5 ms·

“Customer support by AI” sounds even worse than the current practice of having customer support provided by utterly clueless and unhelpful “support” from barely

by sltkr 2y ago

“Customer support by AI” sounds even worse than the current practice of having customer support provided by utterly clueless and unhelpful “support” from barely-English-speaking-people from the lowest-wage-country-we-could find (stereotypically India). How did you come to hate people so deeply that you made it your life goal to make an already abysmal experience even worse?

dragonsky67 2y ago
When the obvious desire of 90% of customer support is to make the customer go away being able to use a box rather than a person to get rid of them seems a perfect solution. Support is a cost centre, the people calling support are not the ones who will be making purchasing decisions so why provide anything close to decent support. Vendor lock in, subscription services and other ways to reduce the chance that the customer will go elsewhere all contribute to the downward spiral in support. Truth is, if they can manage to provide proper feedback to the AI for when the support that is provided is actually useful or successful this may actually learn to be better than employing someone to read off a support flowchart that hasn't been updated in 20 years.
laborcontract 2y ago
I was on a small team of 10 that experienced very strong growth in our product. One of the consequences of success is that we had to take hundreds of calls per day. We eventually had to hire dedicated customer support people but after hours were an issue. There are large outsourced customer service companies but those cost $1 a minute and those people have to deal with CS calls from many different companies, so it took them forever to find scripts relevant to the customer’s request. Most calls failed - that outsourced CS was a glorified voicemail box. Our success rate with outsourced CS was very low, partly because we didn’t have the resources to fly over to Austin to train a workforce that didn’t show to be particularly promising. AI voice bots would have been able to helpfully answer and deflect 90% of our calls, and do so in a much faster and humane way. I understand the sentiment behind your comment, but I cannot agree with your assessment of the outcome.
- stackskipton 2y ago
  >AI voice bots would have reduced been able to helpfully answer and deflect 90% of our calls, and do so in a much faster and humane way. Would they have? I've worked at companies that had AI voice bots, they required a ton of care and feeding like a script. Only reason we used them is they were better at following the script and they didn't cause massive customer backlash compared to customer backlash we got when it was offshored.
  - laborcontract 2y ago
    So many of our calls were informational with a predictable/scriptable answer set. 90% may be high, but I know 80% of the calls would have been squashed by an AI bot. There's definitely a ton of care that you have to put into creating it aside from just retrieval, like control flow, but the state of our outsourced CS was really, really bad. 5 minute wait times just to get connected to someone that mispronounced our company name and mostly told people to call us in the morning. The costs were absurd for the quality of service we got and they ended up making customers angry.
    - tmpz22 2y ago
      > 90% may be high, but I know 80% of the calls would have been squashed by an AI bot. Which one is it 90% or 80%? Is it possible it could've been 70%? 60%? 20%?
hakimk 2y ago
I understand where you're coming from - poor support experiences are are the bane of my existence. Preventing poor support experiences is the exact reason we started Fini. With a long career as a software engineer in the support industry, I have seen first hand when and why these initiatives fail. Unfortunately, when growth-stage companies experience a surge of signups, they face the choice between either 1) delay responding to all support issues by weeks, due to lack of staffing, or 2) trying to automate/offshore support, leading to poor coverage and accuracy. Both of these options usually lead to horrendous support experiences such as the ones you mention. Thankfully, with recent advances such as LLMs, we are one step closer to bridging this gap. In our most successful projects we are able to reach 95% accuracy specifically thanks to keeping a close eye on human evaluations. That's why we are launching this package as well, which we believe brings us one step closer towards our mission of helping millions of end-users receive the high quality support they deserve within seconds rather than weeks. It's a challenge which very few (if any!) companies have pulled off well, which is why we have decided to put all our effort into making scalable and hassle-free support our first priority.
- ethanwillis 2y ago
  > In our most successful projects we are able to reach 95% accuracy Is this the 95% of issues that are common between customers that could be and are handled at scale with a FAQ/knowledgebase that is surfaced to a user automatically through prompts? How does it actually do on the smaller percentage of questions that are complex? And that 95% is on the most successful projects. But what does the distribution of accuracy look like across all projects? If 2 projects out of 1000 are 95% accuracy that doesn't mean much to me.
  - bee_rider 2y ago
    It seems like a very convoluted path, but if people are for some strange reason more apt to end up at the FAQ via “chatting” with a tool like this rather than just looking it up directly, this still seems useful, right? I mean what’s the alternative, is there a “more self-directed users as a services” offering out there?
  - 2y ago

by sltkr 2y ago

dragonsky67 2y ago

When the obvious desire of 90% of customer support is to make the customer go away being able to use a box rather than a person to get rid of them seems a perfect solution. Support is a cost centre, the people calling support are not the ones who will be making purchasing decisions so why provide anything close to decent support. Vendor lock in, subscription services and other ways to reduce the chance that the customer will go elsewhere all contribute to the downward spiral in support. Truth is, if they can manage to provide proper feedback to the AI for when the support that is provided is actually useful or successful this may actually learn to be better than employing someone to read off a support flowchart that hasn't been updated in 20 years.

laborcontract 2y ago

I was on a small team of 10 that experienced very strong growth in our product. One of the consequences of success is that we had to take hundreds of calls per day. We eventually had to hire dedicated customer support people but after hours were an issue. There are large outsourced customer service companies but those cost $1 a minute and those people have to deal with CS calls from many different companies, so it took them forever to find scripts relevant to the customer’s request. Most calls failed - that outsourced CS was a glorified voicemail box. Our success rate with outsourced CS was very low, partly because we didn’t have the resources to fly over to Austin to train a workforce that didn’t show to be particularly promising. AI voice bots would have been able to helpfully answer and deflect 90% of our calls, and do so in a much faster and humane way. I understand the sentiment behind your comment, but I cannot agree with your assessment of the outcome.

stackskipton 2y ago
>AI voice bots would have reduced been able to helpfully answer and deflect 90% of our calls, and do so in a much faster and humane way. Would they have? I've worked at companies that had AI voice bots, they required a ton of care and feeding like a script. Only reason we used them is they were better at following the script and they didn't cause massive customer backlash compared to customer backlash we got when it was offshored.
- laborcontract 2y ago
  So many of our calls were informational with a predictable/scriptable answer set. 90% may be high, but I know 80% of the calls would have been squashed by an AI bot. There's definitely a ton of care that you have to put into creating it aside from just retrieval, like control flow, but the state of our outsourced CS was really, really bad. 5 minute wait times just to get connected to someone that mispronounced our company name and mostly told people to call us in the morning. The costs were absurd for the quality of service we got and they ended up making customers angry.
  - tmpz22 2y ago
    > 90% may be high, but I know 80% of the calls would have been squashed by an AI bot. Which one is it 90% or 80%? Is it possible it could've been 70%? 60%? 20%?

hakimk 2y ago

I understand where you're coming from - poor support experiences are are the bane of my existence. Preventing poor support experiences is the exact reason we started Fini. With a long career as a software engineer in the support industry, I have seen first hand when and why these initiatives fail. Unfortunately, when growth-stage companies experience a surge of signups, they face the choice between either 1) delay responding to all support issues by weeks, due to lack of staffing, or 2) trying to automate/offshore support, leading to poor coverage and accuracy. Both of these options usually lead to horrendous support experiences such as the ones you mention. Thankfully, with recent advances such as LLMs, we are one step closer to bridging this gap. In our most successful projects we are able to reach 95% accuracy specifically thanks to keeping a close eye on human evaluations. That's why we are launching this package as well, which we believe brings us one step closer towards our mission of helping millions of end-users receive the high quality support they deserve within seconds rather than weeks. It's a challenge which very few (if any!) companies have pulled off well, which is why we have decided to put all our effort into making scalable and hassle-free support our first priority.

ethanwillis 2y ago
> In our most successful projects we are able to reach 95% accuracy Is this the 95% of issues that are common between customers that could be and are handled at scale with a FAQ/knowledgebase that is surfaced to a user automatically through prompts? How does it actually do on the smaller percentage of questions that are complex? And that 95% is on the most successful projects. But what does the distribution of accuracy look like across all projects? If 2 projects out of 1000 are 95% accuracy that doesn't mean much to me.
- bee_rider 2y ago
  It seems like a very convoluted path, but if people are for some strange reason more apt to end up at the FAQ via “chatting” with a tool like this rather than just looking it up directly, this still seems useful, right? I mean what’s the alternative, is there a “more self-directed users as a services” offering out there?
- 2y ago