The End to End Principle in Ad Exchange Design (The Thin Exchange, 2)

Where should decisions about which ad runs where be made? Ajay Sravanapudi, in his recent article on AdExchanger says

a DSP is really just a feature on an exchange... A DSP simply uses [RTB API] of an exchange to buy media and run campaigns more effectively. The exchange has an ad server that can deliver campaign pacing, frequency capping, targeting, etc. All that is missing is some intelligence to “auto-magically” buy media on behalf of the campaign... Dozens of ad networks have done this for years on things like YieldManager on the RightMedia Exchange (RMX). If we can simply layer this “auto-magical” intelligence on the exchange then there is no need to pay for a DSP.

I disagree. There's a sort of businessman's view here, where an intuition about power leads to an answer at odds with systemic efficiency.

Exchange 1.0 did not sell real-time. So the exchanges had to have rules in-system. Like the NASDAQ (and other limit order book markets), the exchange hosted the rules about who wanted to buy and who wanted to sell at any given price on any of the thousands of things traded there.

But RTB ad exchanges don't have thousands of things being traded, they have an almost unlimited number of things. Each ad impression--the placement, the context, the viewer--is different than every other. There is no commodity, so there can be no order book.

With an order book, the commodities had to be limited: i.e. "if the ad presented is on the front page of CNN.com, and the person viewing the ad is a male 18-34 years old, then bid $5.00 per thousand." That doesn't work when the ad presented is a photo of a red flower on Photobucket presented to a non-logged in 33 year old male in Northern New Jersey at 12:13am on a Sunday and who searched for gardening tools at an online retailer yesterday but didn't buy anything and whose circle of acquaintances includes several people who bought sunglasses this week." What ad you put in front of this person and at what price is a difficult problem, not one that can be reduced to simple rules.

The only way to move away from rule-driven trading is to use algorithms. On the buy side, developing the right algorithm requires a ton of experience, a ton of data from live campaigns (and the insight into how well each individual impression worked), a huge amount of experimentation, mathematical savvy, and a dose of genius. On the sell-side the algorithms are even more complicated to develop. The algorithms are where the intelligence is.

Where should these algorithms be run? Here's an analogy. Let's say you wanted your computer to run the algorithms. Where do you think they should be coded? In the operating system? Of course not. In the application layer? Almost certainly not. Obviously, you'd code them as routines to be run by a more general application, like Excel. The lower down the stack, the more generic the functionality should be. This is called the End to End Principle:

Using performance to justify placing functions in a low-level subsystem must be done carefully. Sometimes, by examining the problem thoroughly, the same or better performance enhancement can be achieved at the high level. Performing a function at a low level may be more efficient, if the function can be performed with a minimum perturbation of the machinery already included in the low-level subsystem, but just the opposite situation can occur – that is, performing the function at the lower level may cost more – for two reasons. First, since the lower level subsystem is common to many applications, those applications that do not need the function will pay for it anyway. Second, the low-level subsystem may not have as much information as the higher levels, so it cannot do the job as efficiently.

Saltzer, Reed and Clark in "End to End Arguments in System Design". This paper described an idea that has been central to internet architecture since early days: don't put in the center what can be done at the edges. Similarly, David Isenberg's "Rise of the Stupid Networks" (predicting that the internet would beat out the "smart" telecom nets.)

Okay, I can hear all you adtech gearheads: is the exchange really "low" level? It's probably the most complicated piece of software in the whole ecosystem.

But low-level in this argument really means that the functionality is used by the most end-user applications. It has nothing to do with how close to the hardware the function is (the two are correlated, but that's outside my scope.)

Clearly the exchange has the functionality that is shared by the most end-users. Each agent is different: different approaches, different algorithms, different in ways none of us has yet imagined. Why should we attempt to encode this as-yet-undetermined difference into the exchange? Putting DSP functionality into the exchange simply means that everyone has to pay for it, even if they don't use it. This means that other, better ways to be a DSP do not get developed, because then the customer has to pay twice: once for the DSP's DSP and once for the exchange's DSP.

And this brings me to my real beef with the idea that DSP functionality should be built into the exchange, or that any functionality outside of what is absolutely necessary should be built into the exchange. The internet has been so phenomenally successful because the low levels are bare boned and flexible. HTTP, FTP, POP, SMTP, DNS, IMAP, etc. have all been built on top of TCP because TCP does nothing but transfer data from one place to another. It doesn't have much expectation about what that data is or what it should do. If the internet's designers had made TCP more "intelligent", we probably would have never had the Web or Skype. Building low level functionality that is simple and allows layers to be built on top enables innovation. And heaven knows what we need right now in interactive advertising is some innovation.

I don't think the ad exchanges should layer in "auto-magical intelligence." I don't think they should layer in anything. I think they should start dumping functionality like Carl Fredricksen tossing furniture out of his house. The ad exchange should do three things. It should do them fast, it should do them cheaply and it should do them six-sigma. What the ad exchange should do, and all it should do, is cookie-match, cross and clear. The ad exchange should be thin, and as dumb as possible.

Everybody's an ad exchange (The Thin Exchange, 1)

Everybody's a DSP? Everybody's a marketplace.

There's this confusing moving about in the marketplace. AdECN was a pure exchange, and is now part of a publisher. Right Media, same thing. OpenX was a publisher tool, and is now running an exchange. AdMeld similarly. AppNexus was an exchange and is now a DSP (I think.) Same with Turn (who was first an ad net before raising money to become an exchange.) Glam and FIM are publishers and now have some features of an exchange. Several of the ad agency holding companies are making unlikely noises about building their own tech. And AdEx, well... they're doing pretty much everything.

It was so confusing I made a picture of companies moving about.

A good entrepreneur will change strategies as they learn the lay of the land. But some companies who weren't exchanges are becoming exchanges and some companies who were exchanges are becoming something else so, um, how does the land lay?

Here's what I think:

1. The ad exchanges know that running a marketplace should not command what they are charging, and lie awake at night fearing that their customers might someday come to the same awful conclusion.
   
2. The customers already have.
   

Why did the exchanges add other functionality? Because they know that they won't make much money as an exchange. I don't have much to add to my analysis in the linked post, other than to say that it seems all the exchanges agree with me, if you look at what they do rather than what they say.

Why is everyone becoming an exchange? Because it's just not that hard to add exchange-like functionality and escape the 20% transaction fees being levied by Google et al. An exchange is a low marginal cost, high fixed cost system. Once you've built the system, you just need to amortize the cost over a sufficient volume. That means that anyone with good volume is better off building their own than paying someone else.

If you extend this economic logic into the future, you arrive at an inevitable conclusion. Someday, someone will garner a huge amount of volume by offering exchange services at the lowest possible price, somewhere just north of marginal cost, probably in the 1% to 5% range of transaction fees. Everyone else will find that it is cheaper to use this single exchange than it is to run one themselves. Non-exchanges will stop reinventing the RTB wheel. And exchanges will be glad they moved into other lines of business.

I know it's a good deal for you.

In the past few months I did not invest in a couple of companies whose businesses I really liked because they were raising seed financing through convertible debt.

Convertible debt is essentially a loan with an option to invest in the next equity round at a discount. So, for instance, a typical seed stage convertible note might give the lender the right to roll the principal and accrued interest into the Series A at a 20%-25% discount.

Chris Dixon posted about the reasons why convertible seed rounds are a bad idea. I agree with his reasons. But I have another, more investor-centric, one: seed stage convertible debt is usually a bad deal for the investor.

There's an old saying in the venture industry: lemons ripen before pearls are cultured. Startups that are going to fail usually fail quickly. Look at Scott Shane's numbers on startup failure rates, in the graph below. While these numbers are for all small businesses (not just venture-backed ones) and for companies started in 1992, they seem to jibe with what I have observed over my thirteen years of venture investing.

If you invest in companies at founding, 40% will survive to year six. If, on the other hand, you invest in the companies that survived their first year, 53% will survive to year six.

Let's say that, like Fred Wilson, 33% of your early stage investments fail, 33% go sideways (1.5x return) and 33% return 5x-10x*. (I assume that "early stage" means investing one year into the company's life.) Then, if 25% of businesses fail between year zero and year one, seed investors should expect 50% failure, 25% sideways and 25% success.

For a Series A investor (assuming the average successful exit is 7.5x and exits are five years after investment) the IRR of the fund under these assumptions before management fees and carry is 25%. Until the last decade, this was close to the average return for VC as a whole.

As a seed investor, I should expect higher returns (I am taking more risk): I want a 30% return, rather than 25%. I should also expect it takes six years to exit the investment rather than five. For sideways and successful investments, the return is the same dollar amount as the Series A investor (so, if I invest at a 50% discount to the Series A, I get 3x for sideways and 15x for successful exits.)

Running these numbers, I need to invest at a 50%-60% discount to the Series A to get my target return.

Even if my simplifying assumptions are too simplifying, the discount has to be at least 40% for a convertible note to be economically rational**. I doubt whether any Series A investor would let that stand because, in hindsight, the seed investors took very little risk***.

I've had a couple of entrepreneurs tell me that they wanted the round to be a convertible note because they wanted to limit dilution. What they are really saying is that they want me to invest at a higher price than I would if we actually agreed on a price. I understand that, and I sympathize, and I even recognize that many angels turn a blind eye to this because they want to pretend they are not getting a bad price. But let's be honest, they are getting a bad price.

As a matter of discipline, I have a handful of rules I won't break, no matter how much I like your company. One of them is that I won't invest in seed-stage convertible debt.

------------
* There is some evidence that venture backed firms are less likely to fail than startups as a whole, and there's a lot of evidence that Fred Wilson is a better than average investor, so his failure rates may not match Shane's. There is also some evidence that VCs keep companies afloat until about five years after their investment, then give up, so the failure rate curve for VC-funded firms may be distorted.

** Disregarding that some angel investors will take a worse return in the hopes that the Series A will be lead by a great VC, allowing them to take a carry-free ride on that investor's coattails.

*** To be fair, I think the VCs really just take the entirely sensible viewpoint of caring more about the ongoing management of the company than the people who put money in a year ago and haven't been involved since.

The long siege of Corbenic

Darren Herman's article on AdExchanger this week is more important than it might seem at first read: it's the first time a major figure in our little adtech industry has pointed out that what we've built in terms of exchanges and platforms and data is just a small first step to what we will need to build over the next decade. Bear with me.

*****

1. The state of the state

Those of you who have been reading this blog for the past year know I've been thinking about several online display advertising paradoxes:

• "There is an oversupply of inventory" vs. time spent with media (i.e. supply of eyeball time) is decreasing (also here)
  
• Dozens of targeting startups pitching radical lift vs. being told by practically everyone in the industry that targeting does not appreciably increase lift
• The common industry lament that CPMs are falling rapidly vs. the data that seems to contradict that

One fact about what our industry is doing resolves all of these.

*****

2. What we are trying to do for marketers

The first internet marketing rant I heard was "the internet will create true one-to-one, accountable, ROI driven marketing!" That was in 1995.

Still waiting.

But it's true, or will be some day. The ability to dynamically maximize the ROI of your campaigns across all media in real-time is what our industry is building. To do this we need to be able to formulate hypotheses, test them, figure out the ROI, and reallocate into the highest ROI buys. And then, because humans are changeable beasts, we need to do it all over again, forever.

*****

3. What is Quality?

ROI is the value of the customer divided by the cost of acquiring them.

ROI = LTV / CPA

Computing this isn't as simple as it sounds.

Back in the day, at Prodigy, we kept track of each customer by when they started and by what marketing channel they arrived. For each marketing channel we would track the average lifecycle. That is, for a customer arriving in month x, 80% would stay one month (first month was free!), 40% would stay two months, 30% three months, 25% four months, etc.* The churn was different by marketing campaign because some campaigns got people for whom the product was wrong to sign up: they did not stay long.

We then multiplied the probability that any given person would remain in a given month by their contribution margin for that month (say $10, except in the free trial month, when it was -$10) and discounted all that back to month 0 using the company's cost of capital**. This was the customer lifetime value (LTV).

This is a complicated example. If you are promoting an offer where you get paid $2 every time you bring someone a customer, your LTV for that customer is, essentially, $2. But in most cases, the right customers are more likely to remain customers (unless you subsequently scare them away), and the wrong customers are less likely to remain customers, so the LTV is usually dependent on the marketing campaign.

Now, to the cost of acquiring a customer using display advertising. This part most of you know. I'm going to do it the simplest way, though.

CPA = Cost of media x Conversion = CPM/1000 x Conversion

Conversion here is number of people who see the ad divided by the number of people who become customers of the advertiser. That is, they convert from ad viewer to customer. Conversion is key, but it is an incredibly complicated thing to predict because it depends on: who the viewer is, what they are thinking at that moment, what they are looking to buy, where and when they are seeing the ad, what surrounds the ad, how many times they have seen the ad before, what the ad itself is, whether they are interested in what is being advertised, etc.

ROI = LTV /(CPM/1000 x Conversion)

None of LTV, CPM and Conversion are independent of the others, each depends on the particulars of the campaign (i.e. targeting high-net worth individuals changes CPMs , changes conversion, and changes LTV, the latter two in ways depending on the product.)

*****

4. What are we doing now?

What we measure now in the adtech industry is CPM. Some DSPs, with some clients, will claim that they are doing more. But if there is true integration of conversion into their buys--or any integrations that go much beyond "the CMO likes the results"--then it is extremely rare.

Because the adtech people can't measure conversion*** or LTV, they optimize CPM while trying to hold LTV and conversion constant. Using the same creative and target viewer, they find lower CPMs. This explains the puzzles I mention above.

• There isn't more eyeball time (i.e. supply), but we are putting ads in places that previously had no ads. On social networking sites, on UGC sites, on small sites, etc. We do this by using targeting data to find the few "right" viewers in the sea of "wrong" viewers. There isn't more attention to sell, we are just selling more ads per hour of attention. This is the increase in supply. (I mentioned this previously.)
• People do not see "lift" (i.e. increase in the conversion rate) by using targeting data because that's not what we are using it for. We are using targeting data to increase reach and find the same conversion rates at a lower CPM. Targeting data increases ROI by lowering CPMs, not increasing conversion.
• If CPMs are falling at the major media purveyors (as reported) then that is being made up for by increasing CPMs on social networks, etc. On average CPMs are not falling, but there is an equalization of CPMs happening across internet media. We hear the lament of the New York Times' of the world, but not the jubilation of everyone else. Facebook's ad revenue came not entirely by taking time spent with the NYTimes away, it also came by lowering their CPMs. You can find the same person you were targeting at the NYTimes on Facebook, for lower cost.
  

*****

5. What's next?

This is a great service to marketers. Adtech increases ROI. But it's transitory. Once enough ad dollars use a specific way of finding the right viewer on overlooked sites, the price of the inventory on those sites rises. We are right-pricing all the inventory on the internet after years of it being overpriced on the high-profile sites and underpriced everywhere else.

But there are declining returns to this strategy, as the inventory becomes more rationally priced. I've heard some rumblings to this effect already, although I think we have another couple of years before it becomes widespread.

Step two is to focus on conversion. This is what Darren was implying in his article. But to know conversion, there has to be integration back into the marketer: did this particular instance of an ad result in a product being sold? When this feedback loop is in place****, marketer's agents can start to use the algorithms they used to find low CPMs to find higher conversion rates. This will be a sea-change. But implementing it will be difficult because it requires substantial tech investment at the marketer, as well as procedural and probably cultural change. It's also one of the primary reasons the DSPs will have to choose between being a marketer's agent--working directly with marketers--or a technology provider to marketer's agents.

Step two is now. If you are starting a company to do this, contact me.

*****

6. Step 3

Step 3 is the end of advertising. It is using marketing communication to increase LTV.

Kotler spends less than 5% of his seminal textbook on advertising, the rest is about customer value, loyalty, product strategy, pricing, etc. But in the adtech world, it's all about advertising/direct marketing/promotion. When we can measure the effect of marketing communications on LTV, that will change.

Increasing LTV means keeping customers, not getting them. It means figuring out what customers want and building or changing your product to keep them. It means finding the right pricing, providing the right customer service and out-innovating your competitors.

And that is the holy grail of marketing.

-------
* These numbers are illustrative, but directionally correct. I don't remember the real numbers. It's been a long time.
** I don't really want to explain contribution margin, discounting and cost of capital. This post is going to be long enough as it is. Ask Fred Wilson on Monday.
*** Click-through is not conversion. If you've ever been in lead-gen you know you can easily and cheaply generate poor quality click-through. Free i-pod, anyone?
**** Outside of the arbitrage world, where it is being done now. The arbitragers are and always have been the smartest people in marketing.

Tweet Congress: we want jobs and innovation

If a great company shows up on my doorstep, I try to invest in it. If two great companies show up on my doorstep, I try to invest in both of them.

I think it's pretty much agreed that at this point, the bottleneck in startup formation is not money, it's great entrepreneurs and great teams. We've also pretty much settled that startups are the engine of job growth and innovation. If we had a way to increase the number of startups in this country, the entire country would benefit.

The Startup Visa does this and it just took a big step forward, with legislation introduced in the Senate. That's great, but what it really means is that while the hard work may be over, the bruising work is yet to come. Now that this bill has a chance to become law, opponents will emerge to try and kill it. Nervous congresspeople will try to avoid taking a stand on something as controversial as "immigration" (no matter the uncontroversial merits of this particular bill.) We need to let them know we want this to pass.

Brad Feld and Paul Graham did the hard work, now it's our turn. Luckily, our work isn't so hard: go to the Startup Visa site, it has a tool to tweet your congresspeople about the bill.

I can't think of a single rational reason this bill is not a win for everyone. But unless we proactively push our elected representatives to pass it, chances are it will get lost in the twisty corridors of the Capitol, as so many good ideas do. So, go, now, and support it!

Kicked to the curb

I'm a believer in publisher's agents (also known as publisher brokers or--the dreaded TLA*--SSPs.) I think it's pretty obvious that they are a necessary precursor to a reasonably efficient marketplace**.

But a lot of people don't agree with me. Important people. Like Google, and Microsoft.

I've written before that I think publishers showing up in a data-driven marketplace without an agent risk getting ripped off. The value to publishers is pretty clear: if you don't have a good idea what each piece of inventory is worth to buyers, you risk mispricing it. Having an agent with the technology, data and information flow to more accurately value your inventory is critical. Publishers I know who implement the simplest of information/data-driven pricing strategies see an immediate uptick in CPMs. Right now it's easy because so few are doing it; it will get harder. As it gets harder, publishers will need better and better agents.

But the publisher agents are also critical to the marketplaces. The pub agents provide a crucial service to the exchanges: they vet publisher quality (a huge problem at the smaller exchanges), facilitate the back-office functions and will, eventually, act as credit assurance. Right now the big exchanges are reluctant to work with smaller publishers: Google, for instance, pushes them into AdSense--this looks to me like an internal political problem, in part. Microsoft, reportedly, views the pub brokers as skimming margin that is rightfully theirs. That is, they view others making money on services MSFT doesn't offer to customers they don't want as illegitimate, even though MSFT would make more money thereby.

In my opinion, Google and Microsoft are impeding progress in display. Google has a franchise to protect and Microsoft is just plain ambivalent about success. Impeding progress is a good way to get yourself disrupted.

Publishers are realizing the value of publisher agents. At some point, a critical volume of inventory will flow through pub agents and then an interesting thing will happen: the pub agents and marketer agents will all decide, en masse, to join or form an exchange--one that excludes retail buyers and sellers--and the other marketplaces will become illiquid backwaters.
------------
* Three-letter acronym.
** Full disclosure, I'm an advisor to PubGears.

Duck, duck, goose on the demand side

Mike Walrath in an excellent article over at AdExchanger said "What we're really talking about here is the race to build the next-generation digital marketing services company." I would put it more strongly: the DSPs, the ad agencies and the ad networks are on a collision course*.

Once upon a time it was pretty easy to tell who was doing what. Ad agencies worked for marketers and rep firms worked for publishers too small to get the attention of the agencies. But the incredibly quick growth of digital advertising made a hash of all that. Ad networks once were digital rep firms, now they are impossible to categorize, except ad hoc. Digital media buyers once were agencies, now they are impossible to categorize, even ad hoc. There are hundreds of companies that have entered the digital display category in the last five years and almost all of them are difficult to categorize**.

GCA/Savvian put together a Display Advertising Technology Landscape map, showing some 140 companies in 21 categories. Almost every company on that map should be in multiple categories or will be in multiple categories in the next 12 months. Ad networks are becoming DSPs or exchanges, ad exchanges are becoming creative optimizers or DSPs, ad servers and optimizers are becoming exchanges, arbitragers are becoming DSPs or exchanges, rich media companies are becoming optimizers... you name it. It's chaos. Albeit the normal chaos of a promising market.

In broad strokes, I think the future of all this is inevitable. In a few years there will be three layers between the marketer and the publisher: the publisher's agent, the marketplace, and the marketer's agent. The agents will use technology developed by dedicated technology firms to some extent, and they will have proprietary technology to some extent. They will buy data from dedicated data firms to some extent and they will have proprietary data to some extent. The marketplaces will have their own technologies and will come in two flavors: 'exchanges' for fine-tuned purchases and OTC-type enablers for block purchases.

There will be various types of technology firms--from trafficking to API access to optimization--and there will be many types of data firms--from analytics to targeting to ROI estimation. But the number of players between marketer and publisher will shrink from six-ish to three***.

Here's my rough take, in qualitative visual form.

What happens to the agencies, the DSPs and the ad networks in this scenario?

Ad networks will need to choose between becoming a publisher's agent, becoming a marketer's agent, becoming a marketplace, becoming a technology provider or becoming an arbitrageur. They can't be all of these things--or even most of them--anymore, at least not on a large scale (many of them have already chosen a path.) Publishers and marketers aren't thrilled with the inherent conflicts of interest and built-in obfuscation of many of the larger ad nets. As alternatives become more available, they won't put up with them.

DSPs will need to choose between being a technology provider to marketer's agents or being a marketer's agent. Most of the DSPs I know have always claimed to want to be technology providers. They have been pushed into providing agency services because the agency media buyers have not had the requisite competencies to use the DSP tools. The DSPs responded by providing these skills for hire: some of them have become, by now, de facto media buying agencies. Some of them have begun buying media on behalf of marketers, rather than agencies. DSPs are now confronting the classic tradeoff between easy revenue growth as a professional services company and valuable revenue growth as a technology company****.

The agencies see the danger of being displaced as the primary owner of their customer, the marketer. There are two responses: buy or build. The ad agency holding companies will inevitably buy some of the DSPs that choose to become marketer's agents (as they bought the interactive agencies and the SEMs.) The ad agencies will also begin to build in-house expertise in using the DSPs. This will be painful for them because they will be competing to hire people who understand numbers, and these people are far more expensive than their traditional hires. If an agency wants to hire someone with the analytical skills that would enable them to work at an investment bank, then they will need to pay them like an investment bank. How this sits with the purchasing departments of the major marketers--who sometimes seem more concerned with cost than efficiency--remains to be seen.

There won't be one winner, but three years from now there will be fewer than ten companies dominating the marketer's agent piece of the world. The foundations that lead to this dominance will be laid this year.

------------------
* I talk about this incessantly in person but have not written about it because I have every type of conflict of interest here: financial, professional, personal, sentimental, intellectual, moral, etc. Proceed with caution.

** I have a notebook where I have 500+ companies listed under 25+ different categories, all digital display ad tech (no traditional digital agencies or ad networks, that would probably triple the count.) I didn't put this together on purpose, it was a result of talking to entrepreneurs about their potential competitors, so the count is probably low by quite a bit. I tried making a 'map', but found that it was physically impossible in two dimensions. GCA/Savvian did it by simplifying several aspects of their presentation and limiting it to the Web and internet video.

*** In many--most, if you count by dollars rather than impressions--cases, the actual number of players is currently one: the agency. The agency places the buy directly with the publisher. This also will change as publishers realize that if they don't have an agent representing them they are not maximizing revenue. It's interesting that almost no large marketers do their own media buying, but almost all large publishers do their own media selling. That may have made sense when there was one newspaper in each town and three TV stations, but not any more.

**** I have heard from one holding company exec, re one of the DSPs: "we would buy them if they valued themselves as an agency, but they keep insisting they're a technology company." Agencies sell for 6-10 times forward earnings. Tech companies sell for 40 times hope. If you somehow think the former is better, your VC wants a word with you.