How to rake it in by screwing your customers

A long, long time ago I was at Prodigy when we decided to change from hourly pricing to a flat rate of $19.95 per month. Flat rate pricing was clearly preferred by our customers, and our competitors who were offering it were taking them away from us.

Problem was, when people are accessing the internet by dialing into 28.8k modems, more hours online meant more peak demand meant more modems needed, meant more expense.

As we evaluated the price change, I noticed that some of our formerly best customers would now become our worst customers. For instance, there was a bunch of die-hard group text-based RPG fans who spent 100+ hours online per month*. Paying by the hour they were great customers, but with a flat rate they were our worst. We had to ditch the RPG.

In fact, we had to ditch any content that people spent a lot of time with. It turned out that the people who liked the service the most, who spent the most time on it, were our worst customers. Our best customers were the people who never logged on but never got around to turning off the monthly bill.

I lost interest in this business model and moved on; businesses that do better as their customers do worse should not survive.

I was thinking about this today because of my friend Josh's excellent blog entry on the Google Mortgage thing and on retail banking in general.

Unlike most people's mental model of retail banking operations, banks do not make most of their money on the difference between the rates at which they lend versus the rate they offer for savings. American banks, quite distinctly from banks elsewhere in the world, make the bulk of their money from fees and charges. Invisible and often unavoidable consequences of little clauses in contracts that no one ever reads.

Banks' best customers are the ones who are getting screwed by the banks. Banks' worst customers are the ones who are probably pretty happy with their bank. This perverse incentive shouldn't persist, but it has. What's it going to take to change it?

-------
* This was a lot back then.

More on Ad Dollars per Hour

My friend Rob Leathern, over at CPM Advisors* was blogging about the same thing I was blogging about in Details, Details (unbeknownst to me, since he keeps changing where he's blogging from.)

We used the same approach, but seem to have come up with different conclusions. I don't have time today to figure out why, I'll do it tomorrow. In the meantime, read these:

Online Time Not Worth What It Should Be

More Television vs. Online “revenue per user hour”


-----
* Blogger disclosure: I'm an investor in CPM Advisors.

The 300%-500% Lift Meme

I talk to a lot of people trying to make internet advertising more effective, most of them startups or people socializing a new idea before starting a company. Over the past two years, I think about two-thirds of them have told me that their new technology/process is going to provide a "300%-500% lift."

I even heard a story of a VC, after being pitched on a more reasonable lift, say "your approach is interesting, but we need to see you deliver a 300%-500% lift to be competitive in the market." ( I looked at this VC's website and found no ad targeting companies in his portfolio.)

Sometimes upon hearing this, I drift into a daydream about combining behavioral targeting, social targeting, retargeting, creative optimization, rich media, distribution optimization, contextual targeting and offer optimization technologies into one super-arbitrage strategy. The resulting 328,050% - 19,531,250% lift would allow me to buy $0.50 CPMs and pretty much overnight control the US economy*.

Assuming, of course, that these lifts are real, are the average lifts, are replicable at scale, are actually the result of the data/process/technology itself and not some artifact of attention (i.e. the Hawthorne Effect), and are orthogonal (which I'd expect if they are real and not artifacts, maybe not to the extent of the last paragraph, but between, at least: targeting, distribution, creative/offer/media.)

Not sure where this meme originated, but it very clearly says one thing: noone knows nothing. How well any of this technology works is an open question.

--------
* I suppose, in a way, this is kind of what Google did, so I'm not saying it's not possible, just that I doubt we can all sit down at our laptops and replicate it.

Those projects that come to nothing

After I pulled data from umpteen sources and crunched it for Details, Details, I looked at the resulting graph and saw that I had found... nothing. If I was a journalist, I'm sure my hard-bitten, cigar-chomping editor would have chewed me out, "There's no story here, Neumann!"

But that's the way it is. I was hoping for some trend or abrupt change that would tell me something I didn't know, some result I could attribute to a cause.

Let's look again at the nothing, the ad dollars per hour of online use. I'm going to correct for changes in overall spending per hour of media use, because the period in question was a bit up and down for the ad world. Below is a graph of Online ad dollars per hour as a percent of total ad dollars per hour, from 2001 to 2007.

Data sources the same as for Details, Details.

Why is this not much? Because it is a graph of efficacy. This is what we online marketing people do. The online media folk and the online app folk and even the Verizon DSL technicians get people online and get people to spend more time there. This graph factors that out.

Macroeconomic conditions determine how much is spent on advertising. This graph factors that out.

All this graph shows is how much an advertiser is willing to pay to get their message to a consumer over the course of an hour online as a percent of what they are willing to pay over all media. If it is 50%, then that probably means the advertiser feels they are about 50% as likely to sell something in an hour as they would be otherwise. The ads are half as effective.

So, an increase of 5% over four years is not much of anything. Nothing, really.

What happened in this period? Here's a few things that apparently changed nothing:

Starting in 2000, Google revolutionized the purchase of intent.

Starting in 2002, Tacoda and the scores of companies following its lead, revolutionized what advertisers know about the people seeing their ads, allowing precision targeting.

Starting in 2005 (or thereabouts), Right Media and its followers revolutionized the purchasing of ads across the huge internet media landscape.

What has all the work done over the past ten years to build the infrastructure for a true one-to-one ad marketplace gotten us? Are advertisers, with all the data and mathematics and optimizing they have available, really getting no more for their money than they were ten years ago? I find it hard to believe, personally. What am I missing? Or, what are we all missing?

No starting gun

Chris Dixon's blog post, The Ideal Startup Career Path is spot on: if you want to start a company, go work for a startup. I'd add: find a company started by someone who comes out of the entrepreneurial community and has a really big idea, join as early in the company's lifecycle as possible, and make it clear through your actions that you will wear any hat needing a head on any given day.

At my last startup, we actively looked for and hired entrepreneurial people and several of them went on to found their own companies. The ones I've given support to include Greg Yardley with Pinch Media, Viva Chu with Handipoints, and Rob Leathern with CPM Advisors. There are others, and then others who will or are starting something. For a company that only had some 30-odd employees, that's a pretty good conversion rate.

I'm tempted to say that we made it look easy, but I'm pretty sure what went through their heads was "if these idiots can do it, then I certainly can." This realization, in its many forms, is certainly the spur to more startup formation than any other factor ever will be. I meet so many would-be entrepreneurs with good ideas who just can't figure out how to start. Who worry about knowing what to do next. Who are waiting for some intangible starting gun to go off, some sign from heaven. There is no starting gun, there will be no sign from heaven. You just need to know that it isn't really that hard.

Watching someone else manage the building of a company provides the crucial lesson: there's no special secret magic. Go work for a startup, and you'll see.

Details, details

"An important lesson... is that details may matter."
- Kyle Bagwell, The Economic Analysis of Advertising

Supply in the ad market market is impressions created, people seeing an ad. Everything else can be viewed in terms of this. Demand is impressions bought.

Supply = Impressions created = Hours viewing media x Impressions per hour
Demand = Impressions bought = Ad spend x 1000 / CPM

Supply = demand, so:

CPM = Ad Spend / Thousands of hours x Impressions per hour

Here's a graph of total US ad spend divided by total US hours spent with media, both online and offline.

Average CPMs would be this amount divided by average ad impressions per hour.

1) Online spend per hour is much less than offline spend per hour. This is partly a result of people moving online faster than ad spend, as John K. made me aware last week. But also note that offline ad spend per hour was rising until 2007, which means that--absent offline efficacy having improved over the last ten years (ha!)--marketers are getting worse ROI offline than they had been. In fact, the projections say that when the ad market recovers, this trend will continue. Perhaps internet ad spend will grow faster than the forecasters think? The alternative is that marketers are hidebound and insensitive to wasting money. You choose.

2) The difference between online and offline ad spend/hour is significant and not shrinking. If this were entirely marketer ignorance, the gap would have shrunk over the last five years, as knowledgeable online people became more mainstream in the agencies. My lead generator friends are arbitraging the difference between the high price of offline advertising that is built into the cost structure of most products and the low price of online advertising, so the gap can't be entirely efficacy. What else is there? I think the answer has to be uncertainty about what, exactly, the advertiser is buying. There was a time when a media buyer could flip through the newspaper or magazine or watch the TV show and know how his ad fit. When he's out buying a million disparate impressions through an exchange or network, he can't. He could be buying one display ad in a sea of ads, or one in a pristine page. But it could be something else entirely.

3) CPMs. The online CPM is the above number divided by impressions per hour. Note that the online spend/hour is pretty steady from 2002 to 2008, between $58 and $66 per thousand hours. The decline in online CPMs can't be attributed to this, unless there was a large increase in ads/hour. This may be so (I can think of arguments both ways, but have no idea what the facts are) but who cares, really? If I spend an hour on the NYT site and they show me 30 ads at a $2 CPM or 6 ads at a $10 CPM, they make the same amount of money: I assume they show the number of ads that maximizes their revenue. So, two hypotheses:

• Fragmentation of online time: people spend less time on each site, spreading the money around and making it seem like less money is being made; and/or
• Ads are being put where there were no ads before so money is being made by people who didn't make money before, and that money is coming out of the pockets of the established online media.

Both of these would lower CPMs because there are more ads. This is the only 'supply' argument that seems reasonable to me.

4) As Niki Scevak noted in the comments yesterday, search is 5% of the time and 50% of the revenue. Since the numbers in the graph are averages across all online, one explanation for declining display CPMs could be increasing search eCPMs. In any case, if Niki's stat is true, then it might also be that we are simply noticing, as the display ad infrastructure is extended, how low display ad CPMs have always been.

-----------
Data sources: US population, US Census Bureau; Internet Users, ITU and Nielsen (last three via InternetWorldStats.com); time spend on internet, USC Annenberg Center for the Digital Future, Digital Futures Report press releases and IAB summary (too expensive to get actual report); time spent on media, Veronis Suhler Communications Industry Forecast (my friends at VSS use to give me a copy of this every year... I need to invent a reason to go visit them soon); ad spend through 2008, CMR/TNS; ad growth 2009-2014, eMarketer.

Two (More) Hypotheses on Low Online CPMs

1. If low online CPMs were a result of oversupply, why haven't CPMs in other media dropped by the same amounts? If oversupply were why prices are so low, not efficacy, then online media would be a substitute for other media and ad spend would shift to the low-cost medium, equalizing prices. This is not what seems to be happening.

But if online ad spend has not yet caught up with online attention, that supply is temporarily exceeding demand. One way to test this would be to see if offline CPMs are increasing (because demand there would exceed supply, driving up prices.) If it were true, it means that online CPMs should increase over the next few years back to "normal" levels.

2. Since media sells attention but is paid by the impression, then the fragmentation of time spent on media might lower CPMs as the "bad money" of low-attention impressions drives out the "good money" of high-attention impressions.