Hacker News Reader

Comment by ghshephard | Hacker News Reader

You are off by a couple orders of magnitude on the number of people that a real human responds to. They get paid a few (Say, 5) cents for each text, so, to earn around $10/hour, they would need to send 200 texts/hour (which is pretty easy/trivial if you are doing this full time). Each package includes up to 100 texts/month, so, that is 1 real person per two customers/hour, or 16/day, or 320 customers/month per customer representative.

Looked at another way - one customer representative can support two customers/hour (in aggregate, obviously they don't send all their texts to one person in an hour). The customer representative gets paid $10/hour, the two customers pay $50/hour.

Pretty good business model.

View on HN

> they would need to send 200 texts/hour (which is pretty easy/trivial if you are doing this full time).

Context switching would take a while. You can't just reply with random phrases a-la Eliza[1] . In the article, the "boyfriend" responds to a specific question about Downton Abbey. Sure, in this instance the responder may be a fan of DA; but in the general case, it'll require more than 18 seconds (@200/hr) to just type up an intelligent, context-relevant response.

It turns out they watched Downton Abbey, in which case it's a 15 second response. If they hadn't watched Downton Abbey - also 15 second response.

And, from reading the script - it's apparently the case that the same CSA will get scheduled in for short periods of time with the same customer - able to maintain a thread, and presumably, all the CSAs have the thread available to respond.

> all the CSAs have the thread available to respond.

That's the catch: they have to read the full thread and then respond... in 18 seconds on average! That is a lot. I can barely read a decent-sized paragraph in 18 seconds; and the average typing speed of a professional is 50 - 80 wpm; which means even a 10-word response will take around 10 seconds to type. So you have 8 seconds left to grok the entire context and form a coherent response! And do this for hours at an end.

Yes - I concur that if there is a context switch - you'll run into some performance issues, good point. This suggests then that the system performs best when people are doing back-and-forth on a single thread.

Reply on news.ycombinator.com