To label this trend as 'big data' is an oversimplification. While it's certainly true that there's more and more data available to analyze, many of the most common analytics requirements in business today are focused on relatively small datasets. The common theme across every use case is the demand for fresh data — processed overnight or in real-time if possible — presented in a readily digestible format.
Big data jockeys may want to assess the relative merits of AWS Redshift against Hadoop clusters and more traditional data warehouse platforms. Lesser mortals will simply want to run predictive analysis over their Salesforce.com data. What everyone wants is to work with the latest data in a way that allows them to reach timely, effective decisions — fresh analytics, served daily.
With demand rising fast, it should be no surprise that the number of solutions on offer is exploding equally quickly. While choice is in principle a good thing, it can prove confusing, lengthens the selection process and increases the odds of getting it wrong. Many of the up-and-coming analytics vendors are cloud-based — Tableau's focus on desktop and workgroup server implementations is a notable exception — because the elastic capacity of the cloud is well suited to the bursty nature of analytics processing. But there are several different approaches on offer.
While visiting San Francisco earlier this month, I had a chance to chat with Brad Peters, CEO of online analytics provider Birst. He outlined what he considers to be three failed approaches to cloud-based business intelligence:
- Host an existing product on a virtualized cloud platform (otherwise known as the ASP model or SoSaaS). This simply ports all the existing limitations of the legacy product into the cloud.
- Assume that you can productize analysis and offer pre-built packages, on the basis that everyone's analysis needs are the same (they're not). This approach is often seen in sales performance analytics products.
- Put a visualization component in the cloud but leave the heavy lifting of knocking the data into shape to the customer. (As an aside, last week at SuiteWorld I had a demo of the Connection Cloud, the brainchild of database pioneer Roger Sippl, which takes on that 'knocking into shape' role of connecting real-time data out of SaaS applications into analytics packages).
Birst's alternative approach is to include in its platform both the data management (what data warehouse vendors used to call ETL) and the analytics. "Data does need to go on a journey from source to pretty pixel and ... data management is the critical gap to get over," Peters told me.
The same approach is used by another up-and-coming vendor, Tidemark, whose CEO Christian Gheorghe was interviewed here last week. Gheorghe has described the approach — which at Tidemark runs on a powerful Hadoop cluster — as like ELT, but with the order of the 'T' and 'L' reversed: "When you do ELT — that's what cloud as a computing platform enables. You're putting everything in a grid and the transformation actually happens at runtime, when you ask the question."
The variety of contrasting approaches in the market is symptomatic of a category that's in its rapid evolution phase. Users are educating themselves — often expensively through trial-and-error — as to how analytics can best help them achieve better business outcomes. They want the data that goes into the process to be fresh and up-to-date; they want it served up with meaningful visualization, context and presentation. They may still have more to learn about the intervening phase of preparation to ensure that what comes out is fit for consumption.
Disclosure: SAP is a diginomica premium partner at the time of writing.
Photo credit: © fudio - Fotolia.com