The End of Analytics?

by   |   October 24, 2016 8:30 am   |   6 Comments

Thomas Davenport

Thomas Davenport

Earlier this month I hung out at Dreamforce, the Salesforce.com extravaganza that swamps San Francisco each fall with people (170,000 or so) and hoopla (U2, Tony Robbins, etc.).  At first I was impressed that I was asked to give a keynote—I began to think of myself as the “Tony Robbins of analytics and AI,” perhaps a dubious distinction—but then I realized that at any given time there are about ten keynotes being delivered. It was great to be onstage, but mostly I was just glad to find my session among the more than 3000 others. What I really enjoyed, however, was talking with other attendees of Dreamforce who are interested in analytics and AI. I learned much more from them than I ever could from hearing myself talk.

Salesforce has been moving in a more analytical direction for a while now, with its Wave Analytics product that offers do-it-yourself descriptive analytics. But this year, the company took a decided turn in the cognitive direction, with its Einstein suite of cognitive and analytical capabilities that are being baked into the different cloud-based systems for sales, service, and marketing. As Marc Benioff, the company’s CEO, put it in his keynote (which was attended by a few thousand more folks than mine), “We’re on a march to artificial intelligence,” and, “We want to become everybody’s data scientist.” The company has made ten or so acquisitions of AI and analytics startups over the past couple of years, and in the process has “acqui-hired” more than a hundred very smart data scientists.

I thought that the translation of acquired companies into actual Salesforce products was very fast, but John Ball, the company’s General Manager of its Einstein offerings, told me that the intelligent capabilities were planned for several years, and that each acquisition added a piece that had been identified as necessary.

However long it took, I think it’s a great idea to make AI and predictive analytics easier by embedding them into existing transactional systems like CRM. If you offer a salesperson a set of predictive scores on how good their customer leads are (one of the Einstein capabilities), who wouldn’t want to use them to help allocate time and attention? If your analysis of sales data can identify hidden patterns that you didn’t anticipate (which the company’s BeyondCore acquisition can do), who wouldn’t be interested in learning about them? These types of embedded capabilities can, I believe, rapidly accelerate the adoption of AI.

But I was most interested in how customers would respond to these new capabilities, and in the conversations I had with them, I got some insights into that issue. There seemed to be a lot of interest. In fact, several companies (I have promised not to name them) said they were hoping to “leapfrog” over traditional business intelligence and analytics capabilities and go directly into AI-related environments. As one technology executive from a software company put it, “We’ve had enough bar charts. Nobody has time to digest them anyway. We want our analytics to tell people what to do.”

Another manager from a non-profit organization admitted that her organization had used virtually no analytics in the past. But she also wanted to leapfrog to AI. She noted, “Why walk when you can run—or even drive?” She liked both the BeyondCore “smart data discovery” and the Einstein-equipped CRM ideas.

The general feeling of many of these executives seemed to be that the traditional notion of “analytics” can recede into the background, and that what workers really need is to be told at any point the smartest thing to do next. The managers I spoke with at Dreamforce don’t want complete automation—the humans still need to be able to override the recommendations—but at least some don’t seem to feel a need for visible analytics.

This development, were it to actually happen, would have implications for far more than Salesforce.com (though it might imply diminishing demand for the company’s traditional descriptive analytics offerings, which create lots of bar charts). It would change the entire analytics and BI industry. We’ve traditionally felt that companies should walk with descriptive analytics, run with predictive and prescriptive analytics, and drive with autonomous analytics and cognitive technologies. But what if companies want to go straight to driving? What if the traditional distinction between transaction-oriented applications and analytics applications collapses? Is this the end of bar charts and pie charts?

I am not entirely sure that this would be a good thing for the world. Descriptive analytics help people get familiar with their data and understand how far it can be taken into decisions and actions. A bar chart for a single variable may not be the most sophisticated form of analytics, and it does require some human attention. But it can tell you, for example, that you have some outlier values that you didn’t anticipate, or that a different categorization structure for your data might be in order. More advanced analytics and AI based on them are great, but you don’t want to automate a decision before knowing that your data are of sufficient quality.

We’ll see if this new view of analytics catches on. It’s both an exciting and scary prospect for those of us who have spent much of our work lives discussing the traditional categories of analytical activity. Let’s hope that the new tools are used in addition to the old ones, and that we can still find ways to occasionally immerse ourselves in data.

 

Tom Davenport, the author of several best-selling management books on analytics and big data, is the President’s Distinguished Professor of Information Technology and Management at Babson College, a Fellow of the MIT Initiative on the Digital Economy, co-founder of the International Institute for Analytics, and an independent senior adviser to Deloitte Analytics. He also is a member of the Data Informed Board of Advisers.

 

Subscribe to Data Informed for the latest information and news on big data and analytics for the enterprise.

Tags: , , , , , , ,

6 Comments

  1. scott radcliffe
    Posted October 26, 2016 at 9:06 am | Permalink

    Thanks for this Tom. As always, useful insight about emerging trends in analytics. The deployment of predictive/AI into transnational processes is way to expensive, and takes too long. In this area, decreased reliance on data scientists seems inevitable.

    However, this article overlooks the the fact that most organizations aren’t very good at understanding what bit of insight to insert in which process, and the expected ROI. This is especially true of service businesses (or those with large human and physical service components).
    This requires explanatory/inferential analytics, which typically requires broad analytical competence to obtain actionable insights for management level decision making. For example, a service business may seek to reduce OPEX by reducing upstream defects defects that require otherwise unnecessary service interventions. There may be a dozen operational dimensions that are the source of these defects, and a few to many indicators of the occurrence and performance of those operations. Sorting which operational features in which to invest, and which enhancements to test requires keen interaction with the business decision makers, and bring that into the model specification.

    I know Tom knows this already, just wanted to point this out to those new to the application of analytics to business.

  2. Tom Nogles
    Posted October 26, 2016 at 11:12 am | Permalink

    Tom

    How do you differentiate between prescriptive analytics and AI? When I hear about salesforce’s AI initiatives it strikes me as prescriptive rather than traditional AI.

  3. Art Fennelly
    Posted November 2, 2016 at 2:08 pm | Permalink

    I do not see the end of analytics. What I do see is an expansion of analytics, especially in the areas of prescriptive and discovery. There will always be a need for descriptive analytics, businesses need to know the current state and past. Although I suspect that analytics will become more commoditized as companies progress in the analytics maturity model and capabilities are embedded into more and more product offerings.

  4. Sudipta Sarkar
    Posted November 3, 2016 at 6:19 am | Permalink

    Analytical recommendations embedded in the transactional applications at the point of decision would be needed without making it a difficult activity for business to get the analysis. Whether we need AI to always fulfill that is a different discussion.

  5. Nilesh Saraf
    Posted January 12, 2017 at 10:21 pm | Permalink

    I agree with all the above comments….there are many reasons why business decisions can never be on autopilot on some strategic domains even with the kind of analytics products Tom is describing. First, data needs to be integrated from diverse sources (Internal and External), and with changing organizational environment you can never rely on the seamlessness of the plumbing. Two, the metadata are always going to be in flux depending on what managers want to measure and track. The more entrepreneurial the managers, the less stability in necessary data (definitions) and how it is re-combined for insights. Third, if my competitors are using the same prediction models that come pre-packaged, they can easily guess what I am likely to do (depending on the domain of application). Maybe this could drive up user expectations. Fourth, not every decision maker is as comfortable with interpreting and applying those interpretation to their current business decision (a point already made above).

  6. Kamalakant Sharma
    Posted January 22, 2017 at 7:36 am | Permalink

    Innovation, modification, revision, and sensitivity are key to correctness, fitness, and sustainability of any model, including different facets of analytics.

Post a Comment

Your email is never published nor shared. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>