I mean there may be some way to do this with IRC where you don't have to piss around with ident and nicks and nets and netsplits and all this other ridiculous stuff, but the proof is in the pudding.
IRC is great, for geeks.
IRC is a very transparent protocol, there are A LOT of chat applications that you see and use every day that are actually implemented using an underlying IRC protocol or variation of it and provide an IRC interface (an example would be Twitch chat).
Furthermore, I've seen IRC being used in a lot of companies internally by most engineers and not.
I would definitely say that IRC has NOT failed, let alone colossally.
EDIT: To reply to everybody and make it easier, I fully understand that IRC isn't universally used (that is the point of the original post) and that is in a steep decline, I'm just saying that claiming it 'failed' (both as front and back end technology) is simply not true. It's old and not as used anymore, that's it. It's like saying the Nokia 3310 phone has failed colossally because not many people use it anymore.
Whatever the pros and cons of the technology, or the whys and wherefores of the reason for this lack of widespread use, it clearly has not been adopted by the mainstream. If the goal of a protocol is to be used widely, then clearly IRC has failed.
Herein lies the difference between e-mail and IRC - e-mail is fundamentally the same as an underlying technology (we still use IMAP!), but as an end-user product has changed a huge amount over the last twenty years. Things like webmail have progressed a huge amount, and systems like Outlook/Exchange have been built that allow it to be managed in a business centric manner.
With modern e-mail, you sign up for an e-mail, and that's it. You can now go to a website and e-mail people.
With IRC as a product, (aside from the ugly webchats that still expose huge streams of text and NickServ and other things), the end-user experience is still fundamentally the same.
As someone who has spent a non-trivial amount of time in IRC development and as an operator on small and medium sized networks, I've noticed that IRC (especially when services are involved) is pretty counter-intuitive for many users, even otherwise brilliant ones. Modes are obtuse and vary wildly from server to server, the idea of a separate nickname and channel registrations seems unnecessarily complex, channels just disappear when everybody leaves them, and so on. On top of this, the IRC architecture is needlessly limited for the most common IRC network use cases these days. In the past where networks were a federation of autonomous servers, it made sense, but these days a network's servers are partly or entirely controlled by a single entity, and things like netsplits are an unnecessary interruption in user service.
There is a working group to define the next generation of the IRC standard, but the additions tend to be patches for small nigglings in the protocol (such as needing to consistently poll for user away statuses and account names) than aggressive overhauls to the whole architecture, which is okay; there's only so much you can do with an effort like that while still seeing progress, and these are things IRC has needed for a long time anyway.
In my opinion what is needed if IRC is to regain its popularity is an innovative new server implementation that challenges the very workings of IRC from the ground up, maintaining only enough of the protocol to interact with existing clients. If users are to feel comfortable with IRC, they need a server software suite that conforms to their expectations of how a chat service should behave, and no solution currently provides that. The problem is, there is an enormous engineering cost associated with this, and the IRC development community is having a hard enough time as it is finding people who are willing to do more than just basic maintenance tasks.
I have a feeling that if DNS had this capability, and "IRC addresses" became standardised in the same way as emails (e.g. nick!irc.example.com), allowing anyone to use it with the same universality as email, the story might be very different today.
Auto-backfill is really nice. And being able to send rich content is something we kind of take for granted these days.
(disclaimer: I work on Matrix.org)