Web monitoring throws up anomalies. When is a story marked as new and date-stamped? When the page is updated? Or when it is crawled?

Since the process is totally automated errors can occur. For example when an old news item about United Airlines filing for Chapter 11 bankruptcy protection got picked up from The South Florida Sun-Sentinel’s website by a financial newsletter it got pushed into the “most-viewed items” section”. Google news gave it a current datestamp and courtesy of the Bloomberg wire service it was syndicated to the markets. The stock briefly crashed from $12.30 to $3 before the error was spotted.

[Update Saturday 12th September] The Times of London has a good follow-up to this story, tracing the exact sequence of events and drawing attention to the role played throughout by automated systems. Automation caused the story to be listed as the most popular on the South Florida Sun-Sentinel website, Google News picked it up automatically, traders responded to the headline, Bloomberg cited the headline when the stock started moving and automated selling programmes did the rest. Anyone who stopped to actually read and understand the story (complete with references at the stock price of $0.97 which clearly signalled its inaccuracy) would have known the story was old. But they would have been the losers, because every other system was in full “sell” mode.

When we first started pitching to institutions in the City of London a few years back a big fund manager said: can’t the analysis of stories be totally automated? We replied that it could only ever be semi-automated because human intelligence was required to interpret the patterns of sentiment, volume and authority. Part of their questioning leaned on the idea that automated tools for understanding conversations could be linked to automated trading tools. This is a dangerous idea - as the United Airlines story proves.