What is the best AI model for service-venue retention?

A gradient-boosted tree (XGBoost / LightGBM) on engineered RFM + cadence features beats LLMs at pure churn prediction. LLMs add value at the message generation layer — personalising the win-back copy after the model picks the target.

The honest framing: AI is mostly a targeting and timing improvement over rules-based automation, not a replacement for the relationship work. The best operators combine AI for the long tail (lapsed-by-1-week regulars) with human calls for the top decile (lapsed VIPs).

In Vietnam specifically, Zalo OA is the dominant retention channel — read rates of 60–80% beat SMS (15–25%) and email (8–15%) by a wide margin. Any retention playbook that does not put Zalo OA as the default channel underperforms by 2–4×.

PEKO operationalises this as a loyalty layer that sits on top of an existing booking platform — Booksy, Fresha, Mindbody, KiotViet, or paper diaries — rather than replacing it. Onboarding for a small venue typically takes 3–5 days from contract to first automated message.