Today I had the pleasure to attend Donald Farmer and Kamal Hathi's session on the Gemini project.  I've seen several of Mr. Farmer's presentations on BI and Data Mining, and I'm never disappointed at his skills and depth of knowledge!

Gemini is an in-memory, Excel-based, analysis services orientated technology that brings the power of dimensional modeling to a primarily Excel skilled audience--with the promise that the output of Gemini can be forward-engineered into managed Analysis Services solutions. 

Donald's quite entertaining slide deck was a major departure from the typical corporate spin that we're accustomed to seeing (and that I'm personally accustomed to producing) when we educate potential technology users.   

The slide deck casts the analyst, boss and IT admin as silent movie characters, where the analyst and IT professional struggle to keep the boss happy within the limitations of their skills and the existing content in the corporate DW.  It's the all-too-familiar "I need that analysis by tomorrow" versus "it takes time to put that data together".  Terribly apt and consistent with what we see in the "BI trenches" every day.

The silent movie theme stresses that this situation hasn't changed that much in many decades (and in any event in the two decades I've been working on BI solutions).  Of course, the slides were endlessly entertaining while driving home these points!

As for the "Gemini" product, it's quite exciting.  BlueGranite has been in the business of helping our clients' analysts pull info together quickly and make it meaningful since the ProClarity days.  One of my lingering concerns has been that--since ProClarity was acquired and the future of its core technology became uncertain--what tools within the Microsoft product suite would serve these "data diggers"? 

Excel pivot tables against SSAS cubes are fantastic--but how to make sure analysts can get data pulled together by the cube designer fast enough to meet business needs?  If the corporate governed DW/Cubes evolve slowly, how can analysts fill in the gaps? 

Today I still recommend ProClarity for the deep data analysts (as a complement to Excel), as it still provides more flexibility and richer visualization than Excel pivot tables for advanced analysts.  But I think Gemini may one day (hopefully soon) deliver even more, and really become the data digger's tool of choice.

Gemini is clearly intended to fill the analyst gap.  And in so many ways I think it will. It's at once a more approachable way for non-OLAP users to build rich dimensional models, a way to make more dynamic data integrations using the familiar Excel environment, and a rich OLAP query tool to be used against these analyst-generated models.

In truth, I can easily see even seasoned SSAS pros (including myself) using Gemini during prototype and early development, in addition its intended less sophisticated audience.

The things I love about Gemini:

1. Users import data from all types of sources:
   a. Structured relational
   b. Existing OLAP Cubes
   c. Subscribe to "Service Documents" (a form of RSS feed that contains tabular data)
   d. Paste in any tabular data (e.g. copied to clipboard from a web page)

2. Fast processing of large data volumes (100M rows demonstrated, 20M demo'd on a netbook)

3.  Ability to add calculated measures at the Excel pivot-table layer (sweet!)

4.  Ability to connect to Gemini models as SSAS data sources (rocks!)

5.  Tight integration with SharePoint as a basic architectural construct

6.  Translation of most OLAP concepts to Excel terminology more familiar to analysts

Some things I would like to see improved, or clarified. 

I would have asked these questions, but Q&A was cut short today due to time constraints...

1. As simple as they seem to MDX people (like me), my sense is the DAX expressions are going to be too complex for many of the analysts I train.  It reminds me of PPS-P PEL, in that the expressions are intended to be simplified from MDX, but they're still complex and require multidimensional thinking--which isn't a gift many of us are born with.   PEL generated lots of push-back from analysts when I demonstrated it to "real customers", and I fear DAX may as well.

2. To address #1, I hope the product team will consider following the model ProClarity set with it's KPI Designer, which allows users to use wizards to build calculated measures such as ranking and bubble-up exceptions.  KPI designer users build complex MDX without knowing that's what they're doing.  I still train new users on these tools, and it's a positive for them to use wizards (rather than purely language constructs, as with DAX).

3. I'd like to see some MDM tie-in to ensure that already accepted calculations, data sources, etc., can be drawn upon and re-used in a Gemini solution.  I can see a BI governance issue (and IT objection!) if many analysts are building silo BI solutions without some centralized baseline to start from.  Gemini doesn't prohibit a DW/MDM baseline, but it doesn't appear to promote one either.

4. Security really wasn't addressed in the session.  If analysts will be pulling 100M rows of fact data into a desktop solution…how does a corporate IT policy ensure that that such huge volumes of data isn't lost in the back of a taxi?  BitLocker would be a convenient answer to this question, but I hope it isn't the only one.

5. Data mining--not sure if it's possible to incorporate DM models into the Gemini models, but the combination of capabilities on the desktop would be really fantastic!

In-memory OLAP is a hot technology, and as illustrated by Donald's slides, the world has been waiting far too long to put such powerful tools in the hands of typical analysts.  I can't wait to see how this possibly disruptive technology impacts the wide swath of users its intended to benefit!

When planning a deployment of PerformancePoint Monitoring & Analytics (PPS M&A), an often overlooked component in the business solution is the ProClarity Analytics Server (PAS) application layer.  

PAS is sometimes not deployed by customers, and their analysts aren't trained in the effective use of these analytical tools. In my view, this is often a mistake.  Analytics has at least two constituencies--and all need to be considered when planning a successful BI deployment.

A sales specialist I have a great deal of respect for, Mac Hussey, once explained it this way...this isn't a direct quote but my interpretation of his ideas: PerformancePoint server M&A's native scorecards, analytic grids & views cover "Little a" (lightweight analytics), while PAS covers "Big A" (heavy analytics). 

Two A's...so what's the difference? "Little a" is flashy, cool and as easy to use as a web page--often requiring no training beyond a quick overview.  In most industrial and service industries, perhaps 85% of BI dashboard users are executives, managers and line of business users.  These users don't use analytics in their primary job function, but use analytics to support their "real" jobs. 

Even in the long-run, they may only want/need "Little a". If your organization fits this mold, it may seem that reaping the low-hanging fruit using "Little a" is really where all the value lies--and that's certainly true at first. Eventually, though, the "Little a" users want to know root causes, and analysts really need to dig and develop further insights.  That's when "Big A" is needed (ProClarity Analytics). 

"Little a" is certainly a "low hanging fruit picker" and brings in quick ROI in for any BI dashboard deployment.  Yet time spent deploying and training on "Big A" is where the game-changing ROI comes from over the mid/long-term. Currently this means deploying an additional application (PAS) and training end-users.  

Eventually I'm sure we'll see ProClarity Analytics fully integrated into PerformancePoint, and the "ProClarity" product brand will be just a fond memory. When that happens, every user will have access to both the "Big a" and "Little a", even if--as with Excel's features--they only know how to use the most common analytical capabilities offered by the product.  Until then, Microsoft BI practitioners who address the needs of users in the "here and now" will need to continue to pay close attention to the spectrum of user needs when planning BI deployments.

Business Intelligence is a broad topic, spanning a set of technologies and business improvements spanning from enterprise reporting to analytics to performance management.  One of the more interesting areas is in data exploration and visualization.

State-of-the art BI tools enable a level of exploration and data visualization beyond our wildest imagination when I started in BI 15 years ago.  At that time, to achieve acceptable performance virtually everything deployed to a broad audience had to be pre-designed, and with only a limited amount of “ad-hoc” direction by the end-user. The value was still huge—pulling together multiple data sources into a cohesive reporting environment was a jump to light-speed (many companies are still trying to make that jump today).

One of my favorite data visualization tools is the Decomposition Tree.  I love the Decomp Tree because it’s super-intuitive…everyone “gets it” right away. 

The Decomp tree is powerful not just because it can “break down numbers”—we’ve been doing that with hyperlinks on reports for years.  The real power is that it allows users to select his/her own breakdown path, then easily explore that new path, tweak it, and cross-drill across dimension members.

In the screen print above, we started with Geography and switched to fiscal year, then switched to product.  The Decomp tree makes it easy to let the user go in any direction in any order—all very easily. No need for the report designer to anticipate every permutation the user might need!

By delivering this type of visualization to users, everyone wins.  The users win—they get the information they need quickly. Information at your fingertips – delivered! For IT, tools like the Decomp Tree let users access information in the format they need it, rather than prompting yet another custom reporting request.