Trust Half-Life
One bad bot answer and customers skip it for weeks. Hear it tick.
The job
Measures how a failed bot conversation changes customers' next contacts, like asking for a human straight away, how long that lasts for each topic and what it costs. Then fixes the worst topic and makes amends with the customers it let down.
The moment
Each time a customer who had a bad bot experience asks straight away for a human, the app clicks like a Geiger counter and flashes their first words.
The finding: customers let down once on a blocked card ask for a human in their first message for the next 41 days. That costs $310,000 a quarter in lost trust.
The team fixes the bot's handling of blocked cards.
The week after the fix, the room hears the clicks slow down, each one still flashing a real customer typing "agent".
What it does
- Assign a fix
- Draft the fix brief and make-good
- Send the make-good and label the cohort
- Arm the sound
What you see
Geiger-counter sound where each tick flashes a real customer's first words, with radioactive-decay curves per intent
What it moves
How many days it takes for the loss of trust to halve on each topic, and its cost in dollars: the contacts that skip the bot multiplied by the cost difference between a human and the bot.
Built for
- Product & digital
- Finance
- Frontline staff
