Hacker News Reader

Almost nobody pays for impressions? Do you know how much money is spent on display ads on a CPM basis? Billions.

is true, and not just display ads even a lot of facebook is still CPM/CPC

the point of this article though, is that folks are slowly learning that CPM/CPC is completely worthless. No amount of fraud detection is going to get it under control, and when your typical click frauder can spin up 10,000 micro instances on Amazon and generate a couple of billion false "impressions" you realize that its not cost effective to try to play that game. I agree though that many people have yet to learn this lesson, especially on Facebook but it is happening.

oh, totally agree with that. your point not to mention all of the "ad intelligence" bots or "systems" driving up the prices.

You know a lot about this area. How would one flip advertising around?

Not worthless... just more expensive.

If half of all banner ads are never seen, then that just means the cost is $10 CPM instead of $5 CPM.

Marketers then manage their budgets with this in mind.

Also, traditional display advertising (TV, Print, Radio, etc..) aren't directly measured at all. They have to go through indirect survey channels for any feedback. Yet, advertisers still use them and it has worked fine for a century+.

CPM will be fine forever. Intuitive people don't need direct measurements.

There is a common flaw in this line:

    > If half of all banner ads are never seen, then that
    > just means the cost is $10 CPM instead of $5 CPM.

It assumes that web sites are infinite, they are not. Delivering web pages, and processing clicks, costs CPU resources and bandwidth. A fraudster can put up a bogus web site and throw fraudulent clicks at it and steal fractions of a penny in the millions. And because it is all fake they have very low over head. But an actual web site that is putting up actual content is (hopefully) paying authors, doing research, paying license fees for images, etc.

So click fraud drives the price an advertiser is willing to pay downward since, as you point out they know what they are willing to spend to get their ads shown, and it drives the price below what a legitimate web site needs to continue operating.

Or in Facebook's case the cost of combating fraud overwhelms the price people are willing to pay for a click. Same problem.

Or Google for that matter, the fact that CPC rates have fallen over the last 5 years isn't really something they have been able to fix, and its cutting into both their operating margin and their user experience as they try to make up the revenue in "inventory".

   > CPM will be fine forever. Intuitive people don't 
   > need direct measurements.

Come back and tell me that next year on the 1st of July. (seriously I just added a calendar item to check CPM rates in July 2016). Web advertising is either going to change or die. Since there is a lot of money involved I'm betting on change.

Aren't you conflating two things: display ads cpcs vs adwords cpcs? Despite google telling us that adwords cpcs are falling, we also know total clicks are rising and it is quite possibly a change in mix: more mobile, and more low-cost countries. Anecdotal reports from developed countries claim that high-value adwords are inventory constrained and hence prices are rising -- from kenshoo etc.

It's not at all obvious that advertisers are going to pay for performance, where I assume you mean buy ads on a cpa basis.

There's risk, and someone will bear, and get paid to bear, that risk. If advertisers buy on a cpa basis, they pay more for the privilege.

And frankly (imo) that's all for small to medium advertisers anyway. Bigger advertisers buy 3 month ios and, if you don't perform, don't reup. My experience at a large DSP is this: no-one comes in as a big account; they're all grown. You get $50k for 3 months and told to demonstrate performance. If you do, your account managers call up and demand 3-5x the money. Wash rinse repeat. You do get $3-$5mm/year accounts, but they all started under $100k/3 months. And in any case, all DSPs / ad networks will be measured on a cpa basis. Don't perform, and next quarters ios don't materialize. They're just not sold cpa.

If you aren't aware, btw, there are huge fights about attribution - last view is ridiculous but what most people use, leading to all sorts of gaming behavior particularly in light of cookie churn.

I am aware of the fights, and cookie churn is a symptom of the a challenging and, to my way of thinking, a not very good way of identifying origination.

Lets say I'm content provider A, I get your ad and it has an encrypted URL that lands back on my site. When it lands on my site I can decrypt it and know who sent that click to me and note it. Then, on conversion, pay that person.

The challenge this system has is that it doesn't let you throw your ads into a distribution bucket and wait for them to land on some site that is buying some traffic from some traffic provider. You have to know the sites (and presumably their IP addresses) where it lands. I need a relationship between the site showing the ad, and me. Then I can be sure that anyone who clicks on that ad, and then buys something from me, I can credit the sale to that site.

What is useful (and to some extent different) about this scheme is that you can compute ad spend against revenue generated. There is no 'fraud' because you don't pay unless someone "converts" (the term for making transaction). What it does do however is cut off all the shady ad networks and packagers and skimlinkers and what not because they don't add value for a large brand like Macy's or Target or Amazon.

So a blog like TechCrunch with a relationship with Target can give target an exact number of how much business has arrived at Target because of their ad on Techcrunch stories. And Target can compute exactly how much revenue the clicks that landed there from TechCrunch generated. And from their existing understanding of life time value of a customer who has landed on their site and made a purchase, they can price the value of that advertisement fairly precisely. TechCrunch gets feedback on how much their ads are worth to Target and can choose to make every ad a target ad if they think it will maximize revenue, base on the performance not on the click through rate. Target could lie about conversions, but that would be self defeating in the long term (they want people to land on their site and want people to carry their advertising).

If we learned anything from the meteoric rise of Google as an advertising behemoth it is that people will pay for metrics and demonstrable results. If you can provide the best metrics and the best results, you win.

Your encrypted URL scheme solves nothing as far as I can see. We can already track eg the retargeting provider that brought cookie X in. That's not the problem. The problem is

(1) multiple retargeting providers repeatedly bring user X in

(2) once user X clears his or her cookie and becomes user Y, we no longer have a way to target ads to him or her. If he or she then revisits the site, our analytics become confused and that user is entered anew into a retargeting pool.

(3) user X is exposed to ads from multiple ad partners. Who owns the conversion? Fractional share / time decay / last view through? We can track the partners that showed this user an ad (up to cookie churn), but somehow must apportion the credit.

Users regularly repeatedly visit a site and are exposed to multiple ads before converting -- that's a big part of the debate over who gets credit.

The above problems become more acute as you exhaust retargeting and move to lookalikes or other forms of prospecting.

I happen to think trying to cookie someone is going to ultimately be unsuccessful, and as you point out multiple people feel they have a claim on the conversion. My guess is that will simplfy, kick out everyone in the middle except the content site and the advertiser. I can be totally wrong here, but that is my best guess at what will happen.

ok, but in a cookieless world, how do I identify people to target?

The current flow is this: advertiser pushes 1-party data into an exchange or ad network, either directly or via a dsp or dedicated company ala liveramp or via retargeting partner. This can be as simple as "target this user" or very complex, eg catalog integrations ala tellapart.

With no cookie, how can an exchange or network know that anyone wants to show an ad to this viewer?

The other possibility is a single company sees all / has global ident ala fb and google on mobile.

Also, you keep coming back to multiple parties having a claim on a conversion. That is fundamentally unrelated to cookies. It's this: multiple parties showed an ad to a user. Forget, for now, how that user was identified. If that user then converts, which partners get how much credit?

   > how do I identify people to target?

Publication demographics, backed up by readership surveys. (I know pretty old school but it passes muster even in Germany for privacy protection) Places like TechCrunch survey their readership and identify demographic characteristics that you might want to target. (age, sex, buying habits)

   > With no cookie, how can an exchange or network 
   > know that anyone wants to show an ad to this viewer?

Because they have already made the decision based on the readership, if the publication keeps identity information and demographic information it can share it with the advertisers. Their is no "risk" of showing it because they only pay on conversion, there is only the publication wanting to get better at showing the correct ads.

How that would work in practice, if you have a login with a publication they can share demographics about that login when they ask to get an ad[1], if not they can share general demographics.

   > multiple parties showed an ad to a user. Forget, 
   > for now, how that user was identified. If that 
   > user then converts, which partners get how much 
   > credit?

The partner who got brought the customer gets 100% of the credit, everyone else gets 0. For generic items statistics will have that conversion happening randomly amongst all the possible places it is shown. For non-generic items the partner that got the reader to the point of doing the conversion "wins" that particular round.

As you have surmised it does not support the notion of building an ad network through the use of cookie analytics. However I suspect such tracking/monitoring networks are going to be put out of business by a combination of privacy laws and browser changes. They are already considered "bad" by many consumers and transitively by those consumer's representatives in government.

[1] Caveat the privacy policy of course

I completely disagree about the death of the cookie. First, most users just don't care much. Second, killing 3rd party cookies is a huge gift to the handful of companies with, eg, single sign on so cross site visibility: fb and google. This was the problem with firefox's proposed ban on 3rd party cookies: I can't think of a bigger gift to give to google and fb. Now you may still support that under the theory that a few companies are easier to regulate than many, but I'm not sure that's a compelling case to regulators. I think the most likely outcome is we get european style cookie regulation, and the majority of sites put a click-dismissed banner on the top saying if you keep reading we get to cookie you. Oh, and if you do just use technical measures to kill 3rd party cookies, google et al can evade that by having pubs cname them into the first party then doing standard cookie stitching.

Disorganized thoughts:

Your ad solution sounds to me like a trip back to advertising 10 years ago -- lots of brand ads / dems targeting. Surveys are pretty crappy; high income dems tend not to respond; people lie; most publishers don't collect regdata; etc. Obviously a handful of pubs -- those used as proxies for rich demographics -- would love this (nyt, ios review sites, bmw/mercedes fan clubs) but for most pubs it would be a loss.

Even with lots of direct response ads there is too much inventory continually driving ad rates down; I don't see how removing a bunch of demand would improve anything from most pubs' perspectives.

Publishers uploading regdata also brings its own privacy problems.

ps -- replying to self, but hopefully you'll see it -- if you decide to do anything in this space, I'd be interested in talking to you...

I heard of (large) brand advertisers where marketing managers bought fraudulent clicks on their own campaigns to "guarantee" success of these campaigns, for internal politics/career reasons. These advertisers probably like it the way it is, and keep their offline marketing colleagues in the dark.

I founded a startup creating an innovative media for mobile that, while probably not "immune" to fraud, had very little space for, and provided accurate and real measurements on engagement and content exposure.

While trying to sell this, I heard multiple advertisers say that they only buy TV ads because it would look "weird" with their hierarchy if the company is not on prime time TV, despite knowing that TV ads are hardly profitable and nearly impossible to measure. The same thing goes for billboards on roads with heavy traffic in the capital.

> Eventually the advertisers are going to pay for performance, and to do that they are going to have to fix the way they do advertising so that their ad, leads to their landing page which identifies who delivered the ad.

I've setup a system that tracked that.

The result was a 95% drop in payments for referrals.

> Eventually the advertisers are going to pay for performance, and to do that they are going to have to fix the way they do advertising so that their ad, leads to their landing page which identifies who delivered the ad. Then they are going to pay based on whether or not the person who landed on their page actually buys something (service, good, or subscription). No money, no ad payment.

This already happens. It's referred to as an "acquisition" and "cost-per-acquisition" (CPA) is a fairly standard term in the advertising industry these days.

However, it doesn't necessarily involve money (e.g. a car company might define an acquisition as someone signing up for a test drive).

Spent some time a few years back working in web analytics and my experience was similar to what ChuckMcM describes above - advertisers were relatively sophisticated in general and wanted to pay for performance rather than impression. There are certainly agencies/networks that will waste you're money in advertising if you're not paying attention in digital or traditional media but controlling fraud wasn't terribly difficult. Lots of advertisers would run their own beacons to check and make sure they were getting the audience they were sold. The thing that made them the most nervous was when they couldn't audit their campaigns and the person measuring the audience was the same one selling it.

> Its a shame that "affiliate" ads like the ones they let bloggers use have 100X their ROI...

Sounds like an incredible market opportunity.

What are the companies that are building ad marketplaces targeting high performing affiliate bloggers which demonstrate their value better to advertisers, and share a higher proportion of profit with the bloggers?

Are there fundamental issues trying to scale up a business like this? Not everyone can spin their own custom 'sponsership' packages like DaringFireball, but there is a lot of good blog content out there.

I remember seeing some bespoke ad networks in the designer space, can't think of their name off-hand. Had the feel of a high-end web ring, if that's not an oxymoron.

http://decknetwork.net/

http://carbonads.net/

> Eventually the advertisers are going to pay for performance,

That assumes that you can actually measure performance or attribution. This is far from a solved problem. As such many publishers are reluctant to sell on this basis.

Trivial rebuttal:

User clicks on ad, buys product. Who's to say that ad caused the sale? Maybe they've seen that particular ad or a variant on dozens of different publisher sites. Or even better, maybe the true product awareness was driven by an offline interaction (billboard or word of mouth).

In the general case, you simply cannot measure direct attribution or performance with any real sense of confidence.

The majority of mobile is still CPM. I dare say the vast majority. Source: I ran a mobile DSP until recently.