The worst kept secret in the mobile industry is out: disgruntled Spinvox call centre staff have been telling the BBC that they’re not actually robots, or even highly advanced man-machine cyborg hybrids. But this is a surprise only to the gullible, or people who have never used Spinvox.
Spinvox is a voicemail to text service that purports to use “D2 … a combination of artificial intelligence, voice recognition and natural linguistics”, when in fact, the BBC claims today, low-paid sweatshop staff in foreign countries do almost all of the translations. This had been known for ages – with these photographs popping up last year.
The company told the Beeb that “when necessary, parts of messages can be sent to a ‘conversion expert’”, but declined to elaborate on the proportion of messages “automatically converted”.
Spinvox charged a fiver a month for the service, which sent you a text message transcript of a voicemail after around 5 to 10 minutes. The SMS was spoofed to appear as if it came from the caller, allowing you to call or text them back easily. With voicemail expensive and/or time-consuming, and dogged by HI issues, (every voicemail has a different UI,) Spinvox was a hit – it provided instant convenience.
The company was conceived by a former Psion veteran – Daniel Doulton – who had helped launch the first low cost DAB radio, Psion’s WaveFinder, and launched in late 2004. The timing was excellent – Orange had let its expensively acquired Wildfire “virtual PA” service run down, finally shuttering it in May 2005. Vodafone had for years employed real humans to transcribe your messages. But these were native speakers, and more expensive. Spinvox used the much cheaper virtual sweatshop approach.
Spinvox had tried to go public, and raised £100m from Goldman Sachs last year. But it’s evident that problems were mounting.
In its last annual accounts for 2007, SpinVox employed over 300 people, and ran up operating expenses of £30.445m. Of these, $17m was marked for development and operations, $9.1m sales and marketing expenses, and $4.25m admin. The “burn rate”, or net cash absorbed by operating activities, was £24.45m. However, it only recorded £2.057m revenue.
In order to save money this week, it offered to pay its staff in stock. But more significantly, in a few short months the service had deteriorated to the point where it wasn’t worth a fiver a month. For this reporter, one of the final straws was receiving this text:
“Hi Andrew how you doing nice so call use my charm man cool I guess but what happening I got a big bomb that I wanna drop facing call me back and also leaving help I guess to you and your next see ___ but I post there you close but it keep on making forest Everest get it right I get you later on ___ you include there cheers mate bye bye.”
– spoken through Spinvox
Finally, staff in Spinvox call centres in Egypt have come forth and rumbled the operation. Spinvox has operated (or continues to operate) call centres in India, the Philippines and South Africa. The problems only really escalated this year, as the transcriptions became slower, and contained more blanks, more wild guesses, and more errors. I loved the service, but after a few months, it wasn’t hard to quit.
Two points are worth noting.
If a service works well, as Spinvox did for so long, customers don’t really care if the transcription is performed by humans, robots or chimps. Spinvox’s decline set in when it failed to scale its operation while maintaining quality. Secondly, there was undoubtedly a degree of kidology from this modern day Mechanical Turk. But you’d only have to use the service for a day to see that it was human-powered. Any prudent investor would have asked how many staff Spinvox employs, and how many messages they process, to determine its cost base and scalability.
The Artificial Intelligence claims were exaggerated, of course, but the BBC should pause for a moment before patting itself on the back. Spinvox’s business prospects remain rather better than those of Twitter, Spotify, or any revenue-free Web 2.0 company – since Spinvox has real customers paying real money (if not enough to turn a profit) Critically examining the claims of these “new media” wonders is something the BBC hasn’t been particularly good at – partly because it has so many “new media” evangelists and disciples of its own, whose own prospects are heavily tied to these fictions being maintained.
Specifically, when it comes to propagating AI fantasies and other junk science, the Beeb’s coverage has been worse than any tacky little cable channel. It’s BBC exposure that catapulted Kevin Warwick to fame, and big budgets were thrown at the Singularity recently (see BBC abandons science – and your mailbag).
So when it comes to making exaggerated technology claims – the bigger the lie, the more reporters repeat it