Teradata Big Analytics Roadshow in Atlanta

[BrianD]

Friday I was at the Teradata Big Analytics Roadshow in Atlanta. It was a great event – and I met many new folks from Coca Cola, Georgia Pacific, Lowe’s and other companies. I also got a chance to hear from Mike Gualtieri of Forrester Research, and Bill Franks and Randy Lea of Teradata.  Mike noted in his keynote: “The data is valuable, the analysis of the data is more valuable,” and I couldn’t agree more.

I presented the Alteryx session “Hear Customers Speak Through Their Data.” You can see the deck we used for this event on the Alteryx.com/Teradata page. kudos to Jimmy Garrett of Alteryx for a great overview demo, as well as a deep dive on our Aster integration. The demo showed how a business user can easily drag the Aster macro onto the canvas and input a couple of fields, and then run a full n-Path analysis, without needing to understand the underlying code. This integration really extends the power of Aster analytics to business users.

Social Media Panel

I was also on a panel chaired by David Schweer of Teradata, with:

• Atique Shah, VP of Global Guestand Campaign Marketing, Intercontinental Hotels Group
• Bill Rand, Director of the Center for Complexity in Business, University of Maryland
• Michael O'Connell, Chief Data Scientist at TIBCO

This session followed an excellent session by Bill Rand entitled “Analytics for Social Media: Structure and Content,” thus there was a great focus on Social Media on the panel. There were two key questions from the audience. Here is the first question (and the panelist responses):

“What if the social media results are not indicative? We reviewed storm info and it seemed that Manhattan was the worst hit based on the volume of the tweets.”

Bill – One aspect of bias is the sampling / selection bias.  To that extent, this is like the old phone survey bias (Landon vs. Roosevelt in 1936) – only people wealthy enough to have a phone would be surveyed. There is the same bias on social media. In our research, we have ways of normalizing the data to remove this bias.  But the point is that it is still more pervasive and cheaper to collect information than most traditional survey approaches.

Michael – social data are spontaneously observed and suffer from selection bias, in addition to measurement error. We can normalize to account for selection bias (e.g. divide by regional population size in this example), but it is difficult to address measurement error and accurately represent the sentiment in a text snippet. As an extreme example,  how can we address sarcasm?

Brian – you must have context and you can’t rely on social media as the sole source. In the case of Optimum Lightpath, a complete picture of storm outages involved combining CRM data, network system data, and to understand outage causes, a map of power outages. Social media is more useful for sentiment analysis. And that sentiment analysis might be used as a predictor variable. We might look at our customer history, and identify, prior to leaving a service, how many calls did they log in our CRM system? Using network data, how many dropped calls did they have. Were there any negative social media posts? By combining these sources, it gives us a more complete picture of our customer, but social media is not the sole source. (You can hear the presentation by Optimum Lightpath in this slide deck on SlideShare. Download the deck and listen to the recording.

The second question came from another attendee at Cox Communications, who pointed out that Facebook is taking away the “people are talking about this” feature. How will we know if they are engaged?

Bill – Rather than relying on Facebook analytics, which are suspect, it would be better to write an app.  All social media services change over time, not just Facebook but Twitter has changed their API a few times. Using an app provides you with your own (somewhat) controlled way of collecting data about your consumers and customers, and can provide you with much more detailed information than available through traditional APIs

Brian – If you have an app, make sure you give some added value with the app. If you offer TV listings on your website, offer them as a Facebook app. Link them to Facebook pages on each of the shows, so people can look at the schedule and click to see what people are saying about each show.

Attique – Definitely giving away something of value drives engagement. We had over 70 million loyalty program participants, but only 6,000 likes on Facebook. I arranged an offer: Next time you stay with us, like us on Facebook and we will double your points for that stay. We now have hundreds of thousands of likes.

Michael – Loyalty programs provide information-rich data including demographic profiles and behavior e.g. spend, online and in-store customer actions. We acquired a company, Loyalty Lab, to focus on this area; and have integrated this with our analytics and event-enabled products to provide an informative view of customers for marketing programs.

Sustainability

One other theme of the panel was sustainability. We didn’t get to this topic live, but we had a great conversation on sustainability while preparing for the panel. David’s question was, for many companies, they make one heroic effort to have an analytics driven campaign, but then they are never able to recreate it. I think the main reason for this is that the data gathering is such a manual process for these organizations, verses creating a repeatable process that can be iterated.

Alteryx automates these processes, and our customers reap the results – ongoing marketing analytics that drives all of their campaigns. In our session, we profiled Southern State Cooperative, and how their marketing programs are optimized using predictive analytics to choose which offers to match to which customers and prospects. They have cut their mailings by 63% and raised their response rate by 34%.

You can read their story here.

The next Big Analytics Roadshow is in Dallas on July 25th. We will see you there.

Brian Dirking

Director of Product Marketing