AdZeta is an end-to-end data‑intelligence & activation platform that helps advertisers acquire high-value customers using the power of predictive LTV modeling.
Location : Remote (North America time zones)
Team : Ad Signals, Data & Activation (pLTV)
Type : Full-time (open to Senior, Staff, or Principal leveling)
The Mission :
The right candidate will own the end-to-end pipeline that converts predictions into bid-ready, calibrated pLTV signals , and reliably activates those signals across Meta (CAPI), Google (OCI / Enhanced Conversions + Value Adjustments) and Amazon (AMC→DSP) . You’ll build resilient GCP pipelines, shape / monitor predictions, and ship server-side conversions that move ROAS, not just dashboards.
Roles & Responsibilities
Feature & modeling interface
Partner with ML to operationalize user-level pLTV features (RFM, product mix, acquisition metadata) with point-in-time joins .
Productionize Vertex AI training & batch / online scoring (Pipelines, Feature Store, Model Registry).
Own calibration (decile calibration tables, winsorization, shrinkage) for stable platform learning.
Server-side activation
Build stateless “ signal translators ” on Cloud Run for Meta CAPI, Google OCI / EC + Conversion Value Adjustments , and Amazon AMC Inputs / Audiences .
Ensure dedupe (event_id parity), rate-limiting , idempotency , and retry / Replay mechanics with DLQs.
Maintain secrets / keys, rotate tokens, and validate payloads (test modes, acceptance checks).
Observability, reliability, and safety
Ship SLOs & alerts : data freshness, EMQ / match rates, adjustment acceptance rates, API error spikes, queue lag, drift.
Operate kill-switches and fallbacks : immediate revenue mode, decile throttling, shadow testing.
Write runbooks for common incidents (schema drift, API 4xx / 5xx, value restatement rejections).
Experimentation & performance
Design A / A and A / Bs (holdouts, geo-lift, incrementality) to validate pLTV value vs baseline.
Iterate on send timing (D0 vs D1–D7), horizon (H=90 / 180 / 365), and value shaping; quantify impact on tROAS / CAC.
What You'll Own (KPIs)
Meta : Event Match Quality (EMQ), dedupe Google : Import latency, Conversion Adjustment acceptance ≥98% , tROAS lift vs. baseline.
Amazon : AMC audience match rate & size (≥2,000 active), DSP performance vs. all-site.
Data / ML : Pipeline freshness SLOs, coverage of hashed IDs, drift / PSI guardrails, on-time retrains.
Reliability :
Core Qualifications :
7-10+ years building production data+ML activation systems (preferably in adtech / martech).
Proficiency with Meta CAPI, Google Ads offline conversions & Enhanced Conversions, Conversion Value Adjustments, hashed identity formatting, and dedupe mechanics.
Expert SQL (BigQuery) and Python (ETL, API clients, data validation).
Hands-on GCP : BigQuery, Pub / Sub, Dataflow, Cloud Run / Functions, Vertex AI (Pipelines, Feature Store, Endpoints).
Strong grasp of identity graphs, point-in-time feature building, and leakage prevention.
Production observability : Cloud Logging / Monitoring, alerting, SLOs; comfort with queues & DLQs.
Security & privacy hygiene : KMS / Secret Manager, VPC-SC, PII handling, RTBF.
Compensation :
Competitive base + performance bonus tied to platform KPIs (acceptance / EMQ / lift). Senior / Staff / Principal leveling available.
Engineer • Amravati, Maharashtra, India