<oops> I posted this Ask HN wrong - this body of text is supposed to be up there, but I don't want to delete and repost this because I'll lose the comments here.
</oops>
This is just an MVP; the back end does not scale well (this is so not intended for outside of just our small HN community here). I'm actually embarrassed to post this here, but if you're not embarrassed, you've waited too long to launch, right? So here's my first pitch for HN-peer-review.
Q: In a nutshell, what is it?
A: Auto-response, for Twitter. (Twitter for now, possibly Facebook later)
Q: Why the name?
A: It's from Altimeter's Jeremiah Owyang's 18 Use Cases of Social CRM report, I thought this app would help in the use case(s) he listed: http://bit.ly/csoAJC
Q: I don't want to even visit the site, just 'show me' the value prop.
A: Ok, tweet "@jaysern hi !!!"
Q: Anything else in it just for HN readers?
A: Aside from the use cases I listed on the site, here's one I did not list: I also pull tweets from public_timeline, so if you try match on hot trending word, your twitter account will tweet out to strangers ;) Here's my user page: http://www.rapidsocialresponse.com/user/jaysern/ (If none of this makes sense, don't worry - just ignore)
I was delaying posting this on HN because I wanted to make sure I got the Mixpanel analytics working right (since learning and getting feedback early is point!) Thanks much YC alumni & Mixpanel's cofounder @trefn for helping me with the tech support!
1- If 2 people are using your product, and one of them sends a twitter message to the other, is there an infinite loop because they automatically answer each other's messages?
2- The problem I see is how the engine parse the tweets. Is the user expected to hard-code messages, or is there a natural language processing algorithm that parses the messages in a smart way? It's extremly hard to think about each and every line someone might say - that's the main reason why expert systems fails in most cases. Examples like "hi" are extremly easy to program, and are not really usefull. It might help some people that got too many comments / tweets... but then again, some Twitter client that filter the messages received can do that to, and they can have SPAM filters too.
1. Nope, no infinite loops; it will not re-respond to something already responded to.
2. True and yes, there's some NLP user-friendliness baked in. It's case-insensitive, not whitespace sensitive, and it knows that "it's" == "it is". Give it a try?
My question would then be, what would be realistically considered useful to you? Obviously I'm no AI Ph.D claiming that it will pass the Turing test; but I just thought maybe something "good enough" will be helpful to at least a small subset of people. Saying "hello" is kinda pointless, but say if you got a lot of repeat questions over Twitter (because you're a business or something), this will help you auto respond to them.
Hmmm, this comment left me even more confused about why I'd want to use this. I definitely don't fit into the power user market but I still don't see the value add.
It'd make sense for an autoresponder for a FAQ or something. But the examples all seem to be imitating a human conversational response which sounds kind of risky for someone trying to maintain a reputation.
For my understanding, what part of my comment made you more confused? I'm kinda wondering if I should remove the "power user" thing from the main page now.
1) Web design is a lot like pants. Sure, it is annoying and you already have the important bits covered, but please put it on before going out of the door. It takes a few minutes to use a free HTML/CSS template: if you like the Web 2.0 aesthetic, styleshout.com has what you need.
2) Curiosity: Is the backend a person who is going to be composing tweets for me?
3) My worry, which this web site has to sell me on and that it is not doing currently: Is this going to spam my followers or make me look like an idiot to them? (Eliza answers my query: "What resemblance do you see?")
2) No humans involved, other than yours truly alone. You, the user, set what you want it to respond TO and what it will respond WITH. Yes, it's automated ;)
3) Interesting feedback. I actually have no idea how I'm going to monetize, but it has never crossed my mind to use my users to spam their users (that's like the ultimate cardinal sin!) No, I have to earn trust, not the other way around. But you bringing up this concern is a good thing for me to know. Truth is, I don't know if this will ever make a penny. I know this is cliche, but before I can figure out how to extract value, I thought I'd first understand (1) if there is any value, and (2) how much value it's creating for users
For what it is worth, while the design could use some sprucing up to make it more engaging, I don't find the current design troubling. It's simple and to the point. (But I didn't go beyond the first page yet, so maybe the spartan approach wears thin after a while.)
FYI, your survey form refers to this product as "Jaysern Bot", not "Rapid Social Response".
My $0.02
<disclaimer>
I don't use twitter (I have an account, but never tweet, almost never look at the people who I am following) - so maybe I am too old school to have anything interesting to say.
</disclaimer>
This strikes me as severely reducing the S/N of the social networking space - if this gets popular, which I assume is your goal, then a large percentage of tweets will be generated automatically. Now permutate this machine-generated "social" information with all the tools springing up to analyze tweets and derive all kinds of useful crowd-sourced insights. It all becomes that much more diluted with less and less information actually created by a live, thinking human being. This might be the start of a trend - tools which tweet automatically based on incoming tweets, creating that feared endless loop which you cannot detect because you only control one participant. Oops, some kids are out trampling on my lawn, gotta go.
def. a valid point, and my goal isn't to water down the S/N in S/N :) For what you said to happen, this thing would have to take off like wildfire (which it's definitely not at the moment :p). I think if we were ever in danger of that, the platform would probably reach out to me to say "hey!!", which at that point obviously I'll do something.
But the goal is not to create more mess, what's the quote? "Wealth of information creates poverty of attention". There's so much stuff to wade through out there, and I really don't want to contribute to the junk
3) Can the public see my response page? (/user/qeorge). I'd rather they couldn't, but I would definitely like to keep track of the messages I've sent.
4) Bit.ly analytics integration would be awesome, so I can see what messages are working. Maybe even stop using a response if no one is clicking its link?
Still not sure I would use this, because I don't do enough tweeting to automate it, and it still feels spammy.
1) Understandable; and I'm learning through feedback from the comments :) Let's see if I understand you. The response will be whatever you specify it to be.
2) I apologize for the confusion. RSR scans the tweets from 3 sources: from people you are following, tweets that have @mention'd you, and tweets from public_timeline; if something matches from those 3 sources, it will respond. Whether the person is a stranger or not. What I was explaining in [2], I now realize I didn't explain well.. is that the system does not explicitly search for keywords to respond to, outside of the 3 cases I mentioned above. Let's say 'blah' is a rare word. If you set a 'filter' (as you call it) to respond to a tweet containing 'blah', it will only respond only if that tweet was seen from those 3 places. In other words, if there's a person X who tweets "blah", but you're not following him, X does not @mention you, and X's tweet does not show in public_timeline, then there will not be an auto-response. (let me know if that is still confusing).
4) I didn't add a 'count', but I can do that from the back end so that users know exactly which entry auto-responded the most, and which didn't, and so forth.
I can certainly understand it comes off spammy, but would like to know the reasoning behind that, if you care to share more. It's definitely ripe for abuse .. which is something I will have to pay attention to.
If I set up a new "filter", I have to wait until someone triggers it to see what my response will look like. In my case, that could be days or weeks. So it would be great to preview this in some way - ideally, it would scan the timeline and show what I would have responded to, were the filter already engaged. Does that make any sense?
2) public_timeline is the same pool as search.twitter.com, right? I'm interested in Twitter's policy here, as I could see them disapproving of my sending messages to strangers in response to keywords. OTOH, I see people do this all the time, so maybe its a non-issue.
3) Thanks. That would be a great feature. I know my tweets are public anyway, but I'd like to avoid displaying that I am using an automated service.
4) A count would be great, maybe even rate limiting per response (e.g., don't send this message more than once per hour).
Regarding spamminess, I definitely agree that its all about usage. For instance, if someone mentions my brand, I don't think an auto-response is spammy at all. If I'm scanning for anyone mentioning DVDs so I can push my affiliate link, that's entirely different.
For my brand specifically, my concern would be that the responses were so generic that they would come across as spammy. Basically, I'm nervous about automating any part of my communications with my customers.
Note: even though this might not be a fit for my small business, simply because we don't have that much twitter volume, I think it would be really useful to many others. There is a definite need for this product.
1) that makes perfect sense. Basically like a dry-run test when you're crafting the filter. It'd take more dev time to implement that, I just wanted to get a bare bones out, but I'm taking notes :)
2) From what I understand, search.twitter.com, well they have their own special secret-sauce-laced algorithm; and outside looking in I don't know how they relate, but from what I've tested, it's different. public_timeline is a very small subset of the total tweets in the world. Basically, every 60 seconds they give you a snapshot of 20 tweets, randomly picked from the Twitter universe. With search.twitter.com, you're searching for something specific (public timeline is 100% random)
3) Good to know, noted.
4) I thought about a rate limiter, but to need a rate limiter, you've either (a) entered a filter that is overly generic (since it's matching over aggressively) -or- (b) you genuinely are seeing that many tweets with a similar pattern that matches the filter.
In the case of (a), you'd have to make your filter less aggressive. In the case of (b), then you must be a "power" power user. A made-up example would be if you were @starbucks and hundreds of Starbucks fans would each day ask you "did you change the frappucino light formula?" http://twitter.com/ShesShoppingNow/status/12914278780 - which if that was indeed the case, that is the perfect use case ;) You'd be auto-answering a question that gets repeated to you over and over and over ...
In social media, it's all about honesty/transparency/genuine-"ness", so yeah .. definitely craft the response to be as genuine as possible (its your choice to make it not generic sounding), but then yeah, if people see you repeat the same exact words over and over, they might suspect you automated it. I guess then I will throw in a "randomizer" to include a slight variation in the response ;)
And on the last point, I was thinking the same .. the percentage of people who might find this useful is probably not the bulk of Tweet-izens :/ But it was a fun exercise for me, and I thought I'd put it out there for the world and see if anyone cares!
Good to know, because I kind of thought that it'd be useful for people who do exactly just that (customer service via Twitter), so this is some validation for me :)
To answer to your question:
* It will scan the tweets of people you are following too, as well as anyone @mention'ing you - respond if there's matches (so it doesn't have to be directed at you)
* Specifically, you said a search stream. I actually had that months ago, then much to my disappointment, after lots of sweat to get it working reliably, I found out if you're auto-responding based on a search, Twitter will not be happy: http://help.twitter.com/forums/26810/entries/76915 so I had to throw out that part ;(
I think that seems fairly clear cut on their part.
I wonder if they would be open to dialogue about the issue though. For example, if someone is tweeting a specific question about a branding finding that with search and answering seems legitimate and helpful.
Obviously automated search stream based marketing sucks.
Twitter power user here and.. I signed up. Nice and simple. I didn't set up any responses but if it works as described, excellent. Nothing confused me, even though it lacks the "gloss" of a finished product.
All that out of the way though, I didn't set up any responses because I can't figure out any use for it. I looked through my recent @replies and there's no pattern to any of it. I just couldn't figure out any interesting way to use this (beyond mere testing).. sorry :-)
No need for apologies, and I thank you for the feedback :) From the start I knew that this wouldn't be useful to everybody. Truth be told, even for myself, it's not like there's many 'patterns' for me to auto-respond on.
On that note, I have tried a few things that worked and were entertaining for me:
(1) I had it respond to "good morning *", and whenever I woke up in the morning, the first thing I'd do is check my @mentions to see what stranger in the world responded to me, like this one here: http://twitter.com/NinaCinspires/status/10222932469 - I even had a "good night" one too.
(2) the public_timeline hits a few times if you try to match on a top trending keyword, example: http://twitter.com/jaysern/status/12618345889 (#itsbetterwhen was a top hash tag a few days ago)
I signed in expecting a demo. It looks like it activates the service straight away when using the sign in button. I couldn't see any information or way to deactivate it, so I revoked the oAuth access from twitter.
I wanted to reduce the signup friction as much as possible, but a few assumptions of mine that perhaps you debunked:
(1) I thought the front page and the image of the 2 tweets was the demo (at least it was enough to describe the benefits of this service/app to the prospective user). If you try to add an entry, there are examples to guide you. When you say a demo, do you mean a YouTube clip showing how it works? Specifically, did you want to know (a) what this does and/or (b) how to use it ?
(2) Technically, if you click 'sign in' from the main page for the first time, you're then showed a 'sign up' page (which if you leave the site at that point, your account is not created. But then again, it's what the user thinks happened that matters :)
This is just an MVP; the back end does not scale well (this is so not intended for outside of just our small HN community here). I'm actually embarrassed to post this here, but if you're not embarrassed, you've waited too long to launch, right? So here's my first pitch for HN-peer-review.
Q: In a nutshell, what is it?
A: Auto-response, for Twitter. (Twitter for now, possibly Facebook later)
Q: Why the name?
A: It's from Altimeter's Jeremiah Owyang's 18 Use Cases of Social CRM report, I thought this app would help in the use case(s) he listed: http://bit.ly/csoAJC
Q: I don't want to even visit the site, just 'show me' the value prop.
A: Ok, tweet "@jaysern hi !!!"
Q: Anything else in it just for HN readers?
A: Aside from the use cases I listed on the site, here's one I did not list: I also pull tweets from public_timeline, so if you try match on hot trending word, your twitter account will tweet out to strangers ;) Here's my user page: http://www.rapidsocialresponse.com/user/jaysern/ (If none of this makes sense, don't worry - just ignore)
I was delaying posting this on HN because I wanted to make sure I got the Mixpanel analytics working right (since learning and getting feedback early is point!) Thanks much YC alumni & Mixpanel's cofounder @trefn for helping me with the tech support!