Mining the Gold in Powerpoint Decks (9-8-10)

To expand on the Rumsfeld "We don't know what we don't know" pronouncement, most enterprises don't know much about what they do know! This time I will deal with the omnipresent Microsoft Powerpoint and its tendency to make eminently reusable content unavailable.

How many Powerpoint presentations have been created since this Microsoft product became available? Millions upon millions, we suspect -- the fruits of billions of dollars in costly labor. While these decks were created to illuminate a subject or situation, they now imprison within a proprietary format the information they once cameoed.

Because, at CAI/SISCo, Powerpoint deliverables are created to convey the details of every Price To Win (PTW) engagement our firm completes, we have long seen the need to free information trapped thusly. If the nuggets within could be mined and added to our knowledgebase, they could be easily found and re-used. So, what to do?

Many decades ago, I worked for Control Data Corporation (CDC), where the venerable Seymour Cray created generation after generation of early super computers. One of Cray's favorite truisms concerned the need to hire bright, young engineers because they didn't yet know that the problems on the frontiers of technology were impossible to solve. Following this impeccable logic from this early cyber innovator, CAI/SISCo hired a Phd candidate as a summer intern and gave him this tasking:

• Free the information imprisoned in the thousands of Powerpoint presentations that CAI/SISCo has generated over the years; and
• Accession the freed information to our Lotus Notes-based InfoCenter.

It turns out that extracting the useful knowledge from Powerpoints is not a simple task. Several off-the-shelf statistical natural language processing libraries had to be compiled into a new application. The app uses a statistical model that is trained by manually annotating representative samples of presentation subsets. Subsequently, the "trained" model processes each deliverable file and extracts targeted information. In leave-one-out cross-validation within a subset of 300 deliverables, the tool achieved mean 91% accuracy (8% standard deviation) identifying and extracting target information with no human intervention; a GUI tool allows humans to verify the extracted information and correct errors. The process allowed us to accelerate the processing of the backlog and make this data available to CAI/SISCo analysts.

If you have read some of CAI/SISCo's earlier tracts on the subject of Infocentricity, you will know that we consider the husbanding and pre-positioning of information for possible reuse the foundation of modern business development. CAI/SISCo is continuing to adapt our workflow procedures and knowledge-capture strategies to provide the best, most thorough view of the business capture landscape. Accordingly, we have incorporated the fruits of our intern's labors (tentatively called De-Powerpoint) into our InfoCentricity offerings.

To help business development organizations design, develop and implement their own versions of Infocentricity, CAI/SISCo is now offering training, consulting and implementation services that can help enterprises transform from collections of individuals into information powerhouses that support success, growth and survival. Please contact us to request an Infocenticity consultation, a demonstration, or a proposal.

Good luck and happy hunting!