3 ms·

That's a nice theory, in the abstract. In the concrete... replying to email costs money... a lot of money. First of all, just wading through the out of office,

by spolsky 16y ago

That's a nice theory, in the abstract. In the concrete... replying to email costs money... a lot of money. First of all, just wading through the out of office, change of address, and bounces probably costs 10-20 cents each. But answering a real customer email with a real person probably costs $5-$20 depending on the kind of company you are. Now, there are many businesses (like Zappos) for whom it is worth it, and there are businesses (like Facebook) for whom it is not. If you have a million customers and you make about $2 off of each one, well, replying to their email is going to blow any hope you ever had of profiting from them. Customers who like to have penpal relationships with companies are costly customers. Some places might think that losing those customers is no big deal, especially if they have other, well-established ways of getting in touch. For example, the phone company. They're usually a monopoly or a duopoly. They might send out millions of emails to fulfill some legal requirement to notify you that from now on the APR on late fees has gone up. They literally do not need to talk to you about this. They really don't. Anyone that would reply to such a message is suddenly going to become an unprofitable customer. They're monopolies or near-monopolies so they don't care. Don't get me wrong... ALL my email replies work... anything that Fog Creek every sends out and anything that I ever send out to the 60,000 people on my mailing list... I get every single bounce and delete them all by hand. But that's me. The "one size fits all" belief that EVERY company needs to read ALL their email sounds good in theory but it's obvious that plenty of businesses are making a business decision not to send from real email addresses and they're not always doing it out of stupidity.

acangiano 16y ago
> anything that Fog Creek every sends out and anything that I ever send out to the 60,000 people on my mailing list... I get every single bounce and delete them all by hand. Have you considered using a system that handles that for you, like Mailchimp does? Hard bounces never hit my inbox, because they are filtered out by Mailchimp.
ryanwaggoner 16y ago
Some good points, but I think you're exaggerating. First of all, just wading through the out of office, change of address, and bounces probably costs 10-20 cents each. Just like spam, if it's costing you 10-20 cents to filter out these emails, you're doing it wrong. A well-trained bayesian system would probably correctly identify 99.9% of these emails. Sure, you might miss .1%, but you're missing 100% now. But answering a real customer email with a real person probably costs $5-$20 depending on the kind of company you are. And if this is the case, you're also doing it wrong. First of all, probably less than 1% of people are going to respond to these emails, especially things like order and shipping confirmations. Of those that do respond (and assuming you've already filtered out-of-office, bounces, etc), you can probably put the rest into one of three buckets: 1. Emails where no response is required 2. Emails where a simple canned response is sufficient 3. Emails that require some kind of action or lenghty response I'm willing to bet that the majority of emails would fall into bucket #2. The best way to handle this would be a system where the CS agents can see the email and select a pre-defined response with a single click. This is going to cost you almost nothing. Again, bayesian systems would help with suggested responses. Canned responses aren't ideal, but they're better than nothing, and they come from an actual human being who can deal with further messages if needed. Responses to emails in category #3 are where you're right: it would cost $5 - 20 to handle. However, these are customers who need real customer service. Your proposed alternative is to bounce their messages and leave them frustrated. That's exactly what we're trying to avoid. Finally, even if Facebook does make $2 / customer, losing a customer costs them more than $2. PS - Your point about monopolies and legally-required notifications is a good one, though more as an example of why government-sponsored monopolies and arduous regulations are bad ideas.
- kgo 16y ago
  I would argue that if you're really worried about grandma or the stay-at-home mom in Kansas, then even a 0.1% false positive rate is unacceptable, since it'll send a real email into a black hole.
- benatkin 16y ago
  It's also a case of a good problem to have, but a real problem nonetheless. This is where people must be able to come up with a solution. It requires getting comfortable with tags and pasting canned responses. There are many people for whom tools like TypeIt4Me increase email productivity by 100%. This could translate into thousands of dollars a month.
pkaler 16y ago
Ironically, I send replies to a FogBugz mailbox. It works decently. Ever think about building a CRM for developers that connects to a bug database (FogBugz) and knowledge base (StackExchange)? Most CRMs are aimed at sales professionals (Salesforce) or agencies (HighRise).
haroldp 16y ago
I'll agree that there are plenty of companies to whom this does not apply. Facebook is a great example. But I have seen SO MANY small retail companies who send out email to engage customers, and the VERY FIRST thing they say to their customers is, "I don't care what you have to say." It's just nuts. > I get every single bounce and delete them all by hand. Don't do this! Maybe you already know about this, but for the benefit of everyone else following, Email Lesson: 1) Email bounces are not sent to the "From" address. It goes to the "Envelope From". You won't see Envelop From unless you dig through email headers. It's really just part of the SMTP conversation before your mail server starts sending the actual message (header and body). Your desktop email program probably makes these two addresses the same, but your mailing list software needn't. 2) When Joe User replies to an email on purpose, his email client WILL use the From address (or Reply-to, but never mind that). 3) There is a clever idea called VERP (Variable Envelope Return Path) that exploits this, where you make the Envelope From different for every email you send to a list. If you send me an email and your mailing list software made the Envelope From, "newsletter+haroldp=internal.org@joelonsoftware.com" and it bounced, your MTA could route that bounced message to the "newsletter@joelonsoftware.com" account, and use the encoded email address in there to figure out that it was my address that bounced. The To address on the message should point to a human's mailbox. Only real replies (ok, and the stupid out of office messages) will land there. Poorly coded auto-responders are another rant, but I don't hold out any hope of getting everyone to read man vacation. VERP: http://cr.yp.to/proto/verp.txt http://cr.yp.to/proto/verp.txt

by spolsky 16y ago

acangiano 16y ago

> anything that Fog Creek every sends out and anything that I ever send out to the 60,000 people on my mailing list... I get every single bounce and delete them all by hand. Have you considered using a system that handles that for you, like Mailchimp does? Hard bounces never hit my inbox, because they are filtered out by Mailchimp.

ryanwaggoner 16y ago

Some good points, but I think you're exaggerating. First of all, just wading through the out of office, change of address, and bounces probably costs 10-20 cents each. Just like spam, if it's costing you 10-20 cents to filter out these emails, you're doing it wrong. A well-trained bayesian system would probably correctly identify 99.9% of these emails. Sure, you might miss .1%, but you're missing 100% now. But answering a real customer email with a real person probably costs $5-$20 depending on the kind of company you are. And if this is the case, you're also doing it wrong. First of all, probably less than 1% of people are going to respond to these emails, especially things like order and shipping confirmations. Of those that do respond (and assuming you've already filtered out-of-office, bounces, etc), you can probably put the rest into one of three buckets: 1. Emails where no response is required 2. Emails where a simple canned response is sufficient 3. Emails that require some kind of action or lenghty response I'm willing to bet that the majority of emails would fall into bucket #2. The best way to handle this would be a system where the CS agents can see the email and select a pre-defined response with a single click. This is going to cost you almost nothing. Again, bayesian systems would help with suggested responses. Canned responses aren't ideal, but they're better than nothing, and they come from an actual human being who can deal with further messages if needed. Responses to emails in category #3 are where you're right: it would cost $5 - 20 to handle. However, these are customers who need real customer service. Your proposed alternative is to bounce their messages and leave them frustrated. That's exactly what we're trying to avoid. Finally, even if Facebook does make $2 / customer, losing a customer costs them more than $2. PS - Your point about monopolies and legally-required notifications is a good one, though more as an example of why government-sponsored monopolies and arduous regulations are bad ideas.

kgo 16y ago
I would argue that if you're really worried about grandma or the stay-at-home mom in Kansas, then even a 0.1% false positive rate is unacceptable, since it'll send a real email into a black hole.
benatkin 16y ago
It's also a case of a good problem to have, but a real problem nonetheless. This is where people must be able to come up with a solution. It requires getting comfortable with tags and pasting canned responses. There are many people for whom tools like TypeIt4Me increase email productivity by 100%. This could translate into thousands of dollars a month.

pkaler 16y ago

Ironically, I send replies to a FogBugz mailbox. It works decently. Ever think about building a CRM for developers that connects to a bug database (FogBugz) and knowledge base (StackExchange)? Most CRMs are aimed at sales professionals (Salesforce) or agencies (HighRise).

haroldp 16y ago

I'll agree that there are plenty of companies to whom this does not apply. Facebook is a great example. But I have seen SO MANY small retail companies who send out email to engage customers, and the VERY FIRST thing they say to their customers is, "I don't care what you have to say." It's just nuts. > I get every single bounce and delete them all by hand. Don't do this! Maybe you already know about this, but for the benefit of everyone else following, Email Lesson: 1) Email bounces are not sent to the "From" address. It goes to the "Envelope From". You won't see Envelop From unless you dig through email headers. It's really just part of the SMTP conversation before your mail server starts sending the actual message (header and body). Your desktop email program probably makes these two addresses the same, but your mailing list software needn't. 2) When Joe User replies to an email on purpose, his email client WILL use the From address (or Reply-to, but never mind that). 3) There is a clever idea called VERP (Variable Envelope Return Path) that exploits this, where you make the Envelope From different for every email you send to a list. If you send me an email and your mailing list software made the Envelope From, "newsletter+haroldp=internal.org@joelonsoftware.com" and it bounced, your MTA could route that bounced message to the "newsletter@joelonsoftware.com" account, and use the encoded email address in there to figure out that it was my address that bounced. The To address on the message should point to a human's mailbox. Only real replies (ok, and the stupid out of office messages) will land there. Poorly coded auto-responders are another rant, but I don't hold out any hope of getting everyone to read man vacation. VERP: http://cr.yp.to/proto/verp.txt http://cr.yp.to/proto/verp.txt