The thought engine at the core of the Khepri machine-learning algorithm was originally started from a similar concept to the Multi-Armed Bandit method of calculating results. The name “multi-armed bandit” describes a hypothetical experiment where you are faced with several similar slot machines (sometimes referred to as “one-armed bandits”). In this scenario, while the slot machines are similar, they also do have some variables that may affect their payouts and yield different results when you play them.
Let’s say for example, one of the machines costs $1 per play with the biggest jackpot but lower odds of reaching a winning result, while another machine costs only 5 cents to play and has a lower jackpot payout but offers a much higher chance of winning at least something each time. Given these facts, which machine would be a wiser place to put your $100 dollars in bets? Now add in the possibilities offered by a dozen other machines that each add new variables to the mix, to spawn even more variations on the odds and payout amounts previously presented. Then try to factor in complications of caused by changes in the best choice based on elements like time of day or day of the month. It can quickly become a dizzying array of options that confound even the smartest person with access to all of the information.
It’s actually not just a difficult calculation; it becomes an impossible one because you do not know the exact odds of winning and cannot anticipate every possible variable that may be affecting the results. For example, did someone else win recently change the odds on a particular machine? The only way to discover the odds is by testing the machines, and using the results of those tests quickly is the only way to gain a benefit before the odds change again.
To be effective at finding the best machine and earning the maximum revenue from it with your budget, your strategy must include exploration and exploitation. Exploration is the process of by applying some of your money to testing, and exploitation is the process of using the rest of your budget to monetize the option that pays best. The difficulty manual calculations cannot overcome is the balancing those two functions against each other. That’s because manual calculations require you to take a binary approach by experimenting first, and then calculating which option to exploit when your testing is complete. That may cause you to create waste as well, exploring too much and having less of a budget available to exploit the winner. Even worse, you may make a poor decision after exploring and only realize it when your budget is wasted in attempts to exploit poor choices.
Probability matching with Khepri is a much better solution because it’s not binary and it adapts dynamically to the results of each test. Diverting your traffic to the best paying option at every moment during the test instead of waiting for a final result, and optimize its own decisions at every increment along the way to filter out any false positives. It’s almost like recalibrating your choices while pulling the handle on each slot machine without needing to wait and see if you have won or lost your bets on all of them.
As more variables are added and the right solution becomes increasingly difficult to calculate, the ability of anyone to provide a manual answer to the question is diminished. In fact, the casino may close or even go out of business long before a man with a pen and paper is capable of calculating the best place to wager his hundred-dollar budget. Khepri improves your odds as you make your bets because it can make even the most complex determinations from your data in real time!
For those reasons, businesses faced with these kinds of complex variability decisions have become reliant on highly developed mathematical models for managing their budgets– and we are confident that the proprietary Khepri algorithm provides the greatest degree of efficiency, effectiveness and ease of use of all.