931854_en-US

liliconcom

Bayesian Optimization

I’m looking to see if anyone has run into examples of Bayesian Optimization being applied in the marketing, direct mail, or service industries (e.g., call centers, customer experience, field service, hospitality, government).

If you’ve seen a project, customer conversation, research paper, or even something anecdotal, I’d really appreciate hearing about it. I’m gathering examples to strengthen a Pro Opportunity I have.

Re: Bayesian Optimization

Hi @alisa_h_lowery 

Maybe a bit outside of the industries that you listed, but I know that Netflix does some really sophisticated experiments around their customer experience. I've read some good things from their data scientists in the past. So I thought there was a good chance they are using Bayesian optimization and I googled it. I haven't read this properly but it seems to confirm my suspicions: Supercharging A/B tests for adaptive bitrate streaming with Bayesian Optimization | by Lennart Paduch | Medium

Of course, I have to also point out that these industries are outside our core market, so please be careful investing time there.

Hope that helps,

Phil

Re: Bayesian Optimization

This past ISEA presentation by Nathaniel Stevens may be of interest: https://isea-change.org/event-5669661. The vocabulary is confusing as A/B testing can also be multivariate. 

Re: Bayesian Optimization

Hi @alisa_h_lowery ,

I don't have anything specific to your request, but have had a similar one myself. One of my AI's did a reasonable job in framing JMP Pro's value to the 5 most important publications on the topics my customer is interested in. I used Copilot in this case.

Good luck - and I wouldn't mind hearing about your outcomes

 

Best

Florian

Re: Bayesian Optimization

A little late to the party on this - Something that has worked for me is to look through the examples we already have where there is a decent model and then use that as a formula (or collection of formulas) in the response column(s) of the BayesOpt data table.  So, maybe use Churn, or the donation mailer example, make a really nice model for it, and move the model into a DOE table for BayesOpt to learn from.  Maybe add a little noise to the formula for fun using a random number generator.  

That gives you a packaged story, and might let you slot BayesOpt in an existing demo journal without much rework.