Imagine a website whose purpose is to encourage consumers to take actions — for example to click on an ad, click on the next page, or actually make a purchase. Best practices for such a site include:
- An ever-evolving user experience, informed by — among other factors — creativity, brand identity, the vendor’s evolving product line itself, and …
- … predictive modeling.
- Personalization based on predictive modeling.
Those predictive models themselves will keep changing, because:
- Organizations learn.
- Consumer tastes change.
- More or different kinds of data keep becoming available.
In that situation, what would it mean to offer the website owner a predictive modeling “application”?
My immediate spur for this post is that Omer Trajman, who seemed to be advising WibiData from the getgo, has finally joined the firm,* and his announcement blog post stresses a need for “Big Data applications”. Coincidentally, I was advising MarketShare this week on some messaging, and was reminded that they (too) self-describe as an application vendor. I don’t totally hate that terminology; after all, both vendors are taking general technology and applying it to identifiable problem domains. Even so, I’ve disagreed with that phrasing in the past. For when you compare what WibiData or MarketShare offers to what we normally think of as an application — e.g. something that might be termed an “application” by SAP or salesforce.com — what they’re talking about is something that’s less standardized or complete.
*When he told me Omer was joining, WibiData CEO Christophe Bisciglia remarked that WibiData didn’t have much in the way of formal titles and roles. However, I’ll note:
- Among other duties, Omer ran field implementation services at both Vertica and Cloudera.
- I’ve argued for years that his employers should give Omer more marketing responsibilities.
- Omer is obviously empowered at WibiData to make some marketing decisions.
Anyhow — if you search here on Omer’s name, you’ll see that numerous past posts have been based on his input.
Those vendors actually do different things; WibiData is stronger in run-time operation and data wrangling, while MarketShare is more focused on supporting complex modeling itself. But in essence both vendors are offering predictive modeling application subsystems. From a standpoint both of completeness and of needed customization, those subsystems can fairly be said to be more “application-like” than, say, a DBMS, but less so than a traditional SAP app.
The completeness point should be obvious; if you’re not providing the consumer user interface, nor even a sample/customizable one, you’re not providing the whole application. So let’s consider customization. Any kind of packaged application needs to be highly configurable; and no matter how configurable they are, they often wind up getting outright customized as well. Yet even so, predictive applications need to be customizable in multiple ways that more traditional operational applications don’t. In particular, they need to be continuously customizable:
- Conventional applications can be based on fixed — or very slowly changing — database designs, even if that design was customized once at the time of installation. But predictive models evolve to view data in fundamentally different ways.
- Similarly, the logic of a conventional application is fixed, or nearly so. But the basic structure of predictive models may change much more rapidly.
- Changes in the user experience, which keep happening, directly change what’s demanded of the predictive modeling subsystem as well.
Much of what I’ve just said holds true for business intelligence as well; you may buy BI as an “application”, but then you’ll keep programming in it for as long as you have it running. Planning/optimization technology could perhaps be viewed the same way too, although to date that’s been a more troubled and less successful sector. And so I think the idea of an analytic application subsystem could prove useful when considering how analytic technology is consumed.