Blog Post

Want Tempo’s new calendar assistant? You’ll have to wait for its AI to catch up

SRI International’s new virtual assistant venture Tempo AI was hoping for a lot of interest in its new smart calendar app. But it never expected the huge demand it received when it launched last Wednesday. Tempo told GigaOM that on its first day it experienced a load on its servers 24 times higher than it expected. That led the startup on Thursday to start restricting new registrants to a few thousand each hour. This week it is halting new activations completely so Tempo can catch its breath.

Tempo’s new app uses many of the same artificial intelligence technologies that went into Siri to generate a smart calendar that infers appointment details and context from your other social media and messaging services. Tempo parses all of the data in a customer’s email accounts, address books and LinkedIn(s lnkd) and Foursquare profiles in the cloud using Amazon Web Services(s amzn). That’s where it ran into problems.

The Tempo AI team
The Tempo AI team

According to CEO Raj Singh, it takes a huge amount of computing resources to bring new customer online. Its platform must initially cull through all of the data in the customer’s various email and social media accounts. Once the customer is on-boarded the burden on the AI lessens, though it does reprocess all of that data on a regular basis – any time new email or contact data is added to system, Tempo can generate new semantic links between new data and old.

“There is just generally a ton of CPU to make all of this work; processing data takes time and we don’t get a network effect, since we have to process each individual’s data,” Singh said in an email. “We re-process data constantly; to be semantically relevant and contextual, we’re constantly re-processing, this is very expensive (it’s like Google constantly re-crawling)”

Tempo was not only surprised by the sheer volume of new registrants – last week Tempo estimated it had a backlog of more than 100,000, but it now believes that number is conservative – it was also caught off guard by the amount of data each customer had. Singh said the average customer is linking 2.5 email accounts to their calendar. Tempo’s servers are getting slammed in both directions: they’re processing more new customers than expected and each new customer has much more information than anticipated.

That led to Tempo’s decision to put a halt to new activations for the next few days. It will finish parsing all of the current email accounts for those who have successfully registered, and it has submitted to Apple(s aapl) an update to its iPhone app that contains a built-in reservation system (right now the app simply won’t let you sign up). Once the reservation system is in place, it will begin allowing new customers in gradually as CPU resources allow. (Update: the new version of the iPhone app is now live on iTunes.)

Tempo AI screen shotOf course, since Tempo’s platform is hosted in AWS, it could simply buy more CPU time to get over the hump. Singh said he wouldn’t go into the details for competitive reasons of how Tempo is managing its backend, except to say that the amount of computing resources it needs to overcome the backlog would be very expensive. Plus, once Tempo brings all of these new customers on board, the demands on its servers will drop considerably.

Tempo certainly isn’t the only smart calendar app in the market. On Tuesday, Sunrise debuted a new smart calendar app, which my colleague Erica Ogg just wrote about in detail. Meanwhile, personal data search startup Cue (formerly Greplin) has been offering an intelligent calendar since June.

Because of Tempo AI’s pedigree from SRI and its associations with Siri, though, its app was always going to get a lot of attention (some rave reviews about the app also helped). Customers haven’t responded kindly to Tempo’s delays though. Of the 597 reviews on iTunes today, 435 were one-star skewerings. Tempo said that once the reservation system in place, it’s hoping it can do a better job explaining the reasons for the delay.

Feature art courtesy of Shutterstock  user Gena96

9 Responses to “Want Tempo’s new calendar assistant? You’ll have to wait for its AI to catch up”

  1. This is putting my non-Tempo hat on. I think the problem w/ iOS is you can’t really do an invitation system. Unlike Android, where you can distribute a link (and not list it in the App Store), on iOS, you have to distribute via the App Store and so regardless of whether you do an invitation system, the app will still be in the store, downloaded and 1-star’ed b/c a user doesn’t have an invite. I think the reservation system is the new invite model (eg App Launchrock) for mobile and I’m actually surprised companies don’t yet offer this as an SDK ;) It would have certainly helped us as opposed to engineering from scratch!

  2. pankaj garg

    crap crap crap! i would never share all my accounts with anyone to make sense out of it.. its like inviting the trouble!

    it will be hard for me to believe if they claim they have any serious customer..but maybe its a good news that i cant think like people who are not serious :)

  3. Reservations will be the next new annoyance that appears on the App Store until something is done. Mailbox may have created the monster but I suspect others will try to leverage this hype machine generator that “reservations” are becoming.

    Downloading a non-functional app should be banned from the store. If developers want to do a reservation system they should do that on their own website.

      • NoThanks


        I waited patiently for a month, watching the reservation line tick SLOWLY to 800 only to find my place in line had jumped to 14K this morning.

        If these people can’t even manage a line, why would ANYONE trust them with their data?

  4. Kevin Fitchard

    Hey John,

    Actually a lot of startups and even huge companies limit the number of activations for a new service. There’s Pinterest on one end and Gmail on the other. I think we can forgive Tempo for wanting to control how they scale. But I do agree with you that they could have planned for this contingency a little better by having the reservation safeguard in place a tlaunch. I originally tested out the app on an iPad. When I tried to activate it on my iPhone this weekend, all I received was “we’re not taking new registrations” message and that’s it. Not exactly what you want to see if you’ve just gone through the bother of downloading the app.

    The reservation system is now live, btw. There are already well over 21,000 people in the queue already.

  5. John S. Wilson

    I think this is silly and shows poor planning. For one, if they could only handle so many downloads than they either shouldn’t have engineered a huge PR push or they should have priced the app so they could afford more AWS time. Can’t have it both ways. If the PR is 100% then the backend needs to be prepared for the onslaught.

    • Robert Scoble

      I was involved in the PR push. Tempo had many times more users than even their wildest dreams indicated would show up. This is a common problem with startups. Most don’t get many users at all, but when something really is highly valuable (as it is in Tempo’s case) it goes viral.

      Part of my job at Rackspace is to find companies that potentially could go highly viral and make sure our internal teams are ready to act if they are customers. Unfortunately in this case there’s nothing we can do because they are on Amazon, but there’s a lot we can do if they are on Rackspace Cloud and we have many programs to help startups at

      That said, scaling is very difficult. Remember all the problems Twitter had for its first four years of life? It’s very hard for small teams to plan for every possible outcome.