4 ms·

There has to be something between the best customer service and things like letting shitty "AI" ban accounts without any human intervention. How hard can it be

There has to be something between the best customer service and things like letting shitty "AI" ban accounts without any human intervention. How hard can it be to augment crowd sourcing and "AI" with some actual humans? If the "AI" is so crappy that it can't figure out an established, popular account is worthy of human support, what problems can it actually solve? I also wonder why we can't just pay $X to get a human, but I know the answer. Tech companies would just take the money as extra profit and still provide the shittiest quality support possible.

NoPicklez 6y ago
It's easy to say "there has to be something between the best customer service and things like letting shitting "AI" ban accounts" as an armchair hackernews user. But maybe, there isn't, currently. Maybe there simply isn't a solution at the scale of companies like Google.
- Dylan16807 6y ago
  'scale' is often used as a cop-out. If you double your customer count, and double your revenue, then you can double your customer support resources. O(n) does not become impractical just because the number gets really big. If they can't deal with something, the fundamental design is wrong, and would fail even if they downsized a thousandfold.
  - yuliyp 6y ago
    These things are not linear. As your scale grows, the amount of attention from bad actors you receive goes up (from what could basically be 0 for something small). Coordination also becomes more difficult, as does managing the fleet of customer services resources. If you have 1 person doing support, they can deal with the weird cases themselves. If you have 2 people, they need to share information between themselves. If you have 10 people, they might need to set up some sort of escalation system. If you have 100 people, you'll probably set up some sort of routing system (multiple languages/specialization). If you have 500 people, you probably need some sort of internal abuse / anti-bribery effort to catch bad apples. If you have 1000 people you've now got a separate team working on setting up policies. If you have 5000 people you've got a training and evaluation system set up for your customer service staff. If you have 10000, you've got customer support staff in multiple countries around the globe and need to deal with conflicting and ever-changing laws in the countries you do business in. All of this while marginal revenue per customer declines as your scale goes up. So yes, in addition to it being costlier per customer to provide the same level of support as your customer base grows, you have a smaller budget per customer to do it in.
    - close04 6y ago
      A lot of things get exponentially more complicated as the company or customer base scales up. You'll notice that the ones that benefit the company and the bottom line somehow manage to keep up. I take this as a sign that any such challenge can be successfully tackled providing you have a genuine interest in that. Things are as they are not because they are technically unsolvable but because this is the degree of interest each one posed for the company.
    - Dylan16807 6y ago
      > All of this while marginal revenue per customer declines as your scale goes up. I don't know about that. Just a few bad actors going after a small site can be an utterly ridiculous percentage when compared to the abuse a site like youtube gets. > If you have one person [...] If you have 10000 All of that sounds sub-linear to me. If 10000 support staff require 1000 direct managers and 200 people working on policies and training and 100 people working on development and 100 upper managers, that's still O(n). > All of this while marginal revenue per customer declines as your scale goes up. If some of your customers are higher revenue, and some are lower revenue, I'm not sure if that's really a scaling problem. But even with that consideration, it's not an excuse for being unable to give proper support to paying google customers, or youtube accounts with 100k subscribers.
      - yuliyp 6y ago
        > All of that sounds sub-linear to me. All of the things I mentioned are things that you can get away without but eventually scale forces you to tackle. Each of those things brings with it additional cost without increasing the throughput. This shouldn't be too novel an idea: consider a normal supply curve: cost per unit tends to go up with volume.
        Dylan16807 6y ago
        > All of the things I mentioned are things that you can get away without but eventually scale forces you to tackle. Each of those things brings with it additional cost without increasing the throughput. That doesn't mean it costs more than O(n). Let's say you start off with 100 workers providing 100 units of support, and 20 workers of overhead. Let's say doubling support means you need double the existing workers, plus double the existing overhead, plus 20 more workers to handle new kinds of overhead. So 120 x 2 + 20 = 260 workers for 200 units of support. Then 260 x 2 + 20 = 540 workers for 400 units of support. The percentage of overhead grows every single time we double the number of workers. And yet if you do the math, you see that the overhead never goes over 40%. And it was already 20% when we started, so that extra amount isn't hard to afford at all. Despite growing every stage, it's still O(n). The dominating factor is linear. Obviously those numbers are just an example. But the point is that you have to look at how much the costs increase. Even if it's theoretically faster than linear, if the overhead only grows a moderate amount as you grow from a thousand support workers to a billion support workers then it's not an excuse. It's within a small fudge factor of linear for all real-world numbers. > This shouldn't be too novel an idea: consider a normal supply curve: cost per unit tends to go up with volume. Only in the short term. In the long term, production ramps up until the price is pretty close, and on top of that economies of scale drop the price by a significant amount.
  - mattlutze 6y ago
    At some point you start getting lots of customers who purchase less but are more likely to require significant customer support resources for what they do purchase. At scale, things that might have looked O(n) at one volume of customers can change. Even if you could double your customer support resources, it means a very different structure and needing to ensure you both hire in the same quality of employees up and down the escalation chain as well as adapt your organization to handle the vastly larger communication and problem resolution processes. Noe of that is trivial and it's quite easy to take a great org and make it terrible by just adding a bunch of people.
- vorpalhex 6y ago
  Then your company doesn't get to scale. Sorry, but you can't skip support because it's not cost effective, just like you can't skip financial regulations or data regs. You don't get to use the excuse of "Sorry we are too busy making billions, maybe we wouldn't have to cut corners if we weren't so successful."
  - AnthonyMouse 6y ago
    The problem isn't scale. Scale is just how they made a lot of money. If you make a dollar a billion times you made a billion dollars. The problem is that good customer support costs more than $1. It costs more than what they make from the average user, so the average user can't have it. That would still be true if they had 100 times less users. It's a result of the service being free, not the service being large.
    - phreack 6y ago
      Do you think charging users for support could be a feasible option? I'd pay $100 for a call with a representative who can call shots on Youtube's strike reviews.
      - AnthonyMouse 6y ago
        It would solve the cost problem, but people would hate it. You'd have to pay even if you won because they have to pay the customer service rep either way. And it would open up a new trolling vector. Now the troll files complaint after complaint and forces you to pay $100 over and over again. What could work is to do loser pays. If someone wants to object to your video, they have to post $100, and so do you if you want to get it reinstated. Then whoever wins the dispute gets their money back. And extra points for having a builtin troll deterrent.
    - rndgermandude 6y ago
      The vast majority of Google users will never need customer support, and those who do probably will mostly have easy problems. There aren't a billion content creators on Youtube, and even fewer who would be good targets for blackmail like this. The channel in question has 100K+ subscribers and 17M+ views. That's probably a sizable amount of profit for Google from ads. It it really so unfathomable that google could tune their ban bot from "ban this video/channel/account immediately" to "signal a human to have a look and decide on what to do, maybe after talking to the channel creator"? That's just ridiculous to me.

NoPicklez 6y ago

It's easy to say "there has to be something between the best customer service and things like letting shitting "AI" ban accounts" as an armchair hackernews user. But maybe, there isn't, currently. Maybe there simply isn't a solution at the scale of companies like Google.

Dylan16807 6y ago
'scale' is often used as a cop-out. If you double your customer count, and double your revenue, then you can double your customer support resources. O(n) does not become impractical just because the number gets really big. If they can't deal with something, the fundamental design is wrong, and would fail even if they downsized a thousandfold.
- yuliyp 6y ago
  These things are not linear. As your scale grows, the amount of attention from bad actors you receive goes up (from what could basically be 0 for something small). Coordination also becomes more difficult, as does managing the fleet of customer services resources. If you have 1 person doing support, they can deal with the weird cases themselves. If you have 2 people, they need to share information between themselves. If you have 10 people, they might need to set up some sort of escalation system. If you have 100 people, you'll probably set up some sort of routing system (multiple languages/specialization). If you have 500 people, you probably need some sort of internal abuse / anti-bribery effort to catch bad apples. If you have 1000 people you've now got a separate team working on setting up policies. If you have 5000 people you've got a training and evaluation system set up for your customer service staff. If you have 10000, you've got customer support staff in multiple countries around the globe and need to deal with conflicting and ever-changing laws in the countries you do business in. All of this while marginal revenue per customer declines as your scale goes up. So yes, in addition to it being costlier per customer to provide the same level of support as your customer base grows, you have a smaller budget per customer to do it in.
  - close04 6y ago
    A lot of things get exponentially more complicated as the company or customer base scales up. You'll notice that the ones that benefit the company and the bottom line somehow manage to keep up. I take this as a sign that any such challenge can be successfully tackled providing you have a genuine interest in that. Things are as they are not because they are technically unsolvable but because this is the degree of interest each one posed for the company.
  - Dylan16807 6y ago
    > All of this while marginal revenue per customer declines as your scale goes up. I don't know about that. Just a few bad actors going after a small site can be an utterly ridiculous percentage when compared to the abuse a site like youtube gets. > If you have one person [...] If you have 10000 All of that sounds sub-linear to me. If 10000 support staff require 1000 direct managers and 200 people working on policies and training and 100 people working on development and 100 upper managers, that's still O(n). > All of this while marginal revenue per customer declines as your scale goes up. If some of your customers are higher revenue, and some are lower revenue, I'm not sure if that's really a scaling problem. But even with that consideration, it's not an excuse for being unable to give proper support to paying google customers, or youtube accounts with 100k subscribers.
    - yuliyp 6y ago
      > All of that sounds sub-linear to me. All of the things I mentioned are things that you can get away without but eventually scale forces you to tackle. Each of those things brings with it additional cost without increasing the throughput. This shouldn't be too novel an idea: consider a normal supply curve: cost per unit tends to go up with volume.
      - Dylan16807 6y ago
        > All of the things I mentioned are things that you can get away without but eventually scale forces you to tackle. Each of those things brings with it additional cost without increasing the throughput. That doesn't mean it costs more than O(n). Let's say you start off with 100 workers providing 100 units of support, and 20 workers of overhead. Let's say doubling support means you need double the existing workers, plus double the existing overhead, plus 20 more workers to handle new kinds of overhead. So 120 x 2 + 20 = 260 workers for 200 units of support. Then 260 x 2 + 20 = 540 workers for 400 units of support. The percentage of overhead grows every single time we double the number of workers. And yet if you do the math, you see that the overhead never goes over 40%. And it was already 20% when we started, so that extra amount isn't hard to afford at all. Despite growing every stage, it's still O(n). The dominating factor is linear. Obviously those numbers are just an example. But the point is that you have to look at how much the costs increase. Even if it's theoretically faster than linear, if the overhead only grows a moderate amount as you grow from a thousand support workers to a billion support workers then it's not an excuse. It's within a small fudge factor of linear for all real-world numbers. > This shouldn't be too novel an idea: consider a normal supply curve: cost per unit tends to go up with volume. Only in the short term. In the long term, production ramps up until the price is pretty close, and on top of that economies of scale drop the price by a significant amount.
- mattlutze 6y ago
  At some point you start getting lots of customers who purchase less but are more likely to require significant customer support resources for what they do purchase. At scale, things that might have looked O(n) at one volume of customers can change. Even if you could double your customer support resources, it means a very different structure and needing to ensure you both hire in the same quality of employees up and down the escalation chain as well as adapt your organization to handle the vastly larger communication and problem resolution processes. Noe of that is trivial and it's quite easy to take a great org and make it terrible by just adding a bunch of people.
vorpalhex 6y ago
Then your company doesn't get to scale. Sorry, but you can't skip support because it's not cost effective, just like you can't skip financial regulations or data regs. You don't get to use the excuse of "Sorry we are too busy making billions, maybe we wouldn't have to cut corners if we weren't so successful."
- AnthonyMouse 6y ago
  The problem isn't scale. Scale is just how they made a lot of money. If you make a dollar a billion times you made a billion dollars. The problem is that good customer support costs more than $1. It costs more than what they make from the average user, so the average user can't have it. That would still be true if they had 100 times less users. It's a result of the service being free, not the service being large.
  - phreack 6y ago
    Do you think charging users for support could be a feasible option? I'd pay $100 for a call with a representative who can call shots on Youtube's strike reviews.
    - AnthonyMouse 6y ago
      It would solve the cost problem, but people would hate it. You'd have to pay even if you won because they have to pay the customer service rep either way. And it would open up a new trolling vector. Now the troll files complaint after complaint and forces you to pay $100 over and over again. What could work is to do loser pays. If someone wants to object to your video, they have to post $100, and so do you if you want to get it reinstated. Then whoever wins the dispute gets their money back. And extra points for having a builtin troll deterrent.
  - rndgermandude 6y ago
    The vast majority of Google users will never need customer support, and those who do probably will mostly have easy problems. There aren't a billion content creators on Youtube, and even fewer who would be good targets for blackmail like this. The channel in question has 100K+ subscribers and 17M+ views. That's probably a sizable amount of profit for Google from ads. It it really so unfathomable that google could tune their ban bot from "ban this video/channel/account immediately" to "signal a human to have a look and decide on what to do, maybe after talking to the channel creator"? That's just ridiculous to me.