shame is expensive
humans are the reason people clam up
👋 Hey, I’m Suhas, and welcome to another edition of TPF Weekly!
Trying something a little different this week. Less framework, more of a story and an idea that changed how I think about Voice AI.
If you like this format, tell me at the end and I’ll write more like it.
Quick poll:
You’re 3 months behind on an EMI, your phone rings from an unknown number, and you already know who it is and what they want.
Your stomach drops, so you let it ring out and tell yourself you’ll deal with it later, knowing you won’t.
Now flip it around.
Somewhere there’s a person whose whole job is to make that call to you, and they’ve made 40 of them today.
Half don’t pick up, and the ones who do are either ashamed or angry.
They dread the call as much as you do.
That call, the one both sides hate, is a major area of focus for voice AI startups right now, but hardly anyone is talking about it in this context.
Everyone's selling the wrong use case
Sit through any voice AI pitch lately and you’ve heard the same promises along the lines of deflecting support tickets / cutting cost per call / shrinking the contact centre.
Cheaper customer service, basically.
The reason is that the friendly support call in context was never where voice AI had its edge.
The edge is in the call nobody wants to make, the overdue payment, the “did you take your meds” check-in for the 40th time, or the renewal everyone’s avoiding.
Humans are bad at these calls because we have feelings about them, and that’s a problem a machine doesn’t have.
People tell the machine more
Researchers at USC ran a study where people talked to a virtual character named Ellie.
One group was told Ellie was fully automated, the other was told a real person was controlling her in the background.
It was the same character asking the same questions. The only thing different was what people believed about who was listening.
The takeaway is tough but obvious once you see it: we perform for other people, we tidy up the truth so we don’t look bad, and a machine gives us no one to perform for.
A second study with soldiers found the same thing, where they admitted to more PTSD symptoms talking to a virtual interviewer than they did to a human interviewer.
The whole industry repeats “people want to talk to a human,” and for a big slice of conversations that’s just wrong.
The human is the reason people clam up.
What happens when you take the feelings out
Once you see it this way, the companies winning start to make sense.
Take collections, the most miserable, highest-burnout job in finance.
The borrower on the other end often does better with the bot, which sounds counterintuitive until you’ve been the person 3 months behind.
Nobody sighs at you, nobody sounds let down, and you can sort out a payment plan without a stranger hearing the embarrassment in your voice.
The lack of judgment isn’t a worse experience here. It’s the whole point.
You see it in plain support too.
Healthcare runs the same play on the follow-ups nurses are too stretched to keep doing well, like medication reminders and post-discharge check-ins.
Independent studies show AI reminders lifting medication adherence anywhere from 7 to 33%.
The bot isn’t warmer than a good nurse. It just shows up every time.
Why this is big in India
In the US, a lot of these conversations have moved off the phone.
People text, email, tap a link, and pick up the call only when they have to. The dreaded call still exists, but the surface area is shrinking.
India runs the other way.
The phone is still the main channel, the volumes are enormous, and a huge share of the country follows a reminder far better in Hindi / Tamil / Marathi than in English.
On a collections or renewal call, that isn’t a nice-to-have. If someone can’t follow the call, they can’t really agree to anything on it.
That combination of massive call volume + dozens of languages, turned India into one of the largest voice AI markets anywhere in a single year.
Meesho already runs more than 60,000 support calls a day through it, in Hindi and English.
And that’s barely the start.
EMI reminders, insurance renewals, delivery confirmations, KYC follow-ups, win-back calls for lapsed customers, almost none of it has been automated yet, and most of it needs to happen in a language the caller is comfortable in.
The calls people hate making and hate receiving are the ones sitting untouched, at a scale no other market comes close to.
There’s a catch, and it’s where product work is needed: the voice cannot sound like a robot.
The second it does, the embarrassment comes rushing back and the person hangs up. The whole thing only works if the voice is good enough that your guard stays down.
It is harder than it looks, and ElevenLabs is one of the players doing this well.
If you're building anything in this space, it's worth hearing what their agents sound like before you decide what 'good enough' means.
The line I'd draw
None of this means fire the humans, and that's the trap Klarna fell into.
So here’s the bottom line.
Keep your best people on the calls that need one or many of trust / genuine apology / grief / or a hard judgment call.
Put the machine on the calls where shame makes people lie, where doing it 200 times burns a person out, or where the awkwardness makes everyone flinch.
The question I’d carry into your next roadmap meeting is “are we forcing a human to do this call badly, just because they’re human.”
That’s where the good ideas are hiding.
Latest in Tech
Our friends at Nexla just launched MCP Studio
If you work anywhere near agents, this is worth a look.
Nexla’s new MCP Studio lets teams build governed, task-specific MCP servers by describing the outcome they want, with no integration code.
It works out the data, picks the tools, and hands back a production-ready server with access controls and audit logging built in, and it plugs into Claude, ChatGPT, Gemini, and Copilot.
I had a play with it and the setup is slick! Early access is open now if you want to try it.
Upcoming Events
AI Vibe Sprint
Jakarta | June 13
Register Here
PM’s Dilemma: Build Fast or Build Right?
20 June | Hyderabad
Register Here
Exclusive Jobs of the Week
Senior Product Manager
Lead Product Manager
Director Product
These and other roles open across top companies like Meesho, Google, Zomato & many more.
Reply with one phrase your customers use for your product that you'd never use yourself.
I read every one.
Suhas 👋🏻
P.S. This gets better when the right people are in the room. Share it with one.