This is the fourth — and arguably most impor­tant — in a series of posts about the mea­sure­ment of online adver­tis­ing, and why I think time-spent-on-site, which I wrote about last time, is far from the panacea many believe it will be.

Now I come to impres­sions, a big ugly ele­phant in the room that’s been lin­ger­ing around for years…

When adver­tis­ers buy ad space, they actu­ally buy “impres­sions” more than any­thing else.  You are likely aware that this is called a CPM model — or cost per thou­sand impres­sions pur­chased. A $20 CPM rate sim­ply means it costs you $20 to buy 1,000 impres­sions on a site.

(Sure, there are other mod­els like CPC — cost per click — which is the stan­dard in paid search.  And there’s CPA — cost per acqui­si­tion — which resurges every now and again as an alter­na­tive but has yet to become main­stream for a vari­ety of issues.)

In any case, if you look at firms that use audi­ence mea­sure­ment ser­vices to deter­mine which sites they want to adver­tise on, it’s almost always based on CPM. This means that while unique vis­i­tors, time-spent-on-site, and page views are all well and good, in and of them­selves, they are sec­ondary met­rics when it comes to the true nuts and bolts of ad buying.

So iron­i­cally, despite all this media fan­fare about time-spent-on-site and the death of the page view, there is still a great divide between what adver­tis­ers and pub­lish­ers use to eval­u­ate adver­tis­ing oppor­tu­ni­ties and what they use to actu­ally exe­cute on those.

Audi­ence mea­sure­ment firms are pub­lish­ing sec­ondary met­rics that may help guide adver­tis­ers to a par­tic­u­lar site, but they are gen­er­ally dis­con­nected from the adver­tis­ing inven­tory itself.

Strangely enough, adver­tis­ers tend to pur­chase inven­tory based on a met­ric — impres­sions — that is rarely made avail­able to any­one out­side the com­pany.

To be fair, because adver­tis­ers often buy only a por­tion of avail­able impres­sions, you could argue that it doesn’t really mat­ter if they know how much total inven­tory a site can offer. For exam­ple, if a site has 10 mil­lion avail­able impres­sions, but you’re only in the mar­ket for 1 mil­lion, what does it mat­ter if they have 1 mil­lion avail­able or 10 million?

I believe it mat­ters for sev­eral rea­sons.  First, sat­u­ra­tion — if they have 1 mil­lion impres­sions and you want 1 mil­lion — that’s 100% of inven­tory. If they have 10 mil­lion, your 1 mil­lion only rep­re­sents 10%. Depending on your ad objec­tives, 100% sat­u­ra­tion or 10% sat­u­ra­tion may be more beneficial.

But with­out the total inven­tory num­ber, you have no idea.

Sec­ond, ad impres­sions can vary greatly by site.  This is because sites offer a vari­ety of ad place­ments for sale — often more than a dozen. From sim­ple home page ban­ners to run-of-site ads to stream­ing media pre-rolls, there is a myr­iad of place­ments adver­tis­ers can offer — and often, more than one adver­tiser shares a par­tic­u­lar place­ment, rotat­ing depend­ing on how many impres­sions have been served and how many have been purchased.

For­tu­nately the IAB has pub­lished — and con­tin­u­ally updates — ad place­ment stan­dards, so there is a gen­eral con­sen­sus as to dif­fer­ent place­ment types.  But there are no real guide­lines on how many or where these place­ments appear. For exam­ple, one site could have a sin­gle ban­ner avail­able on its home­page, while another could offer 10 place­ments, includ­ing one for stream­ing media.

Because unique vis­i­tors, page views, and even time-spent-on-site have a one-to-many rela­tion­ship with ad impres­sions, it’s impos­si­ble to use those base met­rics to even esti­mate impressions.

So take a step back, and a deep breath. If you’re with me so far, you’re prob­a­bly ask­ing why audi­ence mea­sure­ment firms don’t just pub­lish impres­sions in addi­tion to the other metrics.

Good ques­tion. One chal­lenge is that ad impres­sions are dif­fi­cult to mea­sure for audi­ence mea­sure­ment com­pa­nies.  An impres­sion can be gen­er­ated in mul­ti­ple ways; there are few “ad serv­ing” stan­dards.  And pub­lish­ers fre­quently com­bine sev­eral dif­fer­ent tech­nolo­gies to serve ads, often on the same page!

Even if ad impres­sion serv­ing tech­nolo­gies were stan­dard­ized, and the audi­ence mea­sure­ment com­pa­nies could track them, there is another chal­lenge (It’s some­thing I talked about before, in the sec­ond post of this series): Panel-based method­olo­gies are just not as good for gran­u­lar report­ing as are other method­olo­gies like web ana­lyt­ics.  And many ad cam­paigns are just that — gran­u­lar, tar­geted ini­tia­tives that are muted across user pan­els, if not alto­gether absent.

Fur­ther­more, ads often change by day — whereas most audi­ence mea­sure­ment com­pa­nies report weekly or monthly at best.

So don’t hold your breath on this one. I just don’t see it in the cards.

What should adver­tis­ers — and pub­lish­ers — do?

I think there is a sig­nif­i­cant oppor­tu­nity for adver­tis­ers and pub­lish­ers to cre­ate a more effi­cient mar­ket­place.  In turn, these effi­ciency gains should drive a bet­ter cus­tomer expe­ri­ence — as vis­i­tors enjoy a more rel­e­vant online (and offline) rela­tion­ship with vendors.

How?  Take back con­trol and set a new stan­dard that is publisher-centric.  Let me explain.

First, imag­ine a world where pub­lish­ers used web ana­lyt­ics to cap­ture ad impres­sions.  Since these impres­sions are mea­sured at the browser level, and do not even rely on cook­ies, they are arguably the most accu­rate mea­sure you can achieve for true impressions.

Sec­ond, imag­ine a world where pub­lish­ers also cap­tured clicks, and even “down­stream” suc­cess events (con­ver­sions, pur­chases, form com­ple­tions, etc.) that occurred as a result of those clickthroughs.

Finally, imag­ine a world where pub­lish­ers then pro­vided this infor­ma­tion to prospec­tive adver­tis­ers, who in turn could eval­u­ate if and to what extent they wanted to pur­chase that inventory.

A fairy tale, you say?  Nope.

The real­ity is that pub­lish­ers can do this today, and many are already doing some of it using web analytics.

At Omni­ture, we have already helped cus­tomers cap­ture impres­sions, in real-time, for their web­site and report those back to prospec­tive advertisers.

This approach also cap­tures any clicks that result from these impres­sions, and reports them back in real-time.

Of crit­i­cal impor­tance, it is pos­si­ble to seg­ment this impression-tracking with pro­file infor­ma­tion that cus­tomers vol­un­tar­ily and will­ingly pro­vide you, all accom­plished in accor­dance with your pri­vacy policy.

For exam­ple, you could see the geo­graphic dis­tri­b­u­tion of all peo­ple that clicked on one of your home page banners. Alternatively, you could report on adver­tis­ing and con­tent affini­ties that would actu­ally help you sell even more inven­tory on your site.

Tak­ing this con­cept even fur­ther, you could seg­ment ads, by demo­graphic infor­ma­tion that your customer’s have shared with you (and you have once again cov­ered in your pri­vacy policy). This would allow you to share gen­der, age, and income pref­er­ences with prospec­tive adver­tis­ers by each ad run on your site. 

And tak­ing it even fur­ther, you could use third-party data sources like PRIZM to aug­ment the data even more – which I can imag­ine would pro­vide some com­pelling multi-channel syn­er­gies given PRIZM’s preva­lence in the offline world.

From this point, the key is to estab­lish a vehi­cle or mar­ket­place by which you can share this infor­ma­tion with prospec­tive adver­tis­ers.  This could be done as a col­lab­o­ra­tive exchange between numer­ous publishers.

So for­get about tak­ing back con­trol from the audi­ence mea­sure­ment firms– you’re com­pletely chang­ing the game with this kind of data!  Exist­ing approaches can’t even scratch the sur­face when it comes to pro­vid­ing this kind of insight.  The depth, the speed, and the accu­racy are unparalleled.

Per­haps you think this is impos­si­ble; there’s just too much weight around the cur­rent approach.  Try telling that to Google, Yahoo, Microsoft or any other paid search engine.  They run their entire busi­ness on a model that’s not too dis­sim­i­lar from what I’m advocating.

Think about it. Why does this work only for CPC and search key­words?  While there are some nuances to pub­lish­ing that need to be con­sid­ered (as I’ve high­lighted above), it’s not rad­i­cally dif­fer­ent.  And paid search is a $7 bil­lion mar­ket — so the model has legs.

Is this approach with­out its chal­lenges? Of course not.  Stan­dards will have to emerge, dri­ven by gov­er­nance bod­ies like the IAB.  And that will hap­pen over time, just as it has for search.

But it won’t hap­pen at all with­out pub­lish­ers demand­ing a bet­ter mouse trap… a Brave New World, if you will.

So, long live the impres­sion! And if you have ques­tions about mea­sur­ing and opti­miz­ing impres­sions, I encour­age you to reach out to us.  We’d be happy to help.

1 comments
Amit G.
Amit G.

Great point highlighted - impressions, clicks, conversions are better metrics to base decisions rather than visitors, pageviews, time spent based on panel data simply because relevant impressions driving conversions is what matters. Regarding panel data, I would like to emphasize a point that you made a passing reference- panel data measures the quality of the visit (not the visitor). As a marketer, I believe, the self selection in the panel itself introduces a bias where the panel member group could be significantly different than the demographic of visitors. This in itself should force the advertisers to look deeper. One challenge that I see for the proposed approach is sharing of conversion and value of conversion data between advertisers and publishers. The advertiser might be getting much higher value than the ad cost and may not be willing to share this information with the publisher (the reason it works with google is due to the inherent trust and non negotiable pricing – would be interesting to see how it plays out when google starts using that information to ‘make keyword recommendations to consume the budget’). A work around for this could be sharing click data and saying that conversion is advertiser specific and could be tested. Second challenge is we have to appreciate that web marketing is part of a broader CMO strategy of generating a leads pipeline. So, development of impression and clicks benchmarks across publishers would be key when advertisers have to make a high level decision on how much and which inventory to buy in order to meet their overall goals for penetration. As I know there is no such benchmark available. Omniture is in a great position to create such consortium data. If we could convince a couple of big publishers like WSJ, who hold premium inventory and could benefit from this approach, others would follow suit. Targeting with geographic, demographic and PRIZM data for multichannel and niche strategy could spice it up. This would really put power in the hands of publishers to provide relevant ad placement too.