I guess I should have been clearer. I think having the right formulation of the problem helps both approaches. The algorithmic approach is one that tends to rely more on quantitative/statistical properties and less on a rule of thumb.
Thanks for the feedback. I'm still a little confused, because in my understanding heuristics are also based on statistical properties; e.g. Is there a strong correlation between property x and y as measured by the amount of click-throughs for example.
Both approaches deal with quantitative properties. But one assumes that causes are too difficult to map out and will try to find the most useful correlations, the other will be frustrated by how hard it is to accurately model reality and will end up using heuristics to whatever degree necessary to patch up their incomplete model. I prefer the second, but I guess the author of the article is correct: Whatever works for you...
Yea it's a bit tough to figure out the precise language for my point so that's why I wanted to show an example but that wasn't that much better.
I guess it does translate into something more along the lines of having something that you can explain and back up vs something that feels right. Having some basic heuristics to adjust the algorithm would be the latter but modeling out the actual environment would be the former. There's a time and a place for both but the idea is that for longer term projects you want to take the more rigorous, more difficult approach.
I think the biggest value is framing the problem properly and understanding the space. Once that's done the approaches can be modified and switched around pretty easily.