But Australian Telecom loves it!


There are two popular ways of measuring how widespread a problem is: You can look at the number of times the problem occurs, and you can look at the number of users it happens to.

Back in the late 1990's, I remember a presentation from a product team on things they learned from their telemetry. One of the most important things gathered from telemetry was deciding which bugs were most important to fix. (Related reading.)

The presenter displayed a chart of the most frequent failures, and for the purpose of discussion focused on one of them. The failures were plotted on a distribution curve, where the height of the curve showed the number of times an individual computer encountered it. And although the total number of failures was high, the distribution curve revealed that pretty much all of the failures were coming from a single computer. The problem wasn't so much that the feature wasn't working for a lot of people; rather, the feature was working great for everybody, except this one poor guy where the feature never worked at all.

Why wasn't the feature working for that guy? Who knows. Maybe his hard drive is corrupted. Maybe he has a computer virus. But whatever is wrong, it's affecting just one person. (And since telemetry is gathered anonymously, we have no way of contacting him to find out. This was before we had a way of asking people to leave their phone number so we could call them back with further questions.)

In a similar way, one user complaining 10,000 times is worth far less than 10,000 users complaining once each. (And psychologically, one user complaining 10,000 times is actually worth less than one user complaining once, because one user who complains 10,000 times starts to sound like a crackpot.) And a corporation with 10,000 employees can say "This affects all of my 10,000 users," and they effectively cast 10,000 votes.

But you have to be careful you don't give too much weight to the corporations who can cast 10,000 votes at once.

Windows 95 came with an email client known as Windows Messaging. That was the thing that ran if you double-clicked the Inbox icon on your desktop. This component was developed by the Exchange team, and sometimes I would send them feedback saying "When I use your program to read my MSN mail, XYZ happens, but that doesn't really make sense for MSN." And the Exchange folks dismissed the feedback by saying, "But Australian Telecom loves it!"

Apparently, Australian Telecom (now known as Telestra) was a large install base for Windows Messaging, with tens of thousands of employees. That meant that when they had some feedback, it carried the weight of 10,000 users.

I didn't have the heart to tell them, "That's so cute. You know who your largest customer will be once Windows 95 ships? MSN, with 250,000 users."

Comments (19)
  1. Entegy says:

    Telstra, not Telestra. Also, naming a company without people guessing it beforehand? Is this a ONT first? :O

    1. They did nothing wrong, so no obligation to anonymize.

    2. William says:

      Its name at the time was also "Telecom Australia", hence Telstra (The name makes much more sense in that context).

      If Telstra managed to amass so many votes, I wonder what another Australian government department, who we were told used to proclaim they had the largest WAN in the southern hemisphere for a while (May still be true; it's huge), what they managed to get changed.

    3. Falcon says:

      I'm pretty sure it's not a first.

  2. Boris says:

    I was actually there at the dawn of the third age of mankind, but I merely used Eudora with the email account supplied by my ISP, not MSN or similar services. Later I switched to Outlook Express.

    (I also remember installing Windows 95 on my 386 just in time to catch GoldenEye in a movie theater.)

    1. parkrrrr says:

      I also used Eudora with my ISP's mail account.
      After I helped the ISP debug the problem they were having with the Win95 sockets layer. All of their other users were using Trumpet Winsock, because Win95 was still in beta and I was the only subscriber they had who was crazy enough to use it.
      But even I wasn't crazy enough to use it on a 386. I had a 486SX.

      1. Boris says:

        Sure, but that was before I used a computer for anything serious, so it wasn't like I had to make a cost-benefit analysis. I only had a 386, and it was Windows 95 of all things, with the fancy new desktop and whatnot, so... But yes, I do remember it being slowish.

        (A couple of years earlier, I'd also upgraded from MS-DOS 3.3 to 5.0 on my IBM PC XT clone running at 4.77 MHz with 512 KB of RAM; was also advised that such an upgrade didn't make much sense, but it was DOS 5.0 and it did work without any issues that I can recall.)

        1. DWalker says:

          Wow, I remember the 4.77 MHz processors. Good times!

        2. parkrrrr says:

          The joke here is that the 486SX was the downmarket, pin-compatible version of the 486, with almost no redeeming qualities whatsoever. And because it was pin-compatible, it only had a 24-bit address bus, so system memory maxed out at 16 MB. It may as well have been a 386.

          1. parkrrrr says:

            And I screwed up the punchline. The machine I first used Win95 on was a Cyrix 486SLC, not a 486SX. The SX was actually more or less usable.

  3. BZ says:

    If we're talking about the same thing, Windows 95 actually came with something called "Exchange" (I think named InfoCenter in Beta) which was a combined Email, fax, and CompuServe client. IE 4 (I think) introduced Internet Mail (and News) which evolved into Outlook Express, Windows Mail, and Windows Live Mail. I don't remember whether Exchange and MSIMN ever coexisted.

  4. voo says:

    "And psychologically, one user complaining 10,000 times is actually worth less than one user complaining once, because one user who complains 10,000 times starts to sound like a crackpot"
    I call this the xpclient effect.

    1. Yukkuri says:

      I like this designation.

  5. Joshua says:

    And by the time I was able to run that messenger app it was crashware (yes literally it destabilized the machine it was running on). Putting back the version from the Install CD would fix the machine but break messenger (protocol change?). We finally ended up with this procedure that would block the component from reinstalling itself (*not* like the Windows 10 allegations -- we're pretty sure it was merely the artifact of updates set to auto-approve simply not comprehending we're repeatedly uninstalling this one).

  6. smf says:

    >In a similar way, one user complaining 10,000 times is worth far less than 10,000 users complaining once each

    It can depend on who the user is. Some special snowflakes have the ear of important people and you can get sucked into working round what ever the problem is on their windows install (which can't be rebuilt or blamed) just so that they disappear.

    A lot of problems only appear when you reach a critical mass of users. I've seen it happen so many times and you can usually predict the carnage when someone goes from a really small test rollout to instantly rolling out to all users.

  7. Henri Hein says:

    About your related reading, I suspect bugs follow a Zipf distribution.

  8. David Candy says:

    It was Telecom Australia. My first employer and I and a hundred other people were employed by them when they were 5 days old (5th July 1976). The Post Master General department was broken up and made companies - the Australian Telecommunications Commission trading as Telecom Australian and a sister organisation trading as Australia Post.

  9. Pseudonym says:

    Telstra got in bed with Sun fairly soon after this, so I guess it didn't matter so much in the medium term.

Comments are closed.

Skip to main content