Quick Answer
The Kelly criterion is a bet-sizing rule from a 1956 information theory paper: stake the fraction of your bankroll that maximizes the expected logarithm of your capital, which reduces to your edge divided by the odds. For a founder, edge is evidence a channel reaches buyers, odds are what a unit of work pays back if it works, and bankroll is the hours you can afford to lose. It sizes the bet and it never picks the channel.
Most founders do not have a channel problem. They have a bet-sizing problem.
With ten or twenty hours a week and a budget that would embarrass a mid-market marketing team, the expensive mistake is rarely picking the wrong channel. It is committing a quarter to weak evidence, or spreading so thin that nothing gets enough repetition to compound.
The Kelly criterion is the cleanest frame available for that decision, and it is also the most abused. It was written for a communication channel rather than a marketing one, and the gap between those two is where founders get hurt.
This covers the formula, one worked example sized in hours, and the four places where the analogy stops holding.
What the Kelly Criterion Actually Says About Bet Size
The Kelly criterion says to stake the fraction of your bankroll at which the expected logarithm of your capital is largest. That fraction turns out to be your edge divided by the odds. John Kelly published it at Bell Labs in the Bell System Technical Journal in July 1956, as a result about information rate rather than gambling.
Written out, it is small. If winning pays B for every unit you stake and losing costs the stake, and the bet wins with probability p, the Kelly fraction is B times p, minus the chance of losing, all divided by B. Ziemba and MacLean give the memorable version: edge divided by odds.
For an even-money bet the formula collapses to the edge and nothing else. A bet that pays one for one and wins 55 percent of the time gets ten percent of the bankroll, not half of it.
The number the rule maximizes is not expected value, and that distinction is the whole idea. Betting everything on a favorable coin does maximize expected value, and Kelly pointed out that it also leaves you broke with probability one if you keep playing. The original paper ran to ten pages and never once used the word marketing.
Translating Kelly Criterion Terms Into Marketing Inputs
The Kelly criterion needs three inputs, and marketing has an honest equivalent for each. Edge is evidence the channel reaches people who can buy, odds are what an hour of work pays back when the channel works, and bankroll is what you can lose without stalling the company.
| Kelly term | What it is in a marketing bet | How you estimate it |
|---|---|---|
| Edge | Evidence this channel reaches people who can buy | Qualified replies or demo requests per unit of work |
| Odds | What an hour of work pays back when it works | The ceiling: search demand, audience size, or the reply rate a list sustains |
| Bankroll | Hours, budget and patience you can lose | Runway multiplied by the hours a week you can protect |
| One bet | A fixed block of work with a decision date | One month of effort on one channel, written down |
Bankroll is the input founders get wrong. It is not this week's hours. It is the total you can spend and lose before the runway makes the decision for you.
Counting a pre-revenue budget in hours rather than dollars makes that number visible, and it is usually smaller than it feels. Content marketing for tech companies has the slowest read of anything on the list, because the qualified replies arrive months after the work that caused them.
How to Size a Marketing Bet With the Kelly Criterion
Size a marketing bet with the Kelly criterion by writing the bet down, estimating the edge from a cheap probe, dividing that edge by the odds, halving the answer, and fixing the date you re-estimate. The arithmetic takes five minutes. The estimate is the whole job, and the steps below are ordered so the weakest number gets the most scrutiny.
1. Write Down What One Bet Actually Is
A bet is a fixed block of work with a decision date, not a channel you are running indefinitely. Write it as one sentence: sixty hours of founder-led search work across three product-problem pages, judged at ninety days. The mistake is leaving it open-ended, because a bet with no end can be neither sized nor lost.
You know this step worked when someone else could read the sentence and tell you what would count as a failure.
2. Estimate the Edge From a Cheap Probe
Edge is the chance this channel produces a repeatable stream of qualified conversations. The only honest way to get at it is a probe small enough that being wrong costs a week.
Thirty targeted outbound messages give you a reply rate. Three pages and a month of patience give you impressions and clicks in Search Console, which Google defines as how often someone saw a link to your site and how often someone clicked it.
A search bet is the easiest of these to read honestly, which is why an AI SEO service is normally sized off Search Console numbers rather than off a pitch. The same arithmetic decides what month one with a B2B demand generation agency should look like: a probe with a number attached, not a retainer with a hope attached.
3. Divide the Edge by the Odds
Say your bankroll is 300 hours across two quarters. The probe leaves you with a 40 percent chance that the channel pays back three hours of compounding work for every hour staked, and a 60 percent chance those hours return nothing.
Edge is 3 times 0.4 minus 0.6, which is 0.6. Divide that by odds of 3 and the Kelly fraction is 0.2, or sixty hours.
Sixty hours across a quarter is about five hours a week. That is a real commitment and nothing like a bet-the-quarter one, which is exactly the point of running the calculation.
Now move the estimate and watch what happens. At 30 percent the fraction falls to about 0.067, or twenty hours. At 25 percent it is zero, and the bet does not get placed at all.
A fifteen-point move in a number you guessed takes the answer from sixty hours to none. The formula is more sensitive to your estimate than your estimate is reliable, and that is the argument for the next step.
4. Cut the Answer in Half
Serious practitioners stake less than the formula says, and the practice has a name: fractional Kelly. Half of the calculated amount gives up a modest share of long-run growth in exchange for a large reduction in how bad the bad runs get. Given that your probability came from thirty emails and a hunch, half is generous.
Ziemba and MacLean list miscalculation of probabilities, short-run losses and bad scenarios among the reasons to bet well below the number the formula produces. So thirty hours, not sixty.
5. Set the Date You Re-Estimate
The fraction is a position you hold until the evidence changes, not a plan you execute. Put a re-estimate at the end of the block, recompute the edge from the numbers the probe actually produced, and resize from there.
If the edge went up, the next bet is larger. If it went to zero you stop, and you have lost thirty hours instead of a quarter. The common failure here is quiet renewal: the block ends, nobody recomputes anything, and the bet rolls forward on its original estimate for another six months.
Overbetting a Channel Costs More Than Underbetting It
Overbetting is the costlier error, and the Kelly criterion is unusually blunt about why. Growth and safety both fall once you stake more than the Kelly fraction, so there is never a reason to wager above it. An investor staking exactly twice that amount ends up growing at nothing above the risk-free rate.
In marketing terms, doubling the justified bet does not double the learning. It buys the same information, later, with less runway left to act on it, which is the founder version of risk of ruin.
The reason founders overbet is almost always one strong signal read as an edge. A launch is the classic case, and a spike is not evidence that a channel compounds.
A founder buying an AI SEO service or a B2B demand generation agency should ask for the same shape, which is how Better Marketing starts: one month, one probe with a number on it, and a date to decide.
Where the Kelly Criterion Breaks for Marketing Bets
The Kelly criterion breaks in marketing at its assumptions, not at its arithmetic. Kelly named the conditions himself in the closing paragraphs: "The essential requirements for the validity of the theory are the possibility of reinvestment of profits and the ability to control or vary the amount of money invested or bet in different categories."
Marketing clears the second requirement easily and only half clears the first. Four gaps matter more than the formula does.
- Probabilities are guesses. Kelly assumed a probability that came out of a communication channel; you have a reply rate from thirty emails.
- The bets are not independent. Content lifts outbound reply rates and a newsletter makes a search page land, so the channels move together.
- Odds are never posted. Nobody tells you what a LinkedIn post pays, so the ceiling is an estimate sitting on top of another estimate.
- Some bets are not repeated. A launch happens once, and a rule built for the long run has nothing to say about a single trial.
Ziemba and MacLean make the same point about markets and it lands harder here. Errors in the estimated returns matter roughly twenty times more than errors in how the assets covary, and when conditions turn suddenly, those correlations approach one.
Their own summary is the honest one: "automatic use of the Kelly strategy in any investment situation is risky and can be very dangerous". Edward Thorp, who took the rule from blackjack to the stock market and dubbed it Fortune's Formula, is the reason most founders have heard of it at all.
So the counter-case is a bet with no second round. For a single launch or one conference you either can or cannot afford the whole stake, and the Kelly criterion has nothing useful to say about it.
Kelly Criterion Sizing Compared With a Fixed Budget Split
Kelly criterion sizing and a fixed budget split answer different questions. One asks how much this particular bet deserves. The other asks how the year divides, and most founders need the first while being handed the second.
| Factor | Kelly criterion sizing | Fixed budget split |
|---|---|---|
| What sets the size | Evidence of edge, divided by the ceiling | A percentage agreed in advance |
| What it needs from you | A probe and an honest probability | A plan and a calendar |
| When the evidence is thin | Sizes down, or to nothing | Spends anyway |
| Main failure mode | Overconfidence in the estimate | Funding a channel that never earned it |
| Where it fits | One channel, one decision date | A team with several proven channels |
A fixed split is easier to defend in a board meeting. That is one reason content marketing for tech companies is so often funded at a flat percentage and then judged on how much got published rather than on what it reached.
The ceiling is the input that turns this from a debate into a number. The click math behind a startup search bet is how you work out what the channel could plausibly pay before you decide what it gets.
Size the Bet to the Evidence You Already Have
Stop asking which channel to love. Ask how much evidence you have, how quickly you can get more, and how much of the runway it is rational to risk on the answer.
The Kelly criterion will not pick the channel and will not fix a dishonest probability, but it will stop you betting the quarter on a hunch. That restraint is how Better Marketing starts with founders: small and measured first, larger once the evidence has earned it.