Data can easily miss a whole lot of subtlety. Subject knowledge, theory, and intuition can capture much more of that subtlety, at the cost of being more vulnerable to bias.
Making decisions purely based on data misses out a lot. And can still lead to bias, not caused by the decision maker, but by the data gathering process.
In my mind, data exists to detect bias, and illuminate interesting phenomena. But decisions should ultimately be based on in depth knowledge and theory. And leeway should be given for those to be somewhat misaligned to data.
Data can sometimes reveal what an expert spent years to learn and sometimes much more. So when used as an exploratory tool, it has good benefits. And to a degree, it can predict future outcomes to a good degree if we have a great theory about the target domain that explains almost everything. Unfortunately, this tempts people to try using data in domains where their expertise doesn't justify their predictions and then it becomes a gamble where the winners get a lot of publicity.
I saw a documentary last year on YouTube about how F1 teams are using data (ML) to make decisions on each aspect of the race. On the last lap, they decide to trust their experience and intuition rather than listening to the ML suggestions, which costs them a couple of positions. Maybe not today, but we are not far away from the time when the F1 drivers stop listening to "humans" ;)
When companies try to become more data-driven, something that I find very often is the tendency to just follow the initial gut feeling process but this time find some supporting data without actually be open and objective.
But I also acknowledge the fact that data is what it is: A pile of information trying to gather via imperfect processes and imperfect representations.
I constantly get back to something I did hear around the topic of marketing attribution couple of years ago:
"The new right is the less wrong".
Clearly you can make many mistakes gathering and analyzing data. But your domain knowledge should also help you identify clear mistakes and improve on them. Which then leads to a better data quality (hopefully you stayed objective enough to not create large biased data points).
But for me the most fundamental advantage remains:
Bringing data makes it easy to test and verify assumptions. You are bringing the full package to the table instead of a hard to qualify able / quantifiable feeling that often times is not easy to communicate.
The last element on this Data vs Human input: The limit quite often for algorithms to work perfectly is having all the information available. That is in many situations a real challenge, so the domain knowledge helps to fill in gaps.
Data can easily miss a whole lot of subtlety. Subject knowledge, theory, and intuition can capture much more of that subtlety, at the cost of being more vulnerable to bias. Making decisions purely based on data misses out a lot. And can still lead to bias, not caused by the decision maker, but by the data gathering process.
In my mind, data exists to detect bias, and illuminate interesting phenomena. But decisions should ultimately be based on in depth knowledge and theory. And leeway should be given for those to be somewhat misaligned to data.