Full kelly-criterion sizing maximises long-run geometric growth but produces drawdowns most people cannot hold. Fractional Kelly multiplies the Kelly stake by a constant, usually 0.25 to 0.5.
The trade-off is favourable because the growth curve is flat near its peak. Betting half Kelly gives about 75% of the maximum growth rate with roughly half the volatility; betting quarter Kelly gives about 44% of growth with a quarter of the volatility. Meanwhile the pain falls faster than the return: expected worst drawdown scales roughly with the fraction.
The deeper reason to use a fraction is that Kelly assumes you know your edge exactly. Estimate the win rate 5 points too high and full Kelly is over-betting, which is not merely suboptimal but actively negative-growth. Fractional Kelly buys insurance against your own parameter error.
Related: half-kelly, kelly-criterion, optimal-f, volatility-drag