchanges pose to the old way of doing media buying.
Exchange CPMs apparently average around $1 on average, compared to $8 - $20 the higher quality sites can command for their direct sales. This has led to the creation of branded and unbranded inventory on some of the exchanges, with some premium sites preferring to offer their inventory as unbranded to avoid channel conflict with their direct sales.
Clearly the industry needs to evolve to resolve this problem. Instead of diluting the value of their sites by hiding the brand, what are some possible scenarios for the premium publishers to capture the value that they create?
Reserve pricing
One simple option is for the premium publishers to set a reserve price for their branded inventory on the exchange. If the advertiser cares about being on quality publisher sites, they will instruct their agencies to get the premium inventory. This seems like a logical choice for brand advertising. The challenge I see with this approach is the historical market structure that developed before there was a liquid marketplace. Audience aggregation was either done by the publisher, or by the ad network. Premium inventory was purchased directly, and ad networks were used for cheap inventory. The ad networks were opaque and cheap low quality inventory was expected. In addition, the typical ad network business model is an arbitrage model maximize the margins by buying inventory at the cheapest price. They have no incentive to pay extra to buy premium inventory in that model. Now the ad networks are utilizing the exchanges to source cheap inventory, and in turn exchanges are getting labeled as a source of cheap inventory.
There are two possible solutions to change the current state of the market:
1) force ad networks to shift to a percentage of spend model, and buy the kind of inventory specified by the agency,
2) provide tools that makes it easy for agencies to buy premium inventory from the exchanges, which may or may not be the same trading desks that the big agencies are building today.
It is hard to see the justification of ad networks to perform this task. I think it is more likely that the exchanges like AdX will provide these tools that are tightly integrated with the exchange, and make it easy for the agencies to aggregate premium inventory on the exchange. When they buy premium inventory suitable for brand advertising, they will be willing to pay a price that is fair for premium content, and reduces the channel conflict for the premium publishers.
One related issue is how high does the exchange price have to go before publishers do not feel threatened about channel conflict? And what would be a fair price for the premium inventory? The sales & administrative costs will be significantly less for the inventory sold on the exchange, so the publisher could afford to sell the inventory at a lower price and make the same net income. But is the inventory worth as much on the exchange as the direct buys? This is a difficult issue. In the direct buy model, a huge part of the value is the audience the publisher aggregates. On an exchange, additional data is being used to aggregate a specific audience across multiple publishers. Are the advertisers willing to pay a higher CPM for this aggregation such that the publishers still make the same amount of money? Even if the industry moves in this direction, it remains to be seen how the pricing and revenue distribution works out.
Last weekend I was Fresno with my family, relying completely on my trusty Droid to conduct life. Living in the bay area, I got used to relying on Google maps without a second thought. So I tried to find restaurants in Fresno using Google. For saturday lunch, I was looking for a Mexican restaurant, and picked one randomly from the ones google suggested. Restaurant name: Rocio’s Mexican Restaurant. When we drove to the destination following their directions, it turned out to be a lousy part of town, and there was no restaurant there. Fortunately it was during the day. I would have hated to end up in that part of the town after dark.
Same story repeated for Sunday lunch. This time we looked for a Chinese restaurant, and followed the directions to their top result for the search. Restaurant name: Golden Dynasty. Result: we ended up bang in the middle of a residential neighborhood with no restaurant in sight. Needless to say that my wife has lost all confidence in the Droid’s navigational system, although she does not equate it to google yet. It would be an interesting brand experiment to see what her reaction is if she maps the source of the poor results to google.
In contrast, I looked up restaurant information on Yelp. The usage mode is of course a little bit different, so it’s not an apples to apples comparison -- but they got it right. Yelp’s data is clearly superior to Google’s. Perhaps Google had a good reason to try to acquire Yelp. I think Microsoft should try to take Yelp away from Google’s grasp while they are still independent. That may give their local search a leg up over Google in terms of data quality. Of course I do not know if Yelp has a long term dedicated set of contributors that will continue to produce high quality data.
To make things worse, I was trying to meet up with a friend on Tuesday at the Borders in Milpitas just off McCarthy Blvd. I followed the GPS directions from work – about 5 miles away from the destination. Towards the end of the trip, Google guided me in the wrong direction. Luckily the store was already in sight. Perplexed, I drove past the store and tried the nav system from the opposite direction. Same result – Google once again sent me off in the wrong direction. Now my faith in the google maps is sufficiently shaken where I would probably verify their directions with another source before heading out. Kind of defeats one of the two main reasons I got the Droid – the built-in turn-by-turn directions.
These are certainly difficult search problems, and I doubt that this is an area that a startup can explore. But someone may come up with a better approach to tackling the problem.
To local data providers: Based on my Fresno experience, it is important to know something about the neighborhood for out-of-towners. It would be a useful additional field of information.
|
Came across this interesting article: http://www.sfgate.com/cgi-bin/article.cgi?f=/c/a/2010/01/10/BUQP1BEDSM.DTL There are a couple of interesting points raised by this article. I will ignore the search aspects discussed since I am now out of search. Here are two points with significant implications for online advertising. First one that caught my eye, though to be honest I have been sensitized to this point by somebody else recently, is this one: "Today, you really have to take much more seriously what captures attention" This is so true when it comes to advertising. In today’s attention deficit media consumption, it is hard to assign any value to spend that does not capture the users’ attention. How that is measured remains a huge challenge. CTR is one proxy. Aided or unaided recall is another measure. Perhaps someday monitors will be equipped with multiple cameras that report eye location to infer user engagement. The current trend of correlating ad view-through to subsequent search activity seems a bit tenuous, particularly if not backed up by a recall study. The second part is a damning piece of research, if true: The research, conducted in partnership with an undisclosed national retailer, sought to accurately measure the impact of Internet display advertising across online and offline sales, by tracking people who had registered with both Yahoo and the store. The research found an approximately 5 percent increase in spending among those who had seen the ads - with 93 percent of those sales occurring in stores. The potentially worrisome thing, however, was that among those under 40, the percentage was nearly zero. That could reflect the unpopularity of the particular retailer among that demographic. Or it could underscore a growing immunity to display advertising among the Web-savvy younger generation. The conclusion would seem to imply that display advertising is doing a poor job of capturing attention of the under-40 group. Either the marketers need to complete re-think advertising formats / creatives to engage this group, or a different methodology is needed to assess the impact of advertising on this group. It is entirely possible that this age group does not follow up with offline purchases, or this retailer does not have a strong share of wallet of this age group. I think it is hard to conclude that display advertising does not have impact on this age group based such a narrow study. The point remains that perhaps different metrics and methodologies are needed to measure impact on different demographics. The people that are forecasting growth of advertising dollars in the display world better take note that advertising spend should be all about capturing user attention, and somehow translating that to commerce. The dollars are unlikely to grow just because people are spending more time online unless far better analytics are developed. |
Several premium publishers have been, or expressed the plans to, avoiding putting their inventory up on exchanges because of channel conflict. The belief is that by not putting their remnant inventory on the exchanges, their direct sales force can continue to charge a premium CPM for the impressions.
There is no question that historically premium brand advertisers have preferred to stay with premium sites only for their brand advertising. Now, in a world that is highly fragmented in terms of media consumption, and with surplus impressions, advertisers are shifting to buying audiences wherever they are. I find it hard to believe that more than a handful of publishers – perhaps a couple of hundred sites -- can sustain their premium CPM pricing whether they withold their inventory from exchanges or not. So I decided to do a simple table to see how much revenues publishers may be foregoing to avoid channel conflict, and to sustain their pricing.
Revenue percentage from direct sales
Direct sales multiple over exchange prices
Sell-through 2x 3x 4x 5x 10x 15x 20x
10% 18.2% 25.0% 30.8% 35.7% 52.6% 62.5% 69.0%
20% 33.3% 42.9% 50.0% 55.6% 71.4% 78.9% 83.3%
30% 46.2% 56.3% 63.2% 68.2% 81.1% 86.5% 89.6%
40% 57.1% 66.7% 72.7% 76.9% 87.0% 90.9% 93.0%
50% 66.7% 75.0% 80.0% 83.3% 90.9% 93.8% 95.2%
75% 85.7% 90.0% 92.3% 93.8% 96.8% 97.8% 98.4%
If we assume that the typical premium-to-exchange price multiple is in the 5x range, then a site would need close to 75% premium sell through to avoid a significant revenue drop by shunning exchanges. Though I hear the typical sell through is in the 30%-50% range, depending on the vertical and the publisher.
The other scenario to compare the above revenue table with -- what would be the drop in total revenue if putting the inventory on exchanges did indeed depress premium pricing -- is a bit more complicated to put in a simple table. It needs to modelled for each multiple. So I decided to do the model for the 5x multiple, which is a gut feel average number based on what I have been seeing in media pricing so far, i.e. if the exchange price is around $2 CPM, then the premium price would be typically around $10 CPM.
Revenue from lower pricing due to selling on exchanges
(For premium to exchange pricing ratio of 5x before price erosion)
Premium pricing due to price pressure
Sell through 90% 80% 70% 60% 50% 40%
10% 96.4% 92.9% 89.3% 85.7% 82.1% 78.6%
20% 94.4% 88.9% 83.3% 77.8% 72.2% 66.7%
30% 93.2% 86.4% 79.5% 72.7% 65.9% 59.1%
40% 92.3% 84.6% 76.9% 69.2% 61.5% 53.8%
50% 91.7% 83.3% 75.0% 66.7% 58.3% 50.0%
75% 90.6% 81.3% 71.9% 62.5% 53.1% 43.8%
This would seem to suggest that for a publisher operating at a premium-to-exchange pricing ratio of 5x, at a sell through of 30%, the premium prices would have to erode by 50%, before they are better off withholding inventory from exchanges. But at a 50% sell through, price needs to erode by only 20% before it makes sense to not go the exchange route.
Not as simple a decision as I thought when I first wondered why the publishers are taking this route.
|
A few days back I looked at my behavioral profile identified by Yahoo! I kept meaning to look at my profile on Google, but kept forgetting about it. The last couple of days, I noticed an ad by TheFind about women’s fashion, served by Google, following me everywhere. I am not sure if this is the result of Google’s inference system or just retargeting me based on a clickthrough to TheFind, it reminded me to look up my IBA profile on Google. The image is attached. I do not think I have done anything to deliberately confuse the inference systems, but my profile absolutely amazes me. I would hate to spend an advertiser’s money leveraging this data. However, we in the ad targeting industry will continue to throw the buzzwords and snake oil around to get a slice of the marketing dollars to pay our bills, and attribute any movements in the metrics to the rigorous science behind the process. For the foreseeable future, I think BT will remain a great sales tool but not one that will deliver results. Buyer beware!
|
|
What do you do with idle time? Take a look at the anti-virus scan results!! What an exciting way to kill time! For the first time, I looked at the cookie deletion log on my Trend Micro antivirus s/w on my home computer. I just downloaded the new 2010 version, and everything was set to the default configuration out of the box for a quick scan. Nothing had been tweaked by me. So I assume that it is configured similarly to what the majority of the home installations look like. To my surprise, the quarantine log looks like a Who’s Who of well-known ad networks. The list is below. If I am interpreting this correctly, then that means that every time I run a scan, Trend Micro deletes all the cookies, and any behavioral data these guys had collected about me is lost. That means, for the ad networks touting behavioral targeting, a huge chunk of their cookie list is actually useless. They will see the user again, but they cannot relate the user back to the behavioral data in the absence of personally identifiable information. This obviously impacts ROI calculations as well if a long expiration is used, like 30 days. The ROI item is a different topic. Google offers a plug-in to make the opt-out for IBA persistent despite cookie deletions. That’s a good theoretical approach – but I cannot imagine anybody with any motivation to install the plug-in, and share more data with Google. All they need to do is run a scan with their AV, and for all practical purposes you just opted out again, unless these networks are using IP address to correlate cookies. If Google uses the IP address to map me to the Google DCLK cookie and aggregate my history, that would be worrisome. Something I should try to find out! Just because Trend Micro quarantines these cookies does not mean that all AV scanners are doing the same. Given that they are one of the top 3, I am guessing that they are following established industry practice in this area, but I could be wrong. At Ask, our analytics team said the data seems to indicate very high cookie deletion rates, and we used to speculate why. Well, this list includes Ask.com!! For Ask, it just created a headache for coming up with the right UV, retention & frequency numbers. For the ad networks, this renders the BT story very weak beyond simple search retargeting. Note: After talking to a friend about this today, I realize that the big guys, Google / Yahoo / Microsoft, have alternate means of tracking users even if Trend Micro is deleting their cookies.
|
|
Do the reasons in the article really make sense? I am not sure what the main motivations are behind this deal. I doubt it is to enrich the Place pages. They could easily include Yelp links without having to buy them. Actually I just looked through the Place pages for a couple of local restaurants, and strangely enough – no Yelp links (see screenshot). In the web search world, Google owning content would create a headache in proving that ranking remains neutral. But local is different from web in that there is very little content. So making the best possible use of the Yelp content could be seen as a service improvement rather than unfair play. I can believe that Google wants to use the Yelp real estate to deliver local advertising. But I do not see why Google needs to own the property for that – unless it is for data access reasons. Adsense has done just fine without Google having to own the properties. The one other explanation that I can think of is that Google has some great mobile apps in the pipeline that would significantly benefit from being able ride on Yelp’s content, iPhone presence, and brand. |
