...
Input your goal and per-channel and total budget spend constraints, and receive a budget that is optimized for your selected target (profit, ROI, or revenue)
...
Once you’ve selected your KPI you can adjust the spike calendar for the time period you are optimizing for. Having an accurate spike calendar is important as spikes will make it more or less difficult to hit your objective. Spikes shown in red are spikes that have not happened yet and can be deleted if they are no longer going to happen.
🤔 Strategy
Conservative In the conservative case, the model will primarily include channels that have low uncertainty
Moderate In the moderate case, the model will primarily include channels that have low uncertainty but may also explore channels with greater uncertainty and the possibility of greater returns
Aggressive In the aggressive case, the model will optimize for returns rather than minimizing uncertainty. This means the model will explore channels that may have greater uncertainty but the possibility for greater returnsgeneral, the optimizer iteratively makes small nudges to your budget that help get it closer to your goal. It repeats this process over and over until you have your final recommendation. The Recast model doesn’t just return one simulation of what might happen in the future, it returns hundreds of simulations of what could happen. These hundreds of simulations are what we use to provide things like uncertainty around channel ROIs/CPAs. The important thing to remember is that the strategy you select impacts which of those simulations we’re trying to optimize.
Strategy Options
Base: targets the mean of all the simulations and tries to make budget changes so that the mean outcome is as good as possible. Good for making sure your average result is as good as possible.
Conservative: targets the 20th percentile and tries to make budget changes so that the 20th percentile outcome is as good as possible (and 80% of simulations will be even better than that). Good for making sure your “worst case” scenario is as good as possible, regardless of how good the “best case” scenarios are. “Hedging your bets”.
Aggressive: targets the 80th percentile and tries to make budget changes so that the 80th percentile outcome is as good as possible (whether or not the bottom 80% of simulations are poor). Good for making sure your “best case” scenario is as good as possible, regardless of how bad the “worst case” scenarios are. “High Risk, High Reward”.
...
⚠️Conservative and Aggressive both target percentiles rather than the mean. Percentiles put all the simulations in order from smallest to largest. Then for an Xth percentile, we look for the simulation below which X% of the simulations fall. When you calculate a mean, every single simulation influences that mean, but when you calculate a percentile, it’s looking at a single simulation (whichever one is at the 20th or 80th percentile). In math-y terms, the mean is a smooth function, percentiles are not. The consequence of this is that small changes in the input of Conservative and Aggressive optimizations can cause large changes to the output. We recommend using Base if consistency between different scenarios is important.
Intuition
...
Here the gray vertical lines represent the possible forecasted KPIs we get from our hundreds of simulations, the green dots represent the 20th percentile (a “worst case” scenario: 80% of simulations forecast a higher KPI than this one) and the red dots represent the 80th percentile (a “best case” scenario: only 20% of simulations forecast a higher KPI than this one).
You can see that Budget 2 has a higher floor (green dot is higher than in Budget 1) but a lower ceiling (red dot is lower than in Budget 1). The Aggressive setting will prefer Budget 1, because its “best case” scenario is better than in Budget 2. But Conservative will prefer Budget 2 because its “worst case” scenario is better than in Budget 1. Base would consider these Budgets similar, because their mean predicted KPIs are similar.
...
💸 Total Spend Constraint
...