I was browsing online to find an answer to my question and landed here... so I thought: I could just ask!

Context:

I am a marketer and currently working on a tech project to automate and speed up real-time data analysis and marketing suggestions. I did advanced statistics at uni and recently completed an online course on Machine learning. Said that, I wouldn't call myself an expert, but just somebody who knows what we are talking about .

The Problem:

I am working on a type of paid traffic that is very dynamic, meaning that "quality" and "volumes" change very frequently and with it my Conversion rate (=sales/clciks). This means that across the day I have to make several changes to adapt my spending.

What I need to find out (and I can't):

Is there a statistical way I can predict how many clicks I have to "wait" before I can assess my current conversion rate accurately?

In example with random data:

Hour - Clicks - Sales

00:00 - 54 - 0

01:00 - 156 - 1

02:00 - 71 - 2

03:00 - 352 - 2

04:00 - 215 - 1

total of the day:

Clciks: 848

sales: 6

CR: 0.7%

Now, Is there a formula to identify if the data at x hour is enough to assess my current CR%? Is there a way I can tell, based on my CR, how many clicks should I wait to have a high probability to be correct?

Example 1) are the 215 clicks at hour 4am enough to assess my current conversion rate?

Example 2) I know that the 71 clicks and 2 sales at 2am is not enough to assess my current CR%, so I normally sum up the numbers from 12am to 2am. But still, what if I just need the data from 1am to have a good chance to be right?

Where is the problem?

Because is real-time analysis, I can't use static values like "today_conversionRate" or "last_3_Hours", clicks volumes and CR changes constantly, so, if using historical data is necessary because of low volumes, I need to use it as little as possible. (the more I go back, the more uncomparable my data is)

My idea for a solution?

If there is a formula I can use to predict how many clicks I need to consider for accurately assess my current CR%, then I can write a code that - in case there is not enough data (like 2am) - will look back summing up hour by hour until it gets the x amount.

Anybody smarter then me (well, that's easy) to solve this?

Available for any question or clarification:

Many thanks

Matt

PS: forgot to say, I pay with lots of gratitude as I don't earn anything myself out of solving this