The biggest issue about AI is that it is lacking the in depth knowledge and the ability to give on point feedback.
Take these two tweets for instance (That guy has a interesting blog about engineering). They show pretty much that AI isn’t the go to replacement for real support.
Can you give an example where there’s a support that is using AI that really gives you the answers you want? Including AI in the workflow yes. Replacing agents with ai is not really working.
Actually, I don’t want AI to completely replace customer support. I just want AI to reduce the workload, such as most of the simple questions: how do I SWAP tokens, etc.
Apologies, I should have been a little more clear, I think it’s obvious that AI shouldn’t just be jumped to and relied upon as an immediate full solution that would work flawlessly. It would absolutely fail and not work in this scenario. AI should be tinkered with and tested as a longer term solution that could help streamline support. Companies across the world have streamlined their support funnels without AI, AI is not the immediate solution for Osmosis.
What I do think is a guaranteed path to efficiency today is just smart replies / direction with prompts that funnel a user down solutions until potentially needing either a live agent or a real reply from human eyes.
This can be done with Telegram, a support widget, or email.
A few examples are
BEST BUY and their website support, this funnels users with solutions until finally they may require support from human eyes (images attached at the bottom). This works incredibly well as I buy all my electronics at Best Buy and ask for support (refunds / price matching usually) a few times a year. This even expands to beyond support, but basic information delivery on things like policies (which could also be done for Osmosis, such as details on spread factor, or taker fees, staking, etc)
KAST CARD (Solana Debit Card), they have a concierge service similar to the Best Buy support box. With Kast Card, you can use Telegram, Email, Whatsapp, and likely a few others. In the image attached (at the bottom) you can see the process where it first digests the message and makes a smart reply, when their reply is not sufficient a real person jumps in and takes it further. Sometimes the real person is available instantly, sometimes it ranges between 30 minutes to a few hours. (image attached at the bottom)
Best Buy services millions of customers, while Kast Card being a debit card will naturally have a large support base required. We should assume they do some things right and that we can take some lessons here. I think this is something that the OGP should explore funding.
The existing primary support telegram chat should cease to exist as a support chat, it’s prone to spam and scams, this is not where customers should be directed for support. It should of course still exist as a general place to chat.
This reply is just about the method of funneling / direction, I’ll share more broad thoughts in a separate reply.
The first thing I do when I notice I am talking to a bot is throwing as much questions in a short timeframe as possible… the bot gives up, asks me if I want to speak a real person and I gladly say I do.
I really hate support bots, since they never ever really give a decent answer on a question which goes slightly outside the standard.
I am nowadays also avoiding webshops were I know a bot is doing the primary communication. I do agree on exploring it and see if we can get to a workable solution, but be aware that it might also cause users to go elsewhere.
If the point is “users will go elsewhere because elsewhere has better service” well, that won’t be the case bc that doesn’t exist. (Edit* clarification, I mean support doesn’t exist elsewhere so using an all things equal argument doesn’t make sense here. Ie, reducing support from 100/25 down to 25/25 is still vastly more than any other ecosystem support that might be 5/25)
I’m not worried about that point. Plus, the goal is to do a good enough job with the UX that this point wouldn’t be an issue. If you aren’t confident that the OSL could do this, then the community should source someone that can. This doesn’t mean OSL can’t operate the hands on portion though.
The main thing to keep an eye on is not what we would do with infinite funding, as Osmosis used to fund even just Reddit and discord management with around $1mm a year alone.
The main thing is what is the logical balance, and that needs to be rediscovered bottom up not top down.
I favour L1 support becoming a chatbot, but L2 support should remain with humans and be easy to get to since I agree with others that being trapped in chatbot hell is incredibly offputting. I also don’t think AI is the solution here since it tends to hallucinate; a more typical support chatbot can read from most sources and give cookie-cutter responses that fulfil the majority of queries.
I don’t have access to the dedicated support channels, but most of the public ones from this weekend (5 in Telegram, 1 in Discord) were basic queries that just needed a link to existing docs or articles sent which feels like a misuse of resources to have tier 2/3 support responding to.
Since our main KPI for OSL is the number of tickets, OSL is incentivized to increase the number of tickets rather than provide resources to allow users to help themselves. The other metrics are also not great, the first response being superior with a relevant auto-reply and the median time to close is a tricky one since some tickets will inherently take days to resolve while others are simple requests that offset this by only taking a minute or so.
I’ve used Axelar’s support quite a few times in the last few months and found that to be a great experience, even though that is just an online query form with an auto-response and a human response within a day.
I’d love to see a couple of separate proposals here, one to build a L1 support system, and a more streamlined L2 system that perhaps recognizes Osmosis as a client of ISG with resources allocated based on need, not a separate organization under that umbrella.
This is a tricky one. Please refrain from feeling superior, because that is always followed by a downfall.
I do agree on being sharp on funding, in the end the protocol is not a bottomless pit or infinite pool of money.
That being said, I think @JohnnyWyles also has a point. KPI’s need to be undisputed, but also won’t leave room for abuse. I think the first question is to properly define what we want before any more proposals or threads are started, because they will all go down in the question of not knowing what requirements must be met. In development setting the requirements is all that matters to have a chance of a successful process following on it.
So if we say that a requirement would be to have simple queries answered automatically, then we have a starting point.
Time to response for more difficult questions, costs per ticket, user appreciation, etc can be nice following requirements imo.
Also to be fair, the Telegram group has not been a designated support area for more than a year now. However, the telegram tech group that the foundation created is, and it is one I am fairly active in:
My point isn’t a superiority thing at all. I believe I didn’t convey properly, what I mean is there is no other place with this level of customer support in DeFi.
When comparing all things equal, taking customer support down from 100/25 to 25/25, is still 5x more than a 5/25
This is a small but valuable channel that you are a great contributor in, and largely primarily only you from OSL right?
Technical channel should be separated out. Or should explore a grant with OGP or Foundation imo. Don’t think it makes sense to tie it to an OSL/ISG proposal
