I'll start with one of my own. I ran this through Roast My Plan, the tool I built to pick apart business ideas before anyone spends money on them. It did not go well for me, which is exactly the point.
Why it sounds good
Voice AI and behavioural economics sound incredible on a pitch deck or a Twitter thread. It ticks every modern startup buzzword box. The problem it points at is real too: people overspend when they know they shouldn't.
1The wrong intervention point
Most impulse purchases happen in physical shops, in noisy checkout queues, or on Amazon at 11pm with the checkout page already open. Nobody is going to stand at a till and talk out loud to their phone to argue themselves out of buying something.
The friction you're adding, opening an app and speaking to it, is higher than the friction of just buying the thing.
2Voluntary versus involuntary
People who want to pause already have ways to do it. They use the 24-hour rule, or leave items sitting in the basket. People who don't want to pause won't open your app halfway through a spree.
So the app only works for the people who need it least.
3A feature dressed up as a solution
Voice is a delivery mechanism. It solves for a feature rather than the real problem, which is why people overspend when they know they shouldn't.
It also walks straight into a crowded room. Mint, YNAB, Copilot, Rocket Money and every bank's own spending alerts are already there. And there's a question it can't dodge: why would someone give you access to their transactions when their bank already tracks them?
The £0 smoke test
Test whether the pause moment works at all before building anything voice-based.
Can you get 20 people to text a friend before buying anything over £50? If they won't send a manual text, they definitely won't speak into an app.
If it does work, then think about the form factor. Automatic triggers based on location, time or the shop itself would beat waiting for someone to choose to open an app.
The behavioural economics are real. An app in the middle of a purchase is just not where intervention works.