Three perspectives on Watson and business analytics
Delaney Turner 270003RQ8K Delaney.Turner@ca.ibm.com | | Tags:  ibmwatson business_analytics
0 Comments | 2,768 Visits
If Watson shows us where analytics is going tomorrow, what does it mean for the solutions you use today? I sat down with three IBM Cognos product experts to get their take on the opportunities Watson presents to business analytics software. With me were:
Don Campbell: Increased confidence for better decisions
There are two aspects that are commonly thought of in large-scale analytics: size and speed. The first one says “I have a whole bunch of data that I can analyze” and the second one says “I can do this quickly.” I can do both of these because I can massively scale.
Watson has a few elements that make it different and applicable to business analytics. The first is understanding different kinds of questions, because when we’re in our business we need to understand the questions in front of us as well. Another is that it breaks down the question and attacks it in parallel and in different ways. It looks for the answer from a variety of sources – celebrity factoids, puns and language tricks, structured and statistical information - and then pulls all that information together for a single answer. When we think about business analytics, that’s the value that we bring to customers today by pulling together what Sales says about a customer, what Marketing says about a customer and what Competitive Intelligence says about a customer. We pull all those things together into one answer for a decision.
Watson also introduces the confidence element, which I think business analytics will need to deal with more and more. Yes, there’s lots of information out there to draw from to find an answer - especially once you go out into the social realm and start scouring the blogs and Twitter feeds. But how confident are you that you'll get the right answer? Watson can find possible answers to a question but it also knows how confident it is in the different domains it’s searching. Then it rolls these confidence levels together and looks for corroboration between what it saw when it looked for something similar in previous questions. From there, it can determine whether it has enough overall confidence to buzz in.
The confidence element is really important because in Jeopardy!, you actually lose money for giving the wrong answer. In business, you lose money as well. Confidence is a really big part of the problem that we as humans in management teams frequently overlook, and that’s what I think makes Watson more relevant to a typical business problem than number crunching on its own.
Watson’s Deep QA technology is also a very interesting place for us, because I can imagine all kinds of assisted decision-making capabilities that it could bring, especially at those decision points that are very real time – when people can’t sleep on it, or consult with 10 people before deciding what do to. This is what we enable today through various technology capabilities and human intervention. But with Watson as a growing part of what we have in our back pocket, we can streamline a lot of the things we do today and do it in a more facilitated way. It’s all in line with making informed decisions from disparate content.
Craig Statchuk: Better semantics for more effective query and search
Watson highlights 3 research concepts I am very familiar with.
The first is UIMA (Unstructured Information Management Architecture), which is a way to categorize and add expertise in vertical subject areas. You may buy a generic decision-making system that knows how to deal with HR, but doesn’t know manufacturing, logistics or sales. UIMA is used to train these systems to more fully understand source context so they ultimately do a better job digging for answers.
The second concept deals directly with business intelligence (BI) where we have consistently strived to find “the single version of the truth.” Now we’re starting to better understand a concept that I describe as “the best supported version of the truth.” With deep analysis of Big Data sources, we’re finding the multiple data lineage paths that improve search recall and disambiguation. This will advance our governance and decision-making capabilities because we provide answers that are backed with the greatest amount of primary data while highlighting inconsistent and incorrect data. This technology is still in its infancy but I think Watson is showing us the way.
I came from the search side of the business. We have always done statistical indexing of terms and words. In recent years, we’ve added UIMA-style topic categorization. The result: we find the best match for a handful of simple keyword terms. But answering questions – and actually understanding the implied semantics – is at the other end of the spectrum. Watson teaches us the value of Natural Language Processing or NLP. This helps us understand the context and semantics of your search question – including the subtle nuances of what you meant. When we apply more of this technology to search, we’ll have systems that can disambiguate your questions and ultimately answer queries like “Which manufacturing components had the greatest supplier cost increases last quarter?” or “What are that factors that will negatively affect sales next month?”
The last thing to keep in mind is that Watson is getting smarter the more it “plays” because it has an algorithm that rates its own success. In other words, Watson it is calculating its own “return on investment” with every round it plays. We are doing the same with business analytics software by adding algorithms that manage and improve themselves.
Colin Moden: Thinking beyond the data warehouse
We see Watson as leading-edge research in the area of analytics. Like the Apollo program, we expect to see spin-offs in technology that's developed from projects like this into commercial solutions.
The history of analytics and data warehousing has focused very much on the numeric domain to help people understand of what they did as an organization in the past. But understanding where you've been won't work if you want to make the forward-looking decisions that optimize business outcomes.
Our world is becoming increasingly instrumented, intelligent and interconnected - it's getting smarter. And to optimize outcomes on a smarter planet you need to take into account of a lot of other types of information, We need to answer questions like What do our customers think of our products? What are competitors up to? How many customers enter our store and leave without buying anything? What route did they take? Online retailers have been doing this kind of analysis for a while; now it's possible to do - and even automate - in a physical store as well.
The information sources you need to analyze, though, are much more diffuse, much more unstructured and require a "mining type" mentality. Think about mining the Web, in particular the blogosphere. Think about processing RFID information, web log information, satellite pictures, even weather predictions. Making good decisions means analyzing data from an increasingly wider perspective.
The technology being pioneered in Watson provides an opportunity to continue that journey by further exploring future strategic business analytics capabilities. I'm thinking of searching large data sets, processing text with semantic and context understanding, responding to questions that require statements rather than numbers, evaluating possible responses with a degree of confidence.
That last area is very interesting to me, because not everything is certain or totally unknown. There are many shades of grey in between and it's quite different from the traditional data warehousing mentality. For example: suppose Watson told you that 10 percent of your customers can't figure out the camera on your smartphone and give up on it, and that figure is accurate within five percent 19 times out of 20. Is it definitive? No, but is it good enough to do something about to keep them happy and using your product? Most likely.
Traditional business intelligence - the kind that helps you understand the numbers - isn't going away. But it's going to be part of a broader solution. For example, IBM Cognos Consumer Insights is based on Hadoop.
Watson is ample evidence of the IBM ability to capture the market with an ambitious vision and deliver leading-edge innovation that delivers real value to our clients and the world. As we and our customers move forward on a journey toward better decisions and better business outcomes we'll continue to leverage the innovative technology developed through IBM Research.
What's your take on Watson?
Create an account on the IBMWatson Social Media Aggregator to share your views and see what others are saying in one convenient spot! (Be sure to tag your blog posts and tweets #IBMWatson)
Begin your own journey to better business outcomes