In my last post, The Product, Part II: Technical architecture and the innovator’s paradox I talked about the importance of staying in the game and linked to a Wikipedia article on the Kelly criterion. In the comments, entrepreneur and physicist Max Skibinsky took the idea literally and used the Kelly criterion to calculate the optimal burn of a startup. I was so impressed with Max’s comment that I imported his Google spreadsheet into Excel and played around with it: here is an editable copy of [my updated version] Max’s spreadsheet.
Max’s math:
p = probability of success
b = payout odds per kelly
F = funding
V = valuation
M = valuation multiplier on “win”
R = burn rate per time frame
T = time frame units to develop and prove
What we can learn about optimal burn from the Kelly criterion.
You obviously shouldn’t take this too literally. But I do find that it is a very interesting reality check.
Assumptions in the spreadsheet:
+ Capital raised to $2MM on a $6MM post money valuation.
+ 15% chance of any experiment returning a 10x increase in valuation.
+ 9 months to build and test each experiment in the market.
+ $100,000 per head (includes salary, benefits, rent, computers, marketing).
Using Max’s spreadsheet which is based on the Kelly criterion’s probability of maximizing long-term returns, the optimal monthly burn is $32K, which would cover 4 heads. This would give you capital for 7 experiments.
A few brief thoughts:
For any hit driven (or wildly innovative) business, you should assume that [at least] your first experiment will fail. This will remove pressure and allow for maximum flexibility. It also drives how you should build your product and manage your finances. It also drives the following recommendations:
1. Keep burn very low until you have proof of traction.
Everyone knows this intuitively, but the vast majority of startups spend an order of magnitude greater than their target Kelly burn. You can reduce burn by hiring fewer people, keeping salaries low, working long hours, and hiring very productive people. Most people focus on keeping salaries low bit, but my experience is that hiring a few exceptional people at higher salaries is cheaper than hiring more [less productive] people at lower salaries.
2. Raise more money than you need.
Easier said than done — but if you have the opportunity to raise a bunch of capital, you should seriously consider doing so. Figure out the optimal number of people needed to run an experiment and use the Kelly burn spreadsheet to impute required capital. The cost of giving up more equity early on is often more than offset by the increased flexibility to take chances. There is obviously some equilibrium point in there between loss of present value as a result of taking too much equity capital (for the entrepreneur) versus loss of present value as a result of taking too little capital and putting too much capital into a single bet or few bets. Many entrepreneurs can’t raise more capital, but those who can should.
3. Increase the probability of success on each experiment.
This is clearly the highest leverage point in the model. You can increase your odds of success by (a) picking a big existing market (rather than trying to invent a market, reinvent an existing one); (b) recruiting a killer team; and (c) picking great investors.
You can also increase your odds of success by building and shipping product quickly, by instrumenting your site / product so that you can run tests and make data-driven decisions, and by killing failed experiments quickly.
For an entertaining history on mathematics, information theory, economics, gambling, and the mob check out Fortune’s Formula.

9 Comments
June 28, 2008 at 3:41 pm
[...] fascinated by the recent ‘burn-rate modeling’ over on Laserlike, which would be accurate only if every experiment conducted by a start-up was a completely discrete [...]
June 28, 2008 at 5:26 pm
Mike:
This is a very cool formalism. About 18 months ago I posted something called Fail Fast, Fail Often which described something related in conjunction with Munjal’s changes at Like.com. It’s all about the gamble inherent in consumer web sites. (http://earlystagevc.typepad.com/earlystagevc/2007/01/fail_fast_fail_.html) But this is clearly a more elegant form of the model.
June 28, 2008 at 6:40 pm
Mike,
This is a great way of looking at optimizing burn in the early stages, and it is all the more critical for consumer internet business (than enterprise or business apps). I think, while everyone more of less understands the importance of this, few have figured out the best ways of truly optimizing burn at every small milestone from a initial core prototype code to early alpha, beta and so on. I am building BetterLabs (http://www.betterlabs.net) with that focus of helping founders to build, test, iterate on a milestone level, where we have significant vested interest in the success of the product with equity partnerships. I believe the teams that figure out a sustainable model for optimizing burn until they hit the “product version that works” will win.
June 29, 2008 at 1:42 am
Very good insight.
It’s interesting to speculate about how the market of VC-backed startups would change if the Kelly criterion were applied more often.
I have a feeling that there would be far fewer consumer internet startups, and that the few that did exist would get most of their funding from a very small syndicate.
On the other hand, I think there would more startups in other technological areas, although again I think there would be less syndication.
June 29, 2008 at 6:01 am
“Log Optimal” investment approaches, aka Kelly, is not a panacea.
Stutzer says it well :
It often invests very heavily in risky assets, which the [...] possibilities that invested wealth will fall short of investor goals, even
over the multi-decade horizons [...] The Kelly strategy never risks ruin, but in general it entails a considerable risk of losing a substantial portion of wealth.
Stutzer, the Journal of Econometrics:
http://pages.stern.nyu.edu/~dbackus/Exotic/1Robustness/Stutzer%20large%20dev%20JEc%2003.pdf
July 16, 2008 at 3:23 am
[...] awesome subtitle of “Free Ideas. Just Add Execution”, has a great post titled, “Optimal startup burn rate and the Kelly criterion” that is definately worth a read ~ especially to my friends running small software driven [...]
August 22, 2008 at 10:15 pm
[...] Hive7’s own CEO & Physicist Max Skibinsky used the Kelly criterion to formulate the optimal startup burn rate. [...]
September 24, 2008 at 8:37 pm
Very interesting thought exercise. I’ll use it in my experience to see if I can gain additional insight. I’m a little wary of formulas but I appreciate the latticework approach of taking knowledge from one discipline and applying it to the most exciting discipline of all, entrepreneurship!
October 16, 2008 at 4:54 am
[...] Backlash against outsourcing Reduced wages and a devalued dollar will create incentives for companies to hire locally in America. Furthermore, politicians will gain favor thru protectionist measures and may make it more difficult for companies to outsource jobs abroad. This will likely lead to a contraction in the outsourcing practice at least for US-based companies. Also some companies may focus on fewer, higher performing employees than lots of low costs employees (as Mike Spieser suggests). [...]