• Home
• About
• Writers
• Landscape
• Wizards Index
• Fraud
• Wanna Be Like Larry?

Search Engine Convera Drifts Off

The journey was a long one, beginning with scanning marketing brochures in the 1990s has filed for a certificate of dissolution. I think this means that Convera has moved from the search engine death watch to the list which contains Delphes, Entopia, and other firms.

Convera splash page on February 15, 2010

You can read the official statement for a few more days on the PRNewswire site. The title of the announcement is / was, “Convera Corporation Files Certificate of Dissolution, Trading of Common Stock to Cease after February 8, 2010 Payment Date Set.” I am no attorney so maybe my lay understanding of “dissolution” is flawed, and Convera under another name will come roaring back. For the purposes of this round up of my thoughts, I am going to assume that Convera is comatose. I hope it bounces back with one of those miracles of search science. I am crossing my wings, even thought each has a dusting of snow this morning. Harrod’s Creek has become a mid south version of Nord Kap.

For me, the key passage in the write up was:

Convera Corporation announced today that it filed its Certificate of Dissolution with the Delaware Secretary of State on February 8, 2010, in accordance with its previously announced plan of complete dissolution and liquidation.  As a result of such filing, the company has closed its stock transfer books and will discontinue recording transfers of its common stock, except by will, intestate succession or operation of law.  Accordingly, and as previously announced, trading of the company’s stock on the NASDAQ Stock Market will cease after the close of business on February 8, 2010.

My Overflight search archive suggested that Excalibur Technologies was around in the 1980s. The founder was Jim Dowe, who was interested in neural networks. The notion of pattern matching was a good one. The technology has been successfully exploited by a number of vendors ranging from Autonomy to Verity. Brainware’s approach to search owes a tip of its Prince Heinrich hat to the early content snow plowing at Excalibur. Excalibur used most of the buzzwords and catchphrases that bedevil me today, including “semantic technology.”

Sample of a category search on the Retrieval Ware system. The idea is that you would click a category.

One of my former Booz, Allen & Hamilton colleagues made some dough by selling his ConQuest Software search-related technology to Excalibur Technologies. The reason was that the original Excalibur search system did not work too well. Excalibur, according to my Overflight archive, described itself as “leading provider of knowledge and media asset management solutions.”

Retrieval Ware search result, which you could see via a key word query, a “natural language query,” or by clicking one of the categories in the category “view” interface.

This is a “report style” interface.

The ConQuest search system was essentially a manual approach to query expansion. If you had a document about “trucks” and there was a surge in “SUVs”, then a person had to hook the words together. Excalibur’s management team though this approach was the cat’s pajamas. Manual indexing is a deal breaker unless you use the hybrid method supplemented with semi autonomous agents like our pals in Mountain View do. (Sorry. No footnote because this is in my new Google study.) My recollection is that the ConQuest deal was in the $30 million range, maybe as high as $33 million. Not bad for a system that relied on manual indexing of word lists.

Example of a Convera “visual search”.

The name Convera emerged from a tie up between Intel Corporation and Excalibur Technologies in 2000. The idea was for Intel to build big data centers according to my notes from that time. The Intel crowd looked at a number of search systems, including one developed by Seymour Rubenstein (founder of WordStar) and Inktomi but decided on the Excalibur approach, its manual synonym expansion, and its buzzwords. (I think that note to myself was humorous, but it was a decade ago.)

The deal moved forwards and then about 36 months later became a financial event. Convera report a loss of about $18 million on its 2003 revenues of about $30 million.  By 2004, Convera, according to my notes to myself, had a net operating loss north of $150 million.

Convera as an eCommerce search system. This is the one example I had in my files. The customer was www.eruggallery.com. The double g’s and the double el’s are tricky.

Okay. That’s no surprise. Search is a difficult business and the loss provided me with some supplemental information about the cost of servicing search and retrieval systems, the cost of coding to eliminate deal breaker bugs, the cost of marketing, the cost of customer support, and the cost of research and development.

Convera offered one of the first rich media indexing services. The company cut a deal with the National Basketball Association, and it was a bit of a challenge to the Convera management team and the company’s engineers. I noted that Convera could market but it could not deliver an affordable system.

The Core Search System

I am in the mood for a walk down memory lane. Let’s consider what Retrieval Ware offered a licensee. We need to keep in mind the financial scorched earth that the system seemed to create since 2001 or so. The idea was a “kitchen sink” approach to search and content processing. Convera’s technologies could do any information process a customer needed. Even today I can name a couple of vendors who make the same assertions to their prospects and customers. In my experience, this kitchen sink approach makes sense to the poobahs, satraps, and azure chip consultants but not to me. Frankly I think that the idea of buying one system, having it handle the different content processing tasks in an organization, presenting the results so busy mangers can get answers, and generating ad hoc reports is a big, expensive job for all but a handful of outfits.

The core of the Convera system, based on my experience, was a series of numerical recipes that worked as long as there were sufficient system resources. Without the infrastructure, the Convera system was sluggish, so functions had to be turned off. Chopping down functions and choking content produces a number of challenges for the APRP or Adaptive Pattern Recognition Processing foundation. In fact, my work with Convera provided me with the view that semantic methods should be plumbing and kept out of sight. Furthermore, these methods should be used * only * when there were sufficient resources; otherwise, other parts of the system would either not work or not work in a useful manner.

I have a note about APRP that reminded me, “Why can’t these Convera guys just tell their customers about the bandwidth, hardware and storage demands before signing a deal?”

Convera offered a “distributed process architecture,” a phrase which reminds me of Google’s plumbing.

Convera was a layered search and content processing system. At the base were the methods of the original pattern recognition and neural network technologies. Then there was the ConQuest system. I think Convera’s engineers coded additional components and may have licensed technology from some other vendors.

The result was a system that might have had satisfactory price/performance numbers if the computing infrastructure available today was available in the 2001 to 2005 period. As it turned out, the costs were too great and there was the financial melt down.

Convera kept making deals in US Federal government as the Convera investors and management team tried to find a money making approach. One angle was to index the Web for “vertical search”. I never understood this but the company whipped up a Web indexing system and invested in servers and storage to index the Web. The technology was okay, but it was the cost analysis that was the problem.

Unlike Google, Convera had to charge customers for the service. Google figured out how to charge advertisers to reach a user base. Convera never figured out a business model that would work. Convera caught the eye of smart money firm Allen & Co., and the rest of the Convera story is one that is anchored in financial activities, not search technology.

I used the phrase “Convera approach” to refer to describing great functionality and then struggling to deliver what the sales professionals and marketing collateral described. This is a common problem and it is one that is used even today.

Retrieval Ware

It is entertaining to think about how Convera’s Retrieval Ware solution worked.

The system was described as having modules or subsystems. For example, you as a Retrieval Ware customer, could tap into:

Classification of Processed Content

The idea is that a licensee would shove content into the Retrieval Ware system and the Retrieval Ware software would assign consistent, standardized index terms to the document or other content object. Keep in mind that Convera pitched its video indexing capabilities to the NBA, and the NBA became a non-customer after a period of time.

Profiling

The idea is that Retrieval Ware would perform Selective Dissemination of Information (SDI). The thought was that a manager could plug in the word or phrase germane to his work and then the Retrieval Ware system would pass documents matching that “profile” to the manager. Today we know that this PointCast, BackWeb, Desktop Data approach is problematic. But in the early 2000s, this pitch sounded great to folks who had a problem but did not know what they did not know.

A Workbench

This was the ConQuest component which a licensee would use to fiddle with the indexing dictionaries. The idea was that long before the azure chip crowd discovered taxonomies as a consulting business, Convera provided its licensees with a way to stick a subject matter expert into the system. It doesn’t take much thinking to understand that a human intermediated approach is not much more consistent that automated classification and more humans are needed when content volume goes up.

Cartridges

The idea for a cartridge, according to my notes, came from the database crowd. The idea is that a clunky Codd database could be given some new life by creating code that would add functionality to the row and column framework. Convera offered canned dictionaries for certain disciplines such as law enforcement, defense, etc. These were described as cartridges, which I think appealed to some of the procurement people at the Department of Defense.

Synchronizers

Convera said that it had technology that would access different repositories of content. This is now called federated search or indexing via connectors (close enough for horseshoes). The idea is that Retrieval Ware could “read” the documents in different types of file systems, file formats, and third party systems. When you ran a query, the Retrieval Ware system would deliver results from all these sources. The idea sounded great in 2001 and it still sounds great today. Some companies can deliver this functionality; in 2001 no company could deliver this functionality. (Marketing collateral is easier to create than affordable, stable systems I have learned.)

Web Spider

Convera had a Web spider, but not just a spider that could index text. Nope. The Convera spider, according to my notes, was able to index audio, video, and images. (I found this assertion specious in 2001, and I am today able to point to a couple of outfits who can deliver this functionality; for example, Google and Exalead.

File Room

Convera offered a component that would be a document repository. A user could view a document processed by Convera. FileNet, Teratext, and iPhrase were among companies offering a similar repository function when Convera was making this assertion. Today this is a big business so Convera knew about an opportunity, but it could not nail it.

The Convera Web site is still up, and I would not be surprised if the company continues in some form in the months ahead. The Web site lists “success stories” and offers a free trial. The spin, however, is not search but Web traffic. I think that Convera is mostly a system for generating clicks, not a leading provider of

Screening Room

Convera said it could processing video with proprietary functions for capturing video such as TV news programs, index the rich media, provide a search and retrieval function, and serve the video. Great idea which Google and other companies are still working on this functionality. I don’t recall if the search system was called Visual Retrieval Ware or something else, but the note I made to myself was, “What’s the cost of supporting multiple search systems?” This is a question that I could pose to OpenText and Yahoo today. I wonder if these firms have analyzed the Convera case?

Retrieval Ware SDK

Convera offered a software development kit. The idea was that licensees and developers could build new functionality and tap into the Convera ecosystem. In my files, I have lists of buzzwords attributed to the SDK but like much of the company’s descriptions, the words are surprisingly fresh in 2010 and still in use by some search vendors. (If you look at the write up for a certain azure chip firm’s four square graph of the search industry, you can find the Convera language passed as fact today.)

Observations

I want to capture some of my thoughts after my romp down memory lane.

First, Convera is a great case study about the risks of over promising and then hitting the customer with the costs of building an infrastructure, hiring the system administrators, and paying for the engineering services to make the marketing literature sync with reality. The Convera case shares some elements with the Fast Search & transfer SA business. My position is that serious management skill, technical expertise, and a lot of money will be needed to keep any search vendor afloat who follows the Convera marketing and sales method that does not make clear the resources required to get the system working as advertised.

Second, the Convera case underscores that those people who are confident in their knowledge of search have essentially zero expertise that bears on search and content processing success. Clients continued if I understand the Convera history to license the firm’s technology despite the yellow flashing lights that I saw a decade ago. I find it amusing that self appointed search experts do not understand the basics of figuring out what works and what does not work when it comes to search and content processing. The Convera case reminds us of the weakness of certain procurement processes and the failure of basic financial analyses prior to licensing a system.

Third, Convera provides a useful library of marketing buzzwords. In fact, when I flipped through the Convera information in my Overflight system, I was impressed with the wording and the timeliness of some of the phrases. In fact, I thought I was reading about one of today’s search and content processing systems, not one that was in “dissolution”, whatever that term really means.

To wrap up, if you know more about Convera, its support for its licensees, the status of its code base, please, use the comments section of this Web log to update me. I have a complete work up on Convera, and if you want to talk about a for fee run down, write me at seaky2000 at yahoo dot com. The briefing addresses the question, “Is there something to buy in the Convera property?”

Stephen E Arnold, February 16, 2010

No one paid me to write this. I did have to review Convera’s technology years ago, and I recall objecting in one meeting to the Intel decision to go with Convera. But no money for this opinion piece. I will report this miserable reality to the FAA where high flying is the core focus of that Federal agency.

Comments

Copyright © 2008 Beyond Search · ArnoldIT.com · Disclaimer · Revolution News theme by Brian Gardner · Log in