After years of false starts speech recognition is starting to make inroads into trading rooms.
Three of the largest equity trading desks now speak, rather than type, their instructions for indications of interest and trade advertisements. Goldman Sachs, Credit Suisse First Boston, and UBS Warburg use technology from Novato, Calif.-based UmeVoice to send out more messages and to reduce errors. Merrill Lynch is currently installing and testing the UmeVoice system.
"In fast markets, traders tell us they can't type fast enough," said Adithya Padala, president of UmeVoice (pronounced "you-me voice"). "UmeVoice lets them send significantly more messages in these conditions. And the more messages they send, the greater the likelihood of doing that next trade."
Voice, or speech recognition, involves the conversion of spoken words into computer text. Its most common manifestation is in the automated voice response systems slowly taking over the duties of receptionists and call center employees. But it has popped up as an office automation solution as well, particularly in the medical, legal and financial fields.
Equity traders who work with institutional clients have targeted the sending of pre- and post-trade messages for voice automation. Invitations to trade and announcements of completed trades are typically typed out by dealers and transmitted to the buyside over networks provided by Thomson Financial's AutEx, Bridge Information and Tradeware.
Three kinds of messages are typically sent. Indications of interest' (IOIs) – both the bare bones interest messages and the more detailed super' messages – are transmitted to generate interest among the buyside for a trade. Advertisements' are broadcast after a trade to draw attention to a broker's trading prowess.
They're crucial deal-getters. But because typing them is time-consuming many aren't sent. "We do so much business here it is impossible to advertise everything we do," said Phil Marber, head trader at Cantor Fitzgerald. "I don't have enough people to advertise all the volume we trade." Cantor does advertise most of its big trades. It's the small ones that get away.
UmeVoice is targeting the top 30 brokers and is in negotiations with a major buyside firm, according to Padala. He lists three main selling points. First, and foremost, the technology enables traders to send considerably more messages, leading to even more transactions. A return-on-investment study done by the company found that traders using the product sent 50 percent more supers and interest messages. That translated into a 10 percent increase in orders.
Second, UmeVoice reduces both communication and input errors. Often, a position trader will shout out or scribble down his instructions. A clerk then inputs the information into the AutEx screen. The clerk, however, could misinterpret the trader. He could also type in incorrect data.
Room for Error
A trader speaking his message directly leaves less room for error. Of course, it also leaves less room for AutEx clerks. In fact, UBS Warburg has transferred one of its two clerks to another position since it started using the UmeVoice technology.
Finally, for traders who do type their own messages, the push-button ease of UmeVoice means they no longer need to break stride and type. They can talk and trade without interruption.
In sum, more messages helps to increase revenues while the elimination of clerical staff reduces overall expenses, pros say.
"It makes the world a lot more efficient," said a trader who did not want his name disclosed. "The fact that every trader on the desk can super' and not have to wait in line for the two AutEx operators is a great thing. You send out a lot more supers because everyone can super at the same time." This trader adds that the system rarely goes down and is user-friendly.
Using it certainly seems painless. A small box called UmeEars connects a trader's turret with his computer. He pushes a button on his turret's handset and speaks into a so-called voice user interface' (VUI). A form on the VUI automatically fills itself out with the trader's words after a pause of about two seconds. "IBM, offer a million, 93 3/4," is how a typical super message might read. The trader confirms the information and then says "send."
The message then travels through the UmeVoice infrastructure into the brokerage's message server and then onto the AutEx and Bridge platforms. From there it is broadcast to the designated institutions.
A trader can correct his mistakes. If he says "1,000" when he meant "10,000" he simply says: "quantity, 10,000." The contents of the VUI form then change. The technology even checks the current price of a stock to confirm the trader's input is not out of line with the market. If a trader says "60" when he meant "16" the VUI will enter "16" if that price is within the day's trading range.
Although UmeVoice recognizes "thousands of words," according to Padala, its limited vocabulary reduces the chances of miscommunication. That's a problem that has limited sales of off-the-shelf dictation software, which must contend with all the words in the English language in millions of combinations.
Most of UmeVoice's vocabulary consists of the names of stocks or their symbols. It also recognizes slang. A trader can advertise 1,000 "GTE" or 1,000 "Genny Phone." The software recognizes both transactions as the same. If a trader invents a new way to say GTE, UmeVoice will incorporate that phrasing into its database as well.
Finally, at the end of each day, a computer sifts through all the audio from all users to find phrases that weren't accepted by the VUI. If the computer discovers that a trader failed to communicate with the VUI, it or staff linguists, makes adjustments.
Building Support
UmeVoice and one-time rival Ficomp Systems both had speech recognition products on the market from the early 1990s. But it took several years and improvements before traders would accept them.
Bear Stearns' fixed income department struggled with and finally abandoned technology from Ficomp in the mid-90s. The New York Stock Exchange and the Chicago Merc also tested Ficomp technology and found it inadequate. The American Stock Exchange tested and rejected UmeVoice in the same period.
Both Goldman Sachs and Salomon Brothers worked with UmeVoice on the floor of the NYSE to voice enable certain data-entry tasks at their booths. But the introduction of the superbooths in the 1997-98 period caused those functions to be automated or moved upstairs.'
Despite the setback, Padala says UmeVoice emerged a stronger product. "We have to thank them," he said. "The New York Stock Exchange is very noisy. We invented our [UmeClarity] technology after realizing just how noisy it is."
UmeClarity is the part of the UmeVoice infrastructure' that reduces background noise. Its development has proved critical to the workability of the product. Loud and unmuffled background noise will mingle with a trader's words, confusing and rendering useless voice recognition software.
That problem is especially acute in trading rooms where noise levels can reach 90, or even 100 decibels. A typical office environment operates at 65 decibels.
It was not until around 1998 that voice recognition won brokers' hearts and wallets. That's when Goldman launched UmeVoice's IOI service on its block desk. It's when tiny agency broker Swiss American Securities, a unit of Credit Suisse, chose Ficomp to voice automate its traders' order entry tasks. Swiss American became the only firm to find success with Ficomp's technology.
Ficomp later exited the voice recognition business to focus on systems integration. Now, because of a lack of technical support, Swiss American is phasing the technology out, according to CIO Martin Schwartz.
UmeVoice has the market to itself for now. That doesn't mean speech recognition is an easy sell.
"We don't consider it absolutely necessary," said an executive with a large brokerage. "We left it up to our traders. If any had said they had to have it, we would've put it on their desk. We didn't get a whole lot of demand for it."
Bill Meisel, principal in TMA Associates of Tarzana, Calif., a consultancy that specializes in voice recognition, is not surprised.
"The trader doesn't believe he is getting anything from it," he explained. "They perceive that stuff as being taken care of. Any problems that come up are someone else's problems."
In other words, traders don't see why they should do something already taken care of by their clerks.
Moving Forward
Padala is undeterred. He's promoting UmeVoice in London and says existing clients are requesting additional functions.
UmeVoice isn't just for trade advertisements. It can also be used for order-entry and order routing.
Traditionally, voice recognition in this area had been a hard sell for vendors. That's because traders were scared to use space age technology for such bottom-line critical functions. Advertisements were perceived as safe' though.
None of UmeVoice's customers are using the technology for these advanced features, but Padala says they are all considering it.
"Extensions are being talked about in all directions," he said. "On the desktop, geographically, and into other areas of the firm such as fixed income."
