MLFoGrasso - I think Moneyball started to provide a set of predictive projections on what could happen to players. it was far from perfect, but at least gave the A's enough of a starting point to attempt to do arbitrage on future-facing outcomes. Baseball has enough of a scouting component that they're still a while away from truly prescriptive approaches.
Prescriptive analytics is based upon assumptions that lie in predictive analytics and descriptive analytics, and predictive analytics itself is based upon assumptions that lie in descriptive analytics. So MB was about changing the assumptions.
This is a personal bias, but I love prescriptive analytics that are used to identify optimal drug design for personalized drugs, meaning that the drug has been designed both to solve a medical problem and is optimized for personal chemistry and other drug interactions. It's amazing that what used to be an endlessly iterative process with dead end after dead end can now develop an actual best recommendation based on thousands of interactions.
"Question I have for our speaker: Given that prescriptive analytics can only be as good as the logical sensibilities, access to and knowledge of factual context, and similar factors, what flexibility exists in present and near-future prescriptive analytics tools to allow data scientists in an organization to modify the prescriptive analytics tools?
And, more to the point, how do you keep them from screwing it up with their "human" biases/failures? ;)"
Great question. The level of contextualization associated with decision-making is the real differentiator. We're trying to get context that matches human insight and reasoning. Of course, the flipside is that when machines can make human decisions consistently, we have a separate problem...
@James: It does. I used to be big into fantasy baseball several years ago.
But no, it's just a really basic system. You rank your players in order. The automated program just picks from your list top to bottom of players still available. I think there may be some basic, not very intrusive override feature just to make sure you get all your positions, but that's it.
"It seems to me that the most rudimentary prescriptive analytics-cum-automated system that comes to my mind immediately is an automated Yahoo! fantasy sports league draft"
Assuming that you trust the Yahoo rules, that's spot on. The challenge occurs when Yahoo auto-chooses someone you would have never picked for yourself. That's where the pre-analysis data assumptions end up being so important.
Prescriptive analytics provide a specific answer to a specific problem. In a world where we had robots with human-level motion, articulation, and mobility, we could automate the recommendation from prescriptive analytics.
"Imagine if we could effectively combine predictive analytics, prescriptive analytics, and automation with social media? Twitter would be Guy Kawasaki's robot talking with Robert Scoble's robot all day. ;)"
That would be scary. The Guy robot would know exactly what to say to push the Scoble robot's buttons, as well...
Thanks for the great discussion so far. One of the key aspects of prescriptive analytics I didn't mention is that we're limited so far in what we can do there simply because of our trust of automated systems. There's only so much that we currently trust "rules" to do without human context.
I think if someone is going to raise the topic of prescriptive in their organization to the business side, one of the key things to emphasize will be that the system offers several suggestions with probabilities of success. You folks are right, a single, hard answer could be tricky.
Imagine if we could effectively combine predictive analytics, prescriptive analytics, and automation with social media? Twitter would be Guy Kawasaki's robot talking with Robert Scoble's robot all day. ;)
So realistically, what's the difference from prescriptive analytics and automation? I mean, obviously, some analytics don't just turn a switch or something, but do you just automatically follow the prescriptive analytics like an autopilot?
The problem with the basketball analogy is that statistic only measures how often the player makes the shot when the player decides to take a shot. It doesn't demonstrate when the player will be in a good shot-making position, and it doesn't tell us how much of that 50 percent is because of the player's good shot-making positioning and how much of that 50 percent is due to the player's good judgment in assessing when his positioning is good and how much of that 50 percent is due to his actual shot-making skill and type of shot-making skill.
@Lyndon Why not? Doesn't a medical doctor often have multiple options to help you relieve your health concerns? But his expertise leads him to choose what prescription to attempt first. Similarly, the analyst's expertise gives him the the wisdom (hopefully) to choose the best of the available options.
In other words, predictive analytics may tell me what's going to happen in the future, but I still have to make a human decision -- which may or may not be flawed -- regarding what to do about that future event that my analytics are predicting.
Question I have for our speaker: Given that prescriptive analytics can only be as good as the logical sensibilities, access to and knowledge of factual context, and similar factors, what flexibility exists in present and near-future prescriptive analytics tools to allow data scientists in an organization to modify the prescriptive analytics tools?
And, more to the point, how do you keep them from screwing it up with their "human" biases/failures? ;)
Get up to speed with emerging analytics technologies including Natural Language Processing, Edge Analytics, Machine Learning, Real-time Analytics, and Augmented Analytics. These expert-led sessions are for analytics leaders, professionals and business users.
Get started with tomorrow's analytics technology. Sign up today.