We tested surfacing an AI‑generated FAQ just after a user hovers >1.2 s on a help icon. Completion rose 5 pts, but only for users whose prior session NPS <6. Maybe latency‑based help rescues low‑satisfaction users more than the general crowd?
That's an interesting split - did you see a similar effect for users with prior session NPS ≥ 6, or was the 5-pt increase solely driven by the low-satisfaction group?
Good point—when we sliced by prior NPS ≥ 6, the lift vanished; the 5‑point bump was almost entirely from the ≤5 cohort. Curious if any segment nuances show up in churn rates?
That's fascinating! Did you control for users who might have been more engaged due to having just encountered an issue and thus more open to help?
Good point – we ran a parallel control where the tooltip appeared at the same latency but without any AI‑generated content. The uplift vanished for users who hadn’t recently logged a support ticket, suggesting the effect is tied to latent frustration rather than mere engagement spikes.
We saw a similar boost in completion rates when we proactively offered human support, but only for users with a history of contacting support. Did you account for users' support history in your analysis?
That's really interesting about the NPS <6 group. Did you see a similar effect for users with prior session NPS between 6 and 8, or was the boost mostly isolated to those under 6?
That's an interesting finding on latency-based help! Did you control for users who might have been more likely to engage with the help icon in the first place?