The software company would be paying the support company. It would probably be a fairly predictable cost depending on the amount of calls/ tickets/ whatever metric is used.

These are good points. I think you are right that Series A/B are actually probably too small. This would have to be something that much larger companies use.

The biggest problem that I know of with support for large software companies is communication and access to experts. A lot of the largest SaaS companies put L1 overseas, and it can make escalations a nightmare and produce a really poor customer experience. However, this experience is somewhat rare, so it isn't always a priority for companies, but if this rare situation happens at the wrong time, it can really impact a customer's opinion of the product.

It seems like there would be a way to create a better outsourced support model that was not only focused on L1, but escalations as well.

"It's also much easier to debate future usability changes because you're more familiar with how real life people are using the product."

This is true of smaller teams, but as things get bigger doesn't this process become more dependent on data anyway, so having it in-house wouldn't matter as much?

It just seems like the mis-communication that happens with escalations can be monitized by having much better support people basically multi-table multiple companies' support at once.

One of the frustrations at my job is that after supporting the product for a few years is that support knew how the product worked better than product or engineering because we lived in the interfaces everyday while product and the devs where not, but I was rarely successful using this knowledge to prevent poor decisions from making it into the product. It's frustrating when you see a feature that you know is going to piss off customers during demos when it's too late to do anything about it.

Yeah definitely. You're right that at a certain size there's a communication problem regardless of if you're staying in house or outsourcing.

And you're also right that this seems like a somewhat unsolved problem.

This may sound crazy, but here is an idea I have been thinking of:

What if you could basically run support like a guy on 3 monitors 24 tabling 3 different games. PLO on one screen, NLHE on another, and Draw on the third. Table-ninja like responses based on the things that are most commonly asked and for things that need to be escalated beyond this guy (L1) there is a second tier of support who works on the next issue. This person would be a PLO expert, he only works on L2 escalations for PLO and at this tier (there are NLHE and Draw experts as well obv), maybe communication becomes a little bit slower, he will immediately respond to the customer and get on top of their issue, but then the next communication update may be 30min-2hr from then. If at that point he cannot figure it out, he then sends it to L3, who is the High Stakes PLO support person, he gets on the phone with the customer and and has serious communication and technical skills (Senior Engineer/ Architect type guy) and he handles the issue from here on out with a team of 2-3 people that he brings up to speed and they work on it internally. When he isn't working on issues, he is coaching L1 and L2 and updating their table ninja responses.

I am sure that the people overseas are multi-tasking to a point, but nothing like this. Go to comcast.net and talk to live support, they are so slow and seem to actually have somewhat personalized responses, I have tested it (inconspicuously) and they actually seem to answer things about how they are doing and the weather faster than the things you are trying to get done (pay my comcast bill). It also takes such a substantially longer amount of time than it should. It is like taking a 70 year old limit player and putting him on 24 tables of PLO.

I also say this having pretty intimate knowledge of the support structure for some of the largest software companies and in working with some of the current fastest growing software companies they have the exact same problem (probably because the same people are moving from company to company).

I find it hard to believe you couldn't pay people legit IT wages and still make it profitable by pumping volume. Yes, some people may get burnt out working this many issues at a time, but I know churn & burn companies making inside sales kids right out of college make 150-250 calls a day so it can't be too much different than that.

My first guess would be that at this point the people that are skilled/competent enough to pull off the multitasking required for what you're talking about are skilled/competent enough to get a much better (interns of money and interest) job elsewhere.

In my limited support experience I've also seen the L2 level disappearing. Basically you have the canned responses level and then the actual product engineer level. The nice thing about this is that it forces the engineers to fix annoying bugs that keep getting escalated to them.

So I have my own holdem evaluator which works on 5, 6, 7 card hands. I'm thinking about using it for omaha evaluation. The naive solution would just be to try all (4 choose 2) = 6 possible sets of 2 card hole hands against the board, and use the highest one as your hands rake. Can you do much better than that? Using the LUT method with omaha starting hands would create a much bigger LUT but would be faster...

Anyway curious to hear any thoughts from those of you who've explored this.

What's the evaluator doing exactly? Is it measuring your AI % against a random hand? Or something different?

measuring AI% vs a random hand is a higher level operaton -- you'd use the results of many evaluations for that.

i'm talking about the lower step, the one that simply maps a hand to a universal rank, which is how most holdem evaluators work. then if you have a hand and i have hand, we can run them each through the evaluator function, compare the two resulting numbers, and see who wins.

Ah ok! Mapping a hand to a universal rank should be quite a cheap operation, I don't think there's really any other way to do it apart from looping each of the 6 combos against every combo of 3 board cards and picking the best. The key is making the method to get the hand rank from any given 5 cards as cheap and memory light as possible.

I remember looking into prime numbers for this sort of thing. If you map all of the 52 cards to a prime number and multiply the primes of a 5 card hand together you get a unique index for that hand which is the same regardless of order (so you eliminate the need to sort the cards at any stage like some other methods do). This is very cheap to perform and can help a lot with working out universal rank. It's possible to create smallish LUT's on these values (I think it's also possible to perfect hash them as well).

The largest number possible value from a 5 card hand is 239*233*229*227*223 = 645,535,342,583 which will fit into a 64 bit int.

Edit: I definitely remember it's possible to convert a sum of 5 primes into it's index (for example there are 2,598,960 combos of hands, the above largest possible value would be turned into #2,598,960). With 7,462 distinct hands you need 2 bytes to store each ranking. This means you can have a LUT <5mb big and a method to get the hand ranking from any 5 cards really really fast which makes looping a few hundred combos on the board very quick to do.

well i've got the part converting the best 5 or 7 card hand to a rank done. i'm using a LUT which uses ~120M ram, which is very fast and perfectly acceptable for my needs. i was just curious about the omaho specific part. but looping the 6 combos should be fairly quick so maybe i'll just start with that.

Yeah the last line is a your mom joke

I'm not actually sure what groucho was working on for that. Probably something related to document rendering using pdf.js though

he is a veritable fount of hilarious chat dialogue.

This might be a silly beginner question, but I have no idea how to figure this out.
I have created a bioinformatics tool for my lab that would benefit most of us greatly, but is hard to install on everyone's pc due to database and library usage.

It is in python, and currently I just run it from my terminal. It asks for two separate lines of input and the. Outputs about 20 lines to my terminal and creates 3 text files.

What would be the easiest/best way for me to make this available to the other people in my lab?
Preferably without paying for a domain to host it on. And they need to be able to get those output files.
Is there a way for them to 'contact' my terminal through our intranet and doing it that way? Because I don't mind leaving my pc on all the time for that.

You could put up a simple web interface where the user puts in the 2 lines of input. Then print out the output. You could prob do it in Django pretty quickly and run the server on your machine 24/7.

Are the users all running linux? Library and database dependencies shouldn't be a big deal in that case.

is he a russian or ukrainian contractor?

nah, we're all 'merican.

I realize that may make us insensitive stereotyping bastards, but if so I apologize profusely. It's not intended that way at all. I think there is something about programming nerds and jokes based around the use of language though.

oh not offended at all, but surprised (and a little disappointed). i was hoping it was real. so he's doing a character? what is the character supposed to be?

I have no idea

If that wasn't real it smells of ebonics and isn't as funny...

it doesn't sound even vaguely like ebonics to me. it's supposed to be something foreign i'm pretty sure.

Sorry, that's what I meant, making fun of a group by mocking their language patterns.

I get the objection, although I confess I'm a little sad if I can't enjoy those kinds of jokes. Because they are funny to me.

Like this one

I don't think that is the same thing and I'm will to admit I might be over sensitive. I got the stink eye when I started here in support for suggesting that using a bad Indian accent to answer support calls might be just a little racist...

I don't mean to start any kind of argument about it really. Context matters a lot, and I could understand why it's a good idea to be more sensitive with regards to an indian accent in that context then an old faux-german sign. All I'm saying is that I don't believe the enjoyment I get out of those kinds of things in general is a result of any kind of racial animus, although I also admit it's certainly possible I'm a little insensitive and/or ignorant. I've just always found a lot of humor in language, whether it's monty python or Sasha Baron Cohen doing Ali G or Borat, or the german sign, or what I originally posted.

yeah it's totally context and delivery dependent. i don't find borat or apu offensive. but i could see a certain kind of person doing it a way that was.

Yes, if you setup a tmux session they can ssh to it as if it were any box. Once they attach to your session they will be able to execute code through your terminal while typing on their end. It's one of the lightest weight ways to do pair programming, but in this case you can use it as some type of remote control functionality.

Google around for "tmux session sharing" and similar phrases to find some guides.