Quantcast

Posts tagged with web businesses

Many thanks to Sameer Al-Sakran for compiling the numbers.

Category	 Started Funded	 %F	∑Raised	 TC Posts
Advertising	 3972	 631	 16	 $8B	 996
Biotech	 	 2787	 1770	 64	 $42B	 43
Cleantech	 1302	 719	 55	 $33B	 192
Consulting	 3330	 176	 5	 $2B	 444
Ecommerce	 5383	 868	 16	 $10B	 1587
Education	 522	 60	 11	 $½B	 90
Enterprise	 2389	 652	 27	 $10B	 993
Games video	 3992	 910	 23	 $16B	 3625
Hardware	 1726	 613	 36	 $12B	 3274
Legal	 	 306	 23	 8	 $0.1B	 16
Mobile	 	 4101	 1099	 27	 $20B	 4263
Network hosting	 1782	 340	 19	 $8B	 1375
Other	 	 33068	 2022	 6	 $24B	 2501
Public relations 2531	 468	 7	 $7B	 765
Search	 	 1437	 226	 16	 $2½B	 4625
Security	 710	 218	 31	 $4B	 135
Semiconductor	 620	 381	 61	 $9B	 27
Software	 12405	 3039	 25	 $33B	 3733
Web              12830	 2401	 19	 $29B	 14356


Category	 Acqu	 %A	IPOs	%I 	% on TC	 Avg Funding
Advertising	 221	 5½	15	½	 6½	 $2M
Biotech	         332	 12	143	5	 ½	 $15M
Cleantech	 72	 5½	39	3	 5½	 $25M
Consulting	 107	 3	15	½ 	 1½	 $½M
Ecommerce	 188	 3½	16	⅓	 12½	 $5M
Education	 4	 ¾	1	¼	 3½	 $1M
Enterprise	 257	 11	41	1½	 9	 $4M
Games video	 285	 7	30	¾	 11	 $4M
Hardware	 139	 8	81	4½	 6	 $7M
Legal	 	 2	 ½	0	0	 1	 $⅓M
Mobile	 	 275	 6½	34	1	 13	 $5M
Network hosting	 154	 8½	22	1	 8	 $4½M
Other	 	 2325	 7	101	½	 2½	 $½M
Public relations 171	 18½	32	1½	 19	 $3M
Search	 	 57	 4	4	¼	 10	 $2M
Security	 96	 13½	12	2	 7	 $6M
Semiconductor	 119	 19	51	8	 3	 $15M
Software	 1101	 9	110	1	 5	 $2½M
Web 		 827	 	57	½	 14	 $2½M

My favourite number here is the number of companies started. 12,830 Web companies started up and got a Crunchbase profile. Forget about the Facebooks and Instagrams’ buyout package to the founder, that’s the max of the sample. If you’re looking at the lower-50% CVaR, it may be $0 or less.

 

My second favourite number is that, even among the crème-de-la-crème who play these games, they have less than one-in-five chance of either acquisition or IPO.

As you might expect, stuff that’s harder to do and takes more technical expertise (semiconductors, hardware, biotech/cleantech) has a higher rate of success than stuff that can be learned in a year or two by >1% of the population (build a Rails app!). Software seems to be at a disadvantage except enterprise has a one-in-ten acquisition rate, which is quite a gamble with your life but counts as good odds in this low-probability game.

On the other hand, the software companies are much cheaper to start than cleantech/biotech (cleantech has highest avg funding). Web companies are 1 order of magnitude cheaper to start.

P.R. is also a standout, I’m guessing the 18% acquisition rate is acquihires (Sameer Al-Sakhran alluded to this). But still, this reveals that public relations must be an important part of the SF business ecosystem, or else the market is mispricing PR. But I have enough stereotypes about geeks who can’t negotiate that I can explain away the high valuation of smiley PR folks filling the niche none of the cool hackers want to talk about.

Of course, these are “running tallies” not “final fail/success rates”. It would be good to know

  • for the subset that exited, what’s the year of founding and the year of exit?
  • for the subset that didn’t exit, what’s the year of founding?

That might help us guess at what companies have been abandoned. (Did a lot of Web companies—maybe unfunded ones—make Crunchbase profiles for themselves  to put themselves on display and then quit after a few months?) It would also give a more precise idea of the number of years it takes to develop a company to IPO-ability. (“Eating Ramen” is expected for a few months, but what about if it’s half a decade?) 

If I get around to doing my own scrape, I’ll add those things—as well as some ggplots of distributions for some parameters. I’d also like to compare some Crunch-based estimates of success rates with YCombinator and TechStars, etc. That would be hard because of selection effects but still nice to see a side-by-side.

In the meantime, big thanks to Sameer for doing it first.

(Source: TechCrunch)




One misconception I got from the academic theory of finance is that risk and reward go together. You take on more risk, you get more reward. This is formalised in CAPM theory as a higher expected return associated with a higher standard deviation of investment returns.

In reality, ∃ many stupid risks—mistakes, bad ideas, not doing your homework, believing people you shouldn’t believe, taking on a job without negotiating a floor for your own compensation first, or investing in a company that was bound to tank.

Recently, academics have undercut the premise that risk goes hand-in-hand with reward. Perhaps this pill is easier to swallow after seeing "dumb money in Düsseldorf" vacuum up synthetic CDO pyrite (AAA mortgage bonds) spun from BBB bonds—and then find out, publicly, along with the rest of investment Narnia, that the rewards were nowhere near commensurate with the risks.

I’ve seen this play out a little more in private equity, where models of price paths are less influential than common sense, gut reactions, and balance-sheet research.

I don’t know as much about trading. But I’ve read between the lines on the EliteTrader forum and its cousins, and got the sense that, as academic papers that study the matter report: most day-traders lose money on expectation. Their trading capital approaches $0 faster than would be expected merely by the drag of trading fees on a statistical mean of zero profit.

 

Warren Buffett, the world’s best living investor, is in a business where risk and reward are inverted from the CAPM model. (He’s written about it plenty so I won’t repeat him.)

Steve Schwarzman, another of today’s most successful investors, says in this lecture that he focusses on figuring out every possible angle beforehand, not making any mistakes, controlling every risk and making sure he wins. I’ve read similar things in interviews where Mark Zuckerberg or Peter Thiel talk about “making their own luck”. A lot of questions and decisions go into running a business, and I find it entirely credible that getting that right increases the chances of success—that if an omniscient Arjuna were starting a company today, he would have a very high chance of success (again, what does “chance” mean? Where do the “possible worlds” come from?)

Insurance and reinsurance companies, though they may serve a social function, aren’t actually concerned with actuarially converting risk into reward. They’re interested in collecting as many large premia as possible for risks that will never harm their balance sheet. Why do you think they have three times as many claims adjusters as actuaries? Si guarda al fine.

Michael Price, one of the stars of The Vulture Investors, bought a loan to a bankrupt company for 47¢ on the dollar, covered 15¢ immediately with cash, plus 45¢ in bonds plus 23% of the post-bankruptcy company. He needed the bargaining skills and the capital to buy out other bondholders and negotiate a good rate for 

One last classic example: McDonald’s. Ray Kroc saw a huge return on investment but only took smart risks, doing less of the hard work and spending more time being successful. Mr. Kroc didn’t finish college with a bright-eyed hope to be the world’s greatest entrepreneur (cf. YCombinator). He sold Dixie cups for 17 years before he saw an opportunity—in a B2B space—with high returns and low costs. (Selling malt mixing machines back when malts were the profit centre for burger joints—a malt might cost as much as sandwich + fries, or even as much as sandwich+fries+coffee.) The malt mixer business was a classic play; it would earn 100% checkmarks from a Business 101 textbook. Only after Ray Kroc saw another opportunity related to the business he was in, did he buy up the MacDonald Brothers’ restaurant and multiply it out. Again, this is a textbook private-equity move: find a proven business where somebody has completely figured out how to make money hand over fist, such that the only other thing they need is more money. (Obviously this is very different from an entrepreneur with an idea who just wants some money or thinks their failing idea would be saved if only they had more money.) You provide the money and collect the multiplied profits, i.e. you take on the easy part of the problem, negotiate the terms so you get a huge return on solving it, and then you’ve done little work for great reward. That’s a “smart risk”, not a correlation of risk and reward.

 

We could probably go back and forth with examples of titanic companies. (Sure, Ted Turner threw massive sums into a money pit for over a decade before seeing TNT and its siblings become profitable.)

But still I think the overall message of risk~reward is wrong. There are smart risks, and there are dumb risks. Don’t expect that just because you did something risky, that the return will be good. Work smart, not hard. Cover your *rse and check yourself before you wreck yourself.




People have been saying that ads are going to be more targeted on the Web since the 1990’s. The more data we give to corporations, the better they can “serve us” ads which “meet our needs”.

I read that again on Scott Adams’ blog a little while ago. I’m doubtful that more data → more targeted ads, because Gmail and Facebook each know about a jillion things about me, and I still see irrelevant ads frequently.

Here are my made-up guesses of why ads are still irrelevant:

  • it’s cheap to serve irrelevant ads (true in email)
  • no matter how much data, people don’t currently know what to do with it — their models aren’t good enough to weed out false positives (non-customers)

I don’t bring this up to complain, I bring it up because I wonder if a team of mathematicians, psychologists, dataists / statisticians, and marketers, couldn’t do better ad targeting and make a business of it. Maybe Mathematical Capital would invest in a business like this.




10 Plays • Download

Most internet businesses I have seen think they are going to make money by selling advertising.  I’m not the only person who finds that disheartening.  It’s no wonder that financial journalists decry ever more iterations of investment bubbles into websites that will, in theory, make money.  Twitter has no customers and Facebook’s customers are advertisers, not you and me and all of the other time-wasting browsers.

The audio story above talks about some goods that are actually consumed on the Internet.  Here’s what the more down-to-earth non-American web companies are investing in:

  1. Gambling
  2. Games
  3. Porno
  4. Anything else?

Yet my American internet usage is dominated by Twitter, Facebook, Google, and now Tumblr and Reddit.  It’s a frothy world and very hard to understand — and very hard to predict what will get you rich.

Maybe the Internet isn’t so different than other media after all, though.  Malcolm Gladwell writes Blink and makes a fine dollar, but then he does speaking engagements and mints money.  Nobody knew before that they desperately needed a consultant who could tell them how to improve their snap judgments, but he convinced heads of corporations it would be worth paying him tens of thousands per hour to multiply the bottom line.

I guess there are really lots of situations where one product promotes another, like a loss leader that must be maintained but either loses money outright (like Gladwell maybe lost his time outright while writing his first book) or makes less than it “could”.

Now, one last tangential thought.  If I came into a new business with my MBA and my econometric tools and did a regression on the loss leader, I might find that the item should be cut.  Probably everybody knows not to do that — but there is a more general sense in which the whole is greater than the sum of the parts.  I have been cogitating on how models with no single variables — only interaction terms — might be used in more situations.  And that must wait for another time.

UPDATE: James Grahn and Troy Alexander pointed out more online businesses where customers can be convinced to pay:

  1. Storefronts (Amazon, Buy.com, iTunes, Steam, Impulse) —JG
  2. Server businesses:  hosting, registrars, mail (GoDaddy, BlueHost, Tucows / OpenSRS, Domai.nr) — JG
  3. Cloud computing — JG
  4. Netflix’s Watch Instantly — JG
  5. Dating (eHarmony, Gay.com, Chemistry.com) — TSA
  6. Craigslist — TSA