What a coincidence. Just when I’ve been fighting with a new spam filter at my office, I run across this article about the duty to your users.
The problem with any Bayesian filter is that it starts out extraordinarily dumb. It isn’t until it’s been properly trained over a period of time that it starts to reach the 99.5% accuracy that the vendors always brag about. I knew this, and I installed the filter fully expecting plenty of false positives and negatives the first week.
The problem was that it wasn’t just running on my computer. This was our new server-based solution, running on our brand new Exchange 2003 computer, filtering everybody’s e-mail. So I warned everybody right off that good messages might end up in the Junk folder for the first couple of weeks. But still, within a couple of days, I started getting complaints from around the office that their e-mails were getting marked as junk. Some of the people asked me to go so far as to turn the filter off. I’ve had to reassure everyone three or four times now that this is just a temporary solution, that the filter will get much better very soon. Still, I can start to feel public opinion turning against the idea of a spam filter, since their first experience with one isn’t turning out to be very positive.
Now, I’m not going to be shutting down the filter. I’m confident that it will get better soon, especially now that I solved the problem that was stopping me from training it manually. And there are only 50 people in the office, so we’re all like a little family. I think I can convince everyone to stick it out for a little while, and they can have patience without storming my office with pitchforks. I’m not sure how you’d handle something like this in a larger organization, though. If you have 50 users, and 10 percent of them are fed up with the new filters, that can be handled though a little personal diplomacy. But what if you have 1,000 users, or 10,000? That ten percent figure starts to become a group with some real power, and if you can’t convince them to be patient, where does that leave you and your new spam filter?
The article brings up some good points about the duty to the users, how it’s the IT department’s responsibility to make sure everyone gets all their e-mails. And I definitely agree with them that the best approach is to filter into a separate “Junk E-mail” folder—not send all spams into a black hole. And I know I can’t ask my coworkers to be patient for too long. We’ve had the filter for one week now; if after another week it’s still junking good messages, I’ll get rid of the program. That’s the advantage to trial periods for software!