Sorry. I am not going to cite any research.
My question is... What data? What are you looking for? What is the design decision you want to improve?
My assumption is... depends on what you are looking for. It sounds pretty obvious. Let's see a small list of design decisions.
Let's think about color. That is one design element. What is the target audience? Are they kids? do you want it to look feminine? do you want it to look moody and surreal?
Another topic. Let's think about fonts. You probably want to look "trendy" fonts. Of course, there are two ways to do this search. One opinion based, like reading someone's blog, or using, for example, a font download website and look at the stats. If you use the stats you probably want to do the opposite as the tread marks and look for a "forgotten" font.
One data can be historical data... what fonts were used in the art deco period in Europe? That probably sounds pretty obvious... but it is data.
I would assume there is data on how an image is scanned by the observer. If you see a face looking at you on an add it would drive your sight to it. Yes, you can look for the data showing that probably an anthropological-evolutionary paper. There can be research on eye movement scanning on an image, etc...
A design is full of data, of homework must be done prior to design.
Yes, you could just do a poll and see what option gets more "likes" but if you do not control your samples, your variables you will probably not learn what is the one variable people like more.
Another thing would be preparing a correct poll. This StackExchange website is an example of how people react to "generic" questions about a design decision.
"What do you think looks prettier?" vs "Which color between this two options drives your attention to the name of the product?
Do not give too many options to users. People will be confused.
One important step in design is, do not assume in your mind how it looks. Do a quick draft and actually SEE it. A lot of design ideas are discarded by a designer prior to seeing it. If it does not work, you now can discard it.
What an AI will probably give is a more organized set of design decisions on one report.