For the last few days, I have been digging deeper into ads engineering, and time and again, I am stumbling on Poisson distribution.
Clicks and impressions follow this distribution and hence modeling them as Poisson distribution is used to
- identify unusual spikes or drops in engagement
- flag click fraud when observed significantly deviates from the expected
- estimate required impression volume to achieve desired click counts
It might not be the best to model, because edge cases are more common than we think. Despite this, it is one of the most computationally efficient and “good enough” for most cases.
This makes it extremely suitable for real-time bidding and making high-volume ad-serving decisions.
and, this is why we study statistics in engineering. Whenever you find some time, please explore this deeper; your fav LLM is your friend here. Do not run away from maths, at least attempt to understand it.
⚡ I keep writing and sharing my practical experience and learnings every day, so if you resonate then follow along. I keep it no fluff.
youtube.com/c/ArpitBhayani