What Data Science Can Do for the Enterprise (Extra Compact Edition)

- The following post was authored by experience&co's Justin Williams and originally appeared on Think Experience. -

Now that we’ve set up some of the terminology from Part 1, how do we apply these concepts to real business problems?

What Data Science Can Do for Business and How It Does It – Extra Compact Edition

There’s a lot of things you never see, and you don’t know you don’t see ‘em cuz you don’t see ‘em. You gotta see something first to know you never saw it, then you see it and say, “Hey, I never saw that.” Too late, you just saw it! – George Carlin

The next thing I like to do in a workshop, once we’re all talking the same lingo, is to talk about measurable (qualitative and quantitative) outcomes. The temptation will always be to start with the data, but you must hold onto one without letting go of the other throughout the entire process or you’ll end up with a really expensive spreadsheet that doesn’t have the impact to your organization that you had hoped. This is also a good time to talk, briefly, about the kinds of outcomes that data science is good for, which is a surprisingly wide range. Here I’ll present you a two dimensional representation on both the ‘what’ and ‘how’.

First, let’s talk about what (and why).

• Cost – You need to reduce how much money something costs to maintain.
• Growth – You need to grow or leapfrog your competition. Innovation is in your blood.
• Time (which isn’t money, it is a lot more valuable) – time to market is everything, or you need efficiency. You need to shift where time is spent from less valuable things to more valuable things.
• Brand – How do your customers think about you and your products? What about your employees? Are either loyal?
• Risk – “An ounce of prevention is worth a pound of cure” – Ben Franklin
  
  

Wow, that’s a lot! What are the tools that achieve these broad but important outcomes?

• Customer Experience
  • Taking sentiment analysis to the next level with emotional analysis.
  • Imagine what useful automation and customer routing could do in not just telecommunications but all touch points, web, mobile, even brick and mortar.
  • Not search. At least, not “fuzzy” search as we’ve known it. Sadly, this is still one of the first things people think of though.
• Employee Enablement and Operational Assistance (the vastly underutilized super weapon of the cognitive computing age)
  • So this is about downsizing, isn’t it?! No, boring, routine RPA is close to saturation levels already. The power comes when you combine the low cost, incredible cognitive computing of humans with a new generation of automation tools.
  • Assistive technologies in industries like medicine where a personal assistant who’s up on this morning’s research results could help a physician reduce clinical costs and save lives while spending more time with a patient and less time in front of a screen trying to remember critically important details they encountered hours ago.
  • Assistive technologies that meet behavioral science and react to the environment. If you’re selling tires and want to upsell in the winter, don’t tell your sales team to hit quota, or they’ll ask, “do you want fries snow tires with that?” and lose the sale by customer instinct. Instead, when appropriate, the system could show a little weather dialog to remind the sales associate that there’s a winter storm advisory for the weekend and maybe they could help their customer by asking if they need snow tires too.
  • A culture of risk aversion doesn’t necessitate a culture of beat down, afraid employees who don’t take the good kinds of risks and avoid innovation. Automated policy governance with a positive employee experience can help.
• Predictive Insights (the go-to capability)
  • Deep learning was able to predict seizures long in advance of them happening back in 2014. What do you need to know about in your company or industry before it happens to change the outcome?
  • This is opposite of dinging bells and flashing alerts. If you just need to be notified that something already happened and don’t need help understanding why it did, then this is not a problem for data science.

And because I can’t resist a good model that visualizes two dimensions (one with a drill down!):

I realize this was an extremely brief overview of what and how data science can be applied to business. But it’s such a big topic, we’re dedicating an entire series to these. We’ll get them out there as quickly as we can but in the meantime please consider this setting the stage – and feel free to reach out to discuss directly if you prefer not to wait.

What About the Data?

“Data! Data! Data!” he cried impatiently. “I can’t make bricks without clay.” – Arthur Conan Doyle

“Data is the currency of the modern enterprise”. – Karl Johnson, CEO of experience&co

The first question most people have about big data is how little can the big be. The answer depends upon what you want to do with it. Cognitive computing, in general, is able to act on very little data, actually, but the less data the more hands on people will need to be to make sure that it gets it right. A lot of RPA, for instance, can be very successful with virtually minuscule data sets. You won’t be able to predict the maintenance schedule of expensive equipment in a variety of environments without a lot of data however. Still, there are some ways to get data if you don’t have it. You may be able to buy or get public data that helps create the right models and outcomes. You may even be able to simulate the data you need, with enough sample testing for validation.

What Are the Pitfalls?

“Experts often possess more data than judgment.” – Colin Powell

“Success is this: the right solution and to do the solution right.” – Anton Abramov, Experience and Strategy Director of experience&co

To really demystify data science for business we must rid ourselves of the illusion that it can do anything. It can do a lot, certainly, but to know a thing we must outline all of its boundaries. This is by no means a comprehensive list of all possible problems, and some may become less common in the future, but they do represent today’s usual suspects. Again, we’ll continue to describe these in more detail in future posts.

Pitfall #1: Data Strategy Lacking Engagement… or Strategy

Once you feel comfortable with data science as a tool, even if it isn’t a tool you create yourself, you can stay engaged in a data strategy and its execution. Obviously it is imperative that you have the right talent to work with but it’s equally important to partner with them because everything you hope to achieve is a business solution, not a math problem, and the talent you choose should be specialized in building and tailoring the tools, not running your business. Nothing ever good came from throwing a problem over a wall and hoping for the best – beware of those who advertise that’s what they do. Critically thinking about your evolving business along the way is the real magic. Speaking of evolving, if you come up with a perfect plan, build a 2-year overarching strategic road map then walk away to come back 2 years later to exactly what was specified, what are the chances it’s still the right thing? First, a lean, engaged, focused team shouldn’t be handing anything over after a long, multi-year wait anymore. But more importantly the best value is one that allows the flexibility and agility to provide at least minor change along the way to react to external factors that don’t exist today.

Pitfall #2: P-Hacking

“Facts are stubborn, but statistics are more pliable.” – Mark Twain

One of my favorite sites to prove bogus opinions I might have is the automated spurious correlations engine at:http://tylervigen.com/discover by Tyler Vigen. We all know that correlation is not causation but because people do have biases and motivations, sometimes we can will the data to mean what we want, not what is real. Believe it or not, this is far more often accidental than malicious.

Data science “smells bad” like P-hacking when you’re using deep learning or unsupervised learning techniques and there’s either not much data, too many features, or both. Statistical anomalies just happen and you need a data team that can critically think about the semantics – the meaning again – that the machine simply can’t understand yet. And this is another reason the business needs to stay engaged and subject matter experts on what the meaning really is.

But this brings us to:

Pitfall #3: The Unbiased Machine Fallacy

“There have always been ghosts in the machine. Random segments of code, that have grouped together to form unexpected protocols. … When does a difference engine become the search for truth?” – Isaac Asimov

At a recent conference on AI I attended there was a great panel on the problem of universal morality when it came to self-driving cars. If external factors create a situation in which an automated car must risk killing one person or another, how will it know who to choose to put into harm’s way? Does it have enough data that it would make the same choice as the driver? Does it evaluate some form of “value” of who it knows to be driving and weighs that against the cyclist it can make some assumptions about, such as cardiovascular health and approximate age and potentially decide to directly challenge the human’s instinct of self-preservation?

The panel’s answer was very data sciencey. It went along the lines of, “we’ll just treat that as another big data problem, see how most people would react to various moral situations, and feed that in as inputs”. The disturbing questions are, are the majority right in how they would behave? Will the sample set used truly be universal enough to represent the true majority even if it is? These are difficult problems and the biases and persuasions of both the data team, the business stakeholders, and even the people who have helped to generate a particular data set all can have influence over the eventual algorithms. The point is redundant, but critical thinking and purposeful approaches are the only way to ensure the kind of insights that produce reliable outcomes and don’t just tell you what you want to hear.

Pitfall #4: The Uncanny Valley

The “uncanny valley” is when something artificial is not quite real enough nor fake enough and ends up freaking people out. We like to name our cognitive computing solutions with names like “Einstein 2.0” but this sets very unrealistic expectations and ensures that the people who end up working with such a solution find it disturbing, or at the very least, extremely disappointing. Discount even the small details of the human experiences to a data science solution at your own peril.

Pitfall #5: Garbage In, Gold Out Fallacy

Finally, the push for unstructured data technology has often been a quest in search of taking data that has failed to be cleansed in the past repeatedly, but somehow still gain value from it. That’s not what unstructured NoSQL is for, unfortunately. There is a place for those systems, and they can vastly improve or speed up efforts to cleanse data but data that is outright wrong is more common that we would all like to hope and the cleaning, or finding the right ingestion methods that can compensate for error, is still by far the number one time sink for data science projects today and for the foreseeable future. However, our strategies for handling this continue to steadily improve. For now, if it’s especially messy, avoid unsupervised learning, it probably isn’t the panacea to fix errors unless you have a substantial clean set it can learn from first.

Wrapping Up for Now

“There will come a time when you believe everything is finished. That will be the beginning.” – Louis L’Amour

OK, sure, there are some important pitfalls to avoid, but knowing what to look out for helps us have really successful projects. Hopefully you come away from this post with a better understanding of what data science and the latest technology can do for you and your company, and some of the murkier areas are a little clearer. This was just the first steps, however, we’re going to focus on concrete examples and clever tricks to getting the most out of these types of projects in each of the upcoming posts. This was, as mentioned, the extra compact edition so don’t wait for future posts, if you have questions, want to argue, or just discuss how data science projects could make your company’s strategies come to life, contact us!

- The above post was authored by experience&co's Justin Williams and originally appeared on Think Experience. -