For some time now, I’ve held strong that AI tools in customer service have made it worse for the customer. The view might not have a lot of currency, but has its merits.
Take the parallel of connectivity; last mile connectivity makes or breaks at the micro, at the level of the individual.
Similar is the case of support services, and its last mile, the human. LLMs for all their stochastic prowess elsewhere, do fail when they need to apply a mind. Alas! is that because they truly lack one at this point?
But, a few calls with customer service teams in the last few months have made me rethink this problem a little better. This time around, not as a customer, but someone who thinks of value chains.
I had a service disruption on my fiber since early evening; something I let slide for a couple of hours thanks to how they have been for the last few months. But, when I got notified in the evening that it is an area level disruption, I only had to blame the choices and thoughts of sunk cost into this provider.
A four-hour delay graduates to 8 and a day by the end of a call to resolve. What makes it even more interesting is a notification saying the service is restored; just that it never was, at least for me.
A follow up call to customer service culminated in parroting of responses, gaslighting any understanding (or supposed lack) I have of networks/routers, and a one hour call at that.
Truth be told, I dread customer care calls for this very reason. It is not because I don’t get resolutions but being told by someone less technical/operational that I don’t know either/both.
Zooming out and back into my recent change of thoughts, if this is not very different from the blind responses that an LLM spits out, why would we need a human in the loop. After all, the bar is set low with the AI talking. That a customer would be frustrated about AI is a different concern.
On my rethinking and change of thoughts, I see that investment with off the shelf options does plug revenue leaks fast. Add to that, the cost of running that nearly plateaus. At any rate, can scale easier on the ledgers. A back of the napkin calculation that I did earlier puts it at anywhere between 4-8x less in running costs, and much less in terms of the friction of humans who supposedly manage.
Which gets me to the next and the crescendo of this thought stream of mine. There’s something that comes across very vivid and consistent in all these interactions that left nothing but a bad taste in the mouth. By and large, the teams are both ill-equipped and stripped of agency when it comes to dealing with.
And that is classic poor middle management that has not earned its keep. By making the word escalation taboo, they’ve in effect made themselves the antichrist of the one job they were to strategise for — resolve customer issues. Not to mention, their general status quo about what their teams operate with and can improve the outcome of the one job they have.
Yes, that’s the most simplistic way to look at it no doubt. And it became all the more evident when I ran it up the flagpole as it were; this time as a devil’s advocate. Organisational budgets, Cost centres, risk avoidance as a moat — things that a “pull people and fix the fire” brain has found hard to understand. But now, the most amusing to learn further.
Customer care doesn’t collapse at the last mile because of AI or humans. It collapses upstream — in silos, in risk‑averse playbooks, in middle management that enforces them. Until those choke points shift, the last mile will keep breaking.