What I Learned Burning $14k on YouTube Ads for Candy Japan

Re:What I Learned Burning $14k on YouTube Ads for Candy Japan by sfifs on Sunday February 25, 2018 @06:53PM (#16461965)

There actually are ways like market mix models and multi touch attribution models to account for impact of those "late conversions" across multiple channels but requires fairly sophisticated data integration and modeling and is generally worth it only if you are optimizing millions of dollars in ad spend. The tools Google/Facebook currently provide are not really sophisticated enough for such work.

Source: I work on these techniques

[ Parent | on HN ]

Re:What I Learned Burning $14k on YouTube Ads for Candy Japan by istorical on Thursday March 01, 2018 @03:28PM (#16495918)

I work in analytics in the sales and marketing department at a late stage startup. I was recently tasked with moving from just reporting to working in modeling and prediction.

A first step I thought of was to build some multiple variable models / apply multiple linear regression to try to measure the relative influence of our marketing campaigns and touches to a sale.

With your experience in this area - are there are particular models you recommend trying to build as a first step for someone who's more of a beginner in modeling and predictive analytics? Or business questions you find go the furthest within this specific domain of sales and marketing?

[ Parent | on HN ]

Re:What I Learned Burning $14k on YouTube Ads for Candy Japan by KodiakLabs on Sunday February 25, 2018 @07:09PM (#16462030)

Could you possibly point to by resources on these models?

[ Parent | on HN ]

Re:What I Learned Burning $14k on YouTube Ads for Candy Japan by sfifs on Sunday February 25, 2018 @07:25PM (#16462107)

While details tend to be proprietary, MMMs are well established and lots of academic papers online. For instance

https://web.archive.org/web/20120813035903/http://faculty.fu...

High quality multi touch attribution models are still fairly new and I've not seen much public literature on the designs. We've had to spend a fair amount of time in development ourselves vs. relying on existing methods.

[ Parent | on HN ]