Agency Unveils AI That Predicts The Exact Moment A Client Says ‘Just One Small Tweak’
Independent creative shop Lantern & Vale has announced a breakthrough in artificial intelligence, a system trained to predict the precise moment a client will request a small tweak that quietly resets three weeks of work.
The tool, called Prevision™, was trained on 11 years of meeting transcripts, muted sighs, calendar invites titled "quick catch up", and the subtle facial expression a marketing manager makes just before introducing feedback from "someone internal". Early tests show the system can forecast the phrase "could we explore one more direction" up to 14 minutes before it is spoken. Staff are encouraged to emotionally prepare during this window.
Lantern & Vale says the model has also become unusually skilled at generating the actual feedback in advance. It produces realistic notes such as "can it feel more premium but also more accessible" and "my partner showed it to their running group and they wondered if the headline could smile more". One feature automatically inserts the sentence "we love where this is going" before recommending a completely different approach.
The agency has also released a public version of the training prompt for junior strategists: "You are a stakeholder who has skim read the deck and joined the call late. Suggest three changes that are impossible to measure but sound decisive." Developers say the next update will attempt something even more ambitious, predicting the future sentence "this is great, let's not change a thing". Early simulations suggest the model believes this event may occur sometime around late 2034.
The tool, called Prevision™, was trained on 11 years of meeting transcripts, muted sighs, calendar invites titled "quick catch up", and the subtle facial expression a marketing manager makes just before introducing feedback from "someone internal". Early tests show the system can forecast the phrase "could we explore one more direction" up to 14 minutes before it is spoken. Staff are encouraged to emotionally prepare during this window.
Lantern & Vale says the model has also become unusually skilled at generating the actual feedback in advance. It produces realistic notes such as "can it feel more premium but also more accessible" and "my partner showed it to their running group and they wondered if the headline could smile more". One feature automatically inserts the sentence "we love where this is going" before recommending a completely different approach.
The agency has also released a public version of the training prompt for junior strategists: "You are a stakeholder who has skim read the deck and joined the call late. Suggest three changes that are impossible to measure but sound decisive." Developers say the next update will attempt something even more ambitious, predicting the future sentence "this is great, let's not change a thing". Early simulations suggest the model believes this event may occur sometime around late 2034.