@David - Management is usually investing (at first) in Data Science or similar name because they believe in the potential for upside (top line growth, bottom line savings).
So I'm a strong advocate of not missing the "On Ramp" to an Analytics-Driven culture---which is applying real science and discipline to a business domain. That will lead to much more interesting Discovery for the business in my experience. Does take a strong advoate and resistance to being measured initatially.
@Lyndon_Henry Nice catch. I have noticed these shows as well - refuse to watch them as I know they are painting how "data scientist" are to be precieved. Hollywood has no place in real world applications for the most part.
@Jim, how often really depends on changes in business drivers, number of business lines taking on data science, maturity and size of the organization. No one answer but considerations on how to make the selected model successful
In all seriousness though, yes we have very tailored ways of crafting our data science positions. Usually, we'll hire someone based on thier potentialto be a data scientist, but that is now how they enter our business. They enter as Computer Scientists, Mathemtician, Consultant, etc, and begin the journey toward data science team.
Notice that a lot of the TV crime shows with lots of high-tech forensic science (i.e., analytics) have kind of weird, unconventional, even zany (i.e. creative) characters. Maybe a snapshot of a "data-driven culture"? But is that realistic in the real business world?
@Robert Arvanitis I see. I think we speak of think of it (creativity) differently. What you speak of in my mind is innovation which of course it just a larger aspect of creativity. I like the thought of a systematic approach to introducing changes and comparisons, but in my mind as soon as you apply a "system" the trueness of creativity is lost.
I finally figured out that is why I don't write my jokes down either. : )
@David Wagner - yes, matching the right Questions, Data, and People at the same time is so hard! We never count on having the right data. First step in Data Science is ALWAYS Data Janitor, basically trying to fill gaps, reduce question scope to match available data, interpolate, etc. Questions always evolve, so never count on having the right question to start with.... but to get the Culture right, always insist on having the right people.
The right model for a data science team depends on the size and complexity of the organization as well as your current business drivers. You may start with one model (i.e., centralized) and evolve to another over time
So I think we've talked a lot about that data science culture. And we've talked a fair amount about spreading evangelism. But it seems to me the hardest part of data science is always that conenction between the decision and the data needed for the decision. How do you prepare your culture to making sure you are putting the right data in the hands of the right people and the people making the decisions want the data to begin with?
To help develop our own data science teams and those of our clients we developed a data science competency model. Basedlined against our data scientists and found that innovation, creativity, curiousity, and perserverance were integral to their skills
Would an organization's structure play a part on the effectiveness of a data science team? Also, which would work better - a centralized team that's deployed based on projects? Or a decentralized model, as @EKhalil said?
The conversation reminded me of a favorite quote of mine:
Everyone you'll ever meet knows something you don't. (Often attributed to Bill Nye) Makes me think there is a reason to have a "big tent" data science skills / background approach to finding staff, provided they have the curiosity and relentlessness needed.
@louiswatson. Well, at least a framework for creativity. Like "Analogies, Metaphors, Perspective." A systematic way to introduce changes, and comparisons. Idea is to not overlook something because the current way is too embedded in our thinking.
For companies just starting with data science, considering a centralized model can work best to start building a cadre of data science capabilities. Then organizations can think about diffused (matrixed) or deployed (decentralized) models over time
There is a lifecycle for data models/analytics, can you tell us from your experience of companies doing this right what skill sets they have for the early discover data scientist, vs the maintain models, and best practices for operationalizing analytic models.
I see a number of thoughts on how to learn data science in both training and university settings. Columbia Univ in NY, Stanford, Coursera, and ExploreDataScience.com are all great resources that I've seen used. Some are classroom based and others are "learn by doing" as others have pointed out.
Looking forward to the text chat. I feel like we've opened a lot of interesting lines to talk about. I, for one, would love to know how you interview and recruit for the kind of relentless curiosity we talked about at the beginning of the show.
@lyndon- I'd imagine they would. But some companies seem to rotate people around the business quite a lot. I've never been at a company like that, and I've often wondered how it works. Seems like you run the risk of moving people from places they are succeeding to places that they might fail.
Unfortunately, I have to jump off for a conference call. I would be interested in a follow up conversation if anyone has any ideas on how a NGO can pay for data scientists. Even the curious people in NGO's have other jobs and can't dedicate the time to data mining. Great conversation though! Wish I could stay on!
@katfree- I get that. What I've discovered is that there are a lot of people with no experience in that but with good intellectual skills who can be drafted to do it. i've met data science folks that started with all kinds of backgrounds who just found their way into it through their own curiosity.
Interpetation is really a big problem regardless of the organization doing the analysis. I think a big problem is that most organizations don't understand the connection between the data and strategic direction. It's even worse in social services.
I saw a listing for a Data Scientist role recently and they wanted you to have had work in one or more of the following: NOSQL, Hadoop, Hive, Pig, MapReduce, or other similar tools, Databases and/or data models (Graph databases, semantic frameworks etc), Statistics (Regression, Clustering, Decision trees, Hypergraphs etc), Simulation, scenario analysis, modeling
@lyndon_henry- With all due respect to Watson, i tis great at finding the patterns it is trained to find. It can't yet just look at something it has never been trained to look at and "get it." People sometimes can.