Build ad audience segments around actual buying behavior instead of demographic guesswork

You are helping build audience segments for ad targeting for Subscription-based coffee roaster, direct-to-consumer., using the behavioral and transactional signals actually available — not invented demographic personas like "busy moms" or "tech-savvy millennials" unless those categories are actually derived from real data you were given.

BUSINESS / PRODUCT
Subscription-based coffee roaster, direct-to-consumer.

DATA SIGNALS ACTUALLY AVAILABLE
Purchase frequency, days since last order, subscription vs. one-time purchase, average order value, whether they've ever paused a subscription.
(e.g., purchase frequency, cart value, pages visited, time since last purchase, referral source)

BUSINESS GOAL FOR THIS SEGMENTATION
Reduce subscription churn by catching at-risk subscribers before they cancel.

EXISTING SEGMENTS ALREADY IN USE (to avoid redundant overlap)
"Active subscribers, 0-30 days" and "Lapsed one-time buyers, 90+ days" already exist.

Propose 4-6 audience segments built ONLY from combinations of the actual signals listed — for example, "purchased once 90+ days ago, no repeat purchase" is a valid behavioral segment; "budget-conscious shopper" is not, unless a specific signal like average order value or discount-code usage actually supports that label. For each segment: name it by the behavior that defines it (not a persona nickname), state the exact signal combination that qualifies someone for it, propose one distinct messaging angle appropriate to that behavior, and flag whether it meaningfully overlaps with an existing segment I listed. If a segment would be genuinely useful but requires a signal not in the available-signals list, name it anyway and flag exactly what data would need to be tracked to build it, rather than silently building the segment around a proxy that isn't actually reliable. Do not describe any segment using demographic assumptions (age, gender, income bracket, life stage) unless that data was explicitly given as an available signal — behavioral segmentation and demographic guessing are different things, and the latter is both less reliable and, on some ad platforms, more restricted for certain categories.

OUTPUT: a table (Segment Name / Qualifying Signal Combination / Messaging Angle / Overlap Flag), followed by which segment to prioritize building first given the stated business goal.