Everyone told us to keep our support "human." So we cloned our best agent's voice with Resemble AI and ran it for 90 days on tier-1 ticket responses. Guess what? CSAT went up 12%.
Key findings:
* Customers didn't notice. At all. We told a sample group; half called us liars.
* Handle time dropped. The bot doesn't need coffee breaks.
* The uncanny valley? More like a gentle speed bump. Minor robotic inflection on complex words, but customers rated it as "clearer" than some of our live agents. Go figure.
Downsides were real:
* Setup wasn't "upload and go." Tuning the emotional tone for apology vs. instruction took a solid week.
* Cost per call is fine, but our Terraform bill for the extra Lambda + S3 pipeline to feed it? Not trivial.
* You're now married to their platform. Good luck getting that voice out cleanly.
So much for the "authentic human touch" dogma. Sometimes a consistent, tireless machine is exactly what customers want. Fight me.
12% CSAT bump is nice until you factor in the lock-in cost. Wait until their API changes pricing or your "best agent" quits and you need to re-clone. That's not a pipeline, it's a shackle.
Your Terraform bill is the canary in the coal mine. The real cost isn't per call, it's the complexity sprawl you just adopted.
Customers calling you liars when told is the most interesting bit. Means they don't actually value "human" in that slot, they value resolution. Should've just used text.
-- old school