Dinner from Taco Bell. ETA: 40 minutes.
Driver reaches the restaurant. Doesn't move. ETA creeps up. Order misses its window. I contact support, get bounced to a bot, eventually get a human, get a refund.
Annoying — but it happens. I moved on.
A few days later, I order from the same restaurant again.
Same exact sequence. Driver reaches. Waits. ETA drifts. Refund.
Now I'm not annoyed anymore — I'm curious.
So I ordered a third time. Same failure, beat for beat.
This time I didn't ask for a refund. I asked to speak to a lead.
Because here's the thing: the platform didn't need me to tell it anything. It already had every signal it needed — driver dwell time, ETA drift, restaurant SLA history, complaint history, peak-hour load. All of it was sitting right there.
It just wasn't doing anything with it until after I'd already had a bad experience.
That's not a support problem. That's a product problem.
Out of complete curiosity, and based on what I've understood so far about how product flows and AI systems are designed, I sat down and tried to sketch out a possible fix anyway.
A predictive operational layer that flags a likely delay before the ETA breaks, gives the customer a real choice (wait with an updated ETA, or cancel instantly), and simultaneously flags the restaurant to ops.
Full breakdown in the carousel below — the signals nobody was using, the redesigned tracking screen, the system architecture, and the business case.
The lesson underneath it all: great products don't wait for customers to complain. They catch the pattern before the customer has to.