A first hand experience of DHL’s extremely helpful Virtual Assistant. (Please ignore my shoddy spelling and grammer. Ta.)

  • STUPIDVIPGUY@lemmy.world
    link
    fedilink
    English
    arrow-up
    26
    arrow-down
    5
    ·
    edit-2
    6 months ago

    Not going to ignore your shitty spelling and grammar, because that is precisely what confuses the AI

    Not saying the chatbot is great or something but you need to use straightforward terms and correct grammar. It can sometimes see through spelling mistakes, but at least use the right form of “they’re”. It’s not hard

    Also using british terms like “not in”… Not in what?

    Say “I won’t be home”.

    • SrootusOP
      link
      fedilink
      English
      arrow-up
      13
      arrow-down
      4
      ·
      6 months ago

      Okay, but at that point they should have just pull in a human, surely they can’t expect every single customer to proof read their messages. This is the first point of contact they have, someone who isn’t confident with technology that doesnt understand the quirks of AI would get stuck.

      • STUPIDVIPGUY@lemmy.world
        link
        fedilink
        English
        arrow-up
        15
        ·
        edit-2
        6 months ago

        Yeah I agree I hate this kind of obstructive customer service. Just making it clear that you have to write coherently and simply. Even a human who knows English as a 2nd language might struggle to understand exactly what you were asking for, especially since i dont know how understandable the censored part is

        • sushibowl@feddit.nl
          link
          fedilink
          arrow-up
          11
          ·
          6 months ago

          Yeah I agree I hate this kind of obstructive customer service

          I work as a software engineer on automated customer service systems like these, and boy let me tell you, obstruction is the name of the game. For example: don’t make the phone number too easy to find on the website because it will lead to too many calls. We nudge people toward the FAQ and such first so they can hopefully find their answer there. Then, we have chatbots like this which contain exactly the same information as the FAQ again. And only then might we offer you contact with a human.

          The essential problem is that support is a cost center, so cost savings is the name of the game. We optimize for metrics like:

          • “deflection” (number of calls averted because we pushed the user into automated tools instead)
          • “first call resolution” (percentage of issues resolved in one contact. How do we know if your issue is resolved? Simple, if you don’t contact us again we assume the issue is fixed)
          • “Average contact time” (pretty obvious, get the customer off the phone ASAP)

          If you manage to get on text chat with a human, typically they are handling two other conversations at the same time, that’s why they seem so absent all the time (and why companies love chat. Much cheaper than calling).

          I’m not saying we’re all diabolical here. There is a general agreement among everyone in the industry that we should help the customer as well as we possibly can. Indeed every CS manager will tell you how important we are to our brand image and NPS, how we strive to be the most customer-friendly company etc. etc.

          But the numbers don’t lie. If you look at the metrics that everyone actually optimizes for, it’s cost cost cost.

          • SrootusOP
            link
            fedilink
            English
            arrow-up
            7
            ·
            6 months ago

            it has become cheaper to talk to a machine than it is to your own kind. Thats just sad

          • bjorney@lemmy.ca
            link
            fedilink
            arrow-up
            1
            ·
            6 months ago

            My company has two full time staff members running point on customer service. Directing people to the 5 bullet point FAQ page is about 85% of their job.

            I dabbled with setting up an AI classifier to do this for them, and while it would remove 99% of those FAQ emails, the remaining 1% of the time it was so catastrophically wrong it made us press pause on the whole thing.