1. Official
  2. Clear Seas Research
  3. eu2015_480x60
  4. RN_ESOMAR_WP_GBook_480_60_11_14

Framing the World of Predictive Analytics

During the past month, I was able to attend two premier conferences dedicated to predictive analytics; these were the The Predictive Analytics Summit in San Diego, and Predictive Analytics World in San Francisco. In categories like this, it’s next to impossible to follow the notion of a mutually exclusive and exhaustive set. In spite of this classification issue, I will do my best to provide a bird’s eye view of the sector. I’ve broken it down into three types of predictive analytics and the three types of companies pursuing the predictive analytics market.

 

By Tony Cosentino

During the past month, I was able to attend two premier conferences dedicated to predictive analytics; these were the The Predictive Analytics Summit in San Diego, and Predictive Analytics World in San Francisco. (The conference in San Francisco was back-ended by Text-Analytics World and the Google Analytics conference.) In addition to attending the conferences, I had the privilege of being on the panel, “Winning with Data Science: Transforming Complexity into simplicity” which served as a cap-stone discussion and visioning session for the Predictive Analytics Summit. After listening to my fellow panelists, as well as following the disparate presentations and engaging vendors at their booths, I came away with a lot more knowledge, but also a fair amount of confusion of this emerging field. It seems the category is still defining itself and that many of the business models are overlapping. To be honest, the environment reminded me a bit of the dotcom era in its energy as well as its ambiguity. In order to ease my confusion (and hopefully provide some clarity for the Greenbook.org community as well), I’ve tried to frame my thinking. Note that in categories like this, it’s next to impossible to follow the notion of a mutually exclusive and exhaustive set. In spite of this classification issue, I will do my best to provide a bird’s eye view of the sector. I’ve broken it down into three types of predictive analytics and the three types of companies pursuing the predictive analytics market.

Three Types of Predictive Analytics

  1. 1.       Business Operations and Financial Analytics

This is one of the premier pay-offs currently for predictive analytics. Areas such as supply chain management allow companies to match their stock very closely with customer demand, thereby reducing carrying costs for the channel. Coca-Cola corporation presented on the reduction in inventory levels needed by their partners and the ability to segment the market in an increasingly niche retail environment.

Financial modeling has of course always been a part of the predictive analytics world. The IPR group gave a presentation on predicting computer markets and their Early Warning System (EWS) offer. (The offer appears to look primarily at different leading indicators produced by the government to predict trends in the technology industry). At the end of the presentation, I asked about Black Swan events, and what role disruptive forces such as the exploding tablet market might play in their predictions. The presenter suggested that these are still challenges especially in the technology industry as it blurs the categories and classification becomes more difficult. As one might expect, the way they address the situation is essentially by giving multiple simulated environments based on different technology adoption scenarios.

  1. 2.       Industry Specific Analytics

Things like fraud prevention, risk analysis, and disease prediction are niche areas for predictive analytics. The Bureau of Alcohol, Tobacco, and Firearms gave a presentation regarding how they are better matching resources to threats based on predictive models. Other presentations revolved around sports analytics (think “Moneyball”), fraud prevention, and portfolio valuation. Niche applications in healthcare are predicting disease and using this to influence things like clinician behavior.

  1. 3.       Customer Behavior and Marketing Analytics

This is the area that likely hits closest to home for the market research and insights industry. Loyalty and customer experience seem to be the hottest topics right now and analytical CRM frameworks are providing very sophisticated ways of predicting attrition. Companies like Acxiom as a data provider, and Nuevora and Opera as analytics firms are looking at individual level behavior and wallet share in a holistic manner (encompassing both on-line and off-line environments). This individual level view seems to be the direction of the industry and it is proving to be a very powerful formula when mixed with the right data and approaches.

The other area that jumped out to me was predictive analytics in cross-selling and product positioning. One large technology company that I spoke with off the record told me that they were able to use their install base information, buying trend information, and refresh cycle information to do planning at the account level (the type of planning previously reserved for “named accounts”). This allowed the rep to determine when to sell, and what offers should lead. There are actually a few companies starting to market this type of database information; more on that below.

One thing that kept coming up in my conversations was the definition of predictive analytics in marketing. When  I think of predictive modeling, I think about ‘scoring’ a model and getting to a rank order outcome (i.e. probability scenarios for particular outcomes), but many (especially those coming from a technology perspective) seem to use descriptive analytics (such as the act of segmenting a market), interchangeably with prediction. Business Intelligence tools like Cognos, or Business Objects are primarily descriptive in nature, yet these tools are being called by some, predictive analytics tools. While it is true that certain modules may provide the predictive analytics capabilities, the legacy nature of these software are primarily descriptive in nature. In a broader sense, the idea of predictive analytics (at least in the area of marketing) seems to be converging with decision analysis and therefore, brings in certain theoretical assumptions (along with an ounce of confusion) about choice.

Three Types of Companies

  1. 1.       Plumbing/Web-Analytics

This is a huge area and it includes all the web analytics firms (Omniture, Webtrends, etc.), sophisticated log-data analytics (Splunk, etc.), and areas such as Tag Management (Tealium,etc.). In addition, a lot of attention is being paid (rightfully so) to the idea of “in-memory” analytics. The idea here is that data mining can be done separately from the data warehouse itself, thereby not needing to access the hard-drive of the system. The result is not only faster query time, but the entire need to store data in the traditional manner goes away. (Many predict that in the future, data warehousing, data cubes, and even SQL will not be needed unless we need to go to the “old” storage facility to dust off some long out of use data.)

  1. 2.       Marketing/Sales Analytics Software

Many companies grouped around using prior knowledge to predict buying cycles and cross sell opportunities. The example I mentioned above where the firm was able to analyze install base, buying cycle, and past buying behavior to predict product leads, cross-sell/up-sell, and timely sales calls, is an important one, but on a broader basis, companies like Lattice Engines (backed by Sequoia)  are providing an industry-wide platform for this type of information. What’s perhaps even more interesting for the market research industry is that previous approaches in this category by firms such as RainKing and DiscoverOrg used survey based interviewing approaches; Lattice-Engines uses only firmographic and behavioral data collected in a crowd-sourced fashion.

I’d be remiss if I didn’t mention the 800 pound gorilla in the space. IBM was present at both shows (and really competes in all of the areas of the market). As most of us know, IBM has a lot of momentum in this space, not to mention $14 Billion dollars put aside for predictive analytics initiatives and acquisitions. So far, most of their activity has been comparatively smaller players and they have only used up a fraction of their war chest. (It makes you wonder about a potentially larger acquisition at some point.) In the not too distant future, they will be linking everything together to provide a broad ROI dashboard for marketers and strategists, but at this point, they are still separate. CoreMetrics is primarily a web analytics platform (the demo showed very impressive benchmarking data for retail); Unica provides much off-line retail data; SPSS provides the analytics/modeling engine; and Cognos the traditional BI. (I’m sure I’m missing others in their acquisition spree, but those are the main ones that came up in conversation with the staff at the IBM booth).

  1. 3.       Services pure plays, Analytics pure plays, and SaaS enabled analytics firms

Again, it’s difficult to get clean breaks, but this third bucket may be thought of as companies that lead with services. That being said, the models can be quite different. For example, Mu Sigma is a firm that has collected well over $100 million in funding (backed by Sequoia as well). The approach appears to be a pure services arbitrage model in which they can source talent from India and undercut the traditional consultancies here in the US. (Interestingly for the market research industry, many of their testimonials are from market research managers and directors.)

The more interesting model in this category is represented by a new breed of firms such as Decooda, Nuevora and 1010Data who are not only providing the analytical services, but also offering the analytics platform through cloud based infrastructure. While 1010Data delivers more BI services, Decooda and Nuevora are pioneering the delivery of predictive analytics “Apps” via the cloud. As discussed above, the economics of cloud based computing and the advancements in data mining are allowing these smaller and more nimble firms the ability to put value propositions on the table that heretofore have been reserved for much larger companies; not to mention they can compete without the hefty overhead.

m4s0n501
Share
You can leave a response, or trackback from your own site.

3 Responses to “Framing the World of Predictive Analytics”

  1. Framing the World of Predictive Analytics | Best of Research on Market Research (Quantitative) says:

    April 20th, 2012 at 11:32 pm

    […] via Framing the World of Predictive Analytics. […]

  2. Austin Wells says:

    June 25th, 2012 at 6:20 pm

    Tony,
    Just happened across this post. Quite interesting from a variety of perspectives. I recently joined RainKing as CMO after spending years doing predictive and descriptive analytics (I appreciate your distinction in the article) in the anti-fraud and anti-money-laundering world. We’ll be building upon what we started here. One thing I’m seeing is that the source of the data (e.g. crowdsourced or not) is relevant to the analytic techniques and use cases mostly when it varies in quality (e.g. crowdsourced data often has spotty quality). Equally important, the depth of information in the data has a big impact on its usefulness in being predictive (e.g. knowing someone has a title “director of infrastructure” isn’t as useful as knowing what the “director of infrastructure” is responsible for, what pain points he has, what projects he’s running, where he’s spending, and what technologies he owns.). Many of the use cases I’m seeing aren’t solvable with any one source of data or one analytical tool (e.g. Lattice, RainKing, etc). In fact I am seeing some customer driven use cases (large conglomerates) who we are working with to combine internal data (e.g. install base, new product lines) with even driven data, with CRM data (e.g. pipeline stage, duration, what products, who, how much) with Eloqua/Marketo driven “lead scores” (not predictive, just weighted scores) with deeper intelligence (e.g. role, technology, pain points and spending info from RainKing) to predict what variables correlate with purchase (who, what products, how much, where, when, etc). Without the right underlying data against which to compare “good” (ie buyers) and “bad” (ie non buyers), they can predict pieces (e.g. what new products are customers of old products likely to want) but miss other aspects (e.g. who should they give the new proposal to!?). In any case, I too find this an interesting area with a lot of opportunity for growth, so I thought I’d reach out as I liked your coverage of the broader topic.
    Best,
    Austin Wells
    CMO
    RainKing

  3. Tony Cosentino says:

    August 3rd, 2012 at 12:36 pm

    Austin,
    Sorry for the delayed response!…Very thoughtful comments. Thank you. You and I should chat at some point. I’m getting quite deep into all of this now that I’m with Ventana and I think you and I could have a fruitful conversation. Tony

Leave a Reply

*

%d bloggers like this: