ChartSearch: Information Is Out There
Even though BI search is an established technology, information retrieval problems still plague many organizations. ChartSearch says it can help.
- By Stephen Swoyer
- June 18, 2008
When business intelligence (BI) search first arrived, proponents touted it as an ideal solution for the information retrieval problems which plague many organizations. Unfortunately, now that BI search is an established technology, those information retrieval problems haven't gone away.
As a result, a number of start-ups (and even a few veterans) claim that BI search doesn't go far enough, or is too vague, or is -- quite simply -- insufficient, when it comes to helping users ferret out pertinent business information.
ChartSearch Inc. is a BI start-up that champions what founder and CEO Chris Modzelewski likes to call a "search-driven" approach to numerical analysis and business intelligence. In other words, Modzelewski claims, ChartSearch picks up where vanilla BI search tools leave off.
"Among users, there is an expectation of being able to access and search for information, but the traditional BI paradigm falls short. The model has always been the traditional reporting-centric model, the push model of information, whereby an analyst would design dashboards and design reports," Modzelewski says. "The consequence of that is that it doesn't quite satisfy what average users are looking for, which is why we see relatively low usage levels of BI outside of certain well-defined [constituencies]."
A host of other players -- including start-up competitor QL2 Software Inc. -- tout a similar take on BI and analytics. Data federation specialist Composite Software Inc. recently announced its own BI search and analytic offering, Composite Discovery (see http://www.tdwi.org/News/display.aspx?ID=8982). Discovery bridges a chasm, Composite officials claim, between existing BI tools -- which are difficult to set up, use, and maintain -- and BI search, which (they argue) is unaccountably vague when it comes to delivering actionable analytic insights.
Modzelewski concedes that the problem ChartSearch is attacking isn't exactly new; what is new is the approach he's taken with the ChartSearch platform.
"We want to provide a solution for clients that enables them to give a Google-simple tool to the actual BI end users so that they can actually search for the data that they need," he explains.
That isn't an entirely new approach, either (see http://www.tdwi.org/News/display.aspx?ID=7933), but Modzelewski isn't just a fast-talking executive of a hot new start-up. He's a programmer/geek who knows whereof he speaks. Before he launched ChartSearch, he spent several years in the EU building Gemius SA, Europe's second-largest media-rating firm. Along the way, he became convinced that the way in which most organizations go about doing data analysis is inefficient.
For example, users frequently don't have access to the right data, don't want to do the analysis themselves, or (at an organizational level) companies don't want to distract their analysts from doing profit-generating work for clients. The original impetus for ChartSearch stems from this realization, he claims, as does much of the original coding, which Modzelewski originally performed himself.
The idea is to bring analytic heft to bear not just against structured data stored in transactional databases but data that's federated across a range of other data sources. That's precisely what the ChartSearch platform enables, he maintains.
"The architecture of the platform is this meta-layer of information, so when a user enters in a search query, what the system does is it semantically parses that query in order to figure out what it thinks the user is asking about," he explains, citing an example in which a user enters a declarative statement (i.e., "Give me the top 5 sales managers in the Northeast region by accounts receivable.")
ChartSearch maintains an index (the aptly-titled ChartSearch Knowledge Base) of available data -- along with a semantic knowledge base that helps it properly interpret user's queries. The Knowledge Base uses a proprietary XML-based language that ChartSearch calls DataSearch Markup Language, or DSML. According to Modzelewski, DSML generates a near-complete semantic layer of metadata that corresponds to the numerical data in an organization.
"It tells our system what [data sets are] out there, how to connect into them, what the structure of those data sets [is], what business rules to apply to them -- for example, how to actually calculate and aggregate information -- and how to integrate with existing authentication systems."
As far as its actual analytic heft is concerned, ChartSearch uses a Search Term Parser that interrogates the Knowledge Base to construct on-the-fly interpretations of these user-submitted natural language queries. From there, the ChartSearch Engine constructs dynamic SQL queries, which it can submit either to application middleware or directly to data sources.
The final piece of the puzzle is a Chart Generator, which executes queries and determines how to display the data that's returned. "It really has to do with the evolution of user expectations. Google has become synonymous with information access, [so this] sets the expectation that users should be able to search for information like they do when they search the Web. What we do is we empower users down the line with an easier tool, with an easier ability to access information economically," he indicates.
ChartSearch can also go after non-traditional sources of data, such as information from subscription services or raw market research data. For a variety of reasons, Modzelewski explains, dumping data of this kind into an enterprise data warehouse (EDW) just hasn't been feasible -- regardless of the fact that large and influential business user constituencies need to consume it.
"This data all has a certain structure to it, but because of the economics of producing BI reports, and because of the complexity of [parsing something like] market research data, it hasn't been feasible for companies to load some of this data into their data warehouses, regardless of how many users actually consume that data. Today, users have to send off e-mails to analytical teams to research those reports, or pay market research firms to do it. This is another, better way."