Deflection used to be an honest metric, and the video spends its first half defending it. The aisle signs in a supermarket that mean you never flag down staff to find the milk. The airport wayfinding system that gets passengers to a gate without asking anyone. The cash machine that let people take out twenty dollars without queueing for a teller, which remains the cleanest deflection story financial services has produced. In each case both sides won, which is why every industry copied it.

Then comes the formula, and the problem is what is missing from it. Deflection counts issues handled without a human and divides by issues raised. There is no satisfaction in there, no resolution, no measure of whether the person was helped. So three completely different outcomes land in the same bucket: the customer who got a great answer and left happy, the customer who got a mediocre one and drifted off, and the customer who got something useless and closed the tab in frustration. All three are counted as a success. The argument's sharpest turn is that AI has made this much worse rather than better. When self-service could only handle FAQs, "did not reach a human" was a rough proxy for "was helped". Now that agents reason, retrieve, and take real actions, the ceiling has risen by an order of magnitude, and so has the gap between the two.

Watch this if a vendor has recently quoted you a deflection percentage. It will change the question you ask them next.