Using AI to Stress-Test Creative vs Media Budget Splits
Every brand I've ever worked with has said "If you can increase the ROAS, we can increase the spend." However, I've never found that to be a very helpful framework if you need to actually plan for your upcoming quarter. How many dollars should you spend on graphic designers? Or editors? Or creative strategists? If you could just create more ads would that be better or would it be better to just put all of those dollars into FB?
I set out to try and answer that question so I asked for ChatGPT's help and built an app.
Let's take a simple example: Let's imagine you are a head of growth for a brand currently spending $80,000/month on ads. Your current pace is about 4 new ads per week and you estimate that it roughly costs you $1000 per ad (assuming salary costs are included). You need to plan out your next six months.
I came up with this framework for the potential outcome of those new ads called the ad multiplier. In this scenario there are 1 of 10 outcomes an ad can produce. It can be a 1x ad, 2x ad, 3x ad, and so on. That multiplier is essentially ROAS: for every dollar spent on an ad, it generates that many dollars in return ($1 spent on a 2x ad generates $2). But there's a catch. Every ad reverts to the mean eventually. Maybe it’s just less exciting. Maybe it’s ad fatigue. This is what we call adDecay. It's the rate at which the ad loses it's multiplier. For the sake of this example, I have choosen a 0.1 which means we're losing 10% of the multipler each week.
If we assume our initial multipler (aka ROAS) is 3x and we decided to never make another ad again based on this model, our ROAS next will be 2.7x (3x decayed by 10%) and then 2.43x, and so on. Ultimately there is a floor but we haven’t included that in this model.
So you’re currently producing 4 new ads per week, what do you think that multipler on each new ad is? Are two a 1x, one is a 3x, and one is a 4x? I took a stab at what I believed to be a standard distribution for most brands.
Prob: 1x: 0.50049 (50.05%) 2x: 0.25024 (25.02%) 3x: 0.12512 (12.51%) 4x: 0.06256 (6.26%) 5x: 0.03128 (3.13%) 6x: 0.01564 (1.56%) 7x: 0.00782 (0.78%) 8x: 0.00391 (0.39%) 9x: 0.00196 (0.20%) 10x: 0.00098 (0.10%
If we assume our initial multipler (aka ROAS) is 3x and we generate a 1x, 1x, 3x, and 4x in a given week that means our 4x is now our multipler. With a 4x ad and $20,000 in weekly spend, we generate $80,000 in revenue. Assuming next week, all four ads are at a 2x, we’ll keep our current multipler but the decayRate will not effect it. At a 10% decay rate our multiplier is now 3.6x so $20k in spend gets us $72,000. This will continue on until we generate an ad above the current multiplier. We keep this going until we’ve completed 6 months.
Now imagine let’s run that scenario again but with 8 ad generated per week (ad costs get pulled from the total budget). And again with 6 ads generated per week. And let’s do it 50 times for each scenario. This is what I asked ChatGPT to help me with and we created this.
