Zoho CRM is where most Australian SMBs want AI first: score leads, route owners, suggest next actions, draft follow-ups. The failure mode is turning on Zia on a pipeline nobody trusts, then blaming the AI.
Phase 1 is data readiness: dedupe leads and contacts, agree picklists, reduce stages to what reps actually use, and ensure source fields are populated. Without that, scoring models and GPT classifications inherit noise.
Phase 2 is native Zia where licensed: prediction and scoring, anomaly alerts on deal slippage, and workflow suggestions. Train sales leads on what Zia is optimising for; otherwise scores get ignored like any other unused field.
Phase 3 is workflow automation with optional AI steps: web form submission → Flow → classify text → assign owner and task → notify Slack or Teams. Deluge handles API calls to OpenAI or Azure with try/catch, admin alerts on failure, and no auto-send email without a human gate early on.
GPT boundaries we document: which fields leave Zoho, retention policy, prompt version, and fallback when API is down (queue for manual triage). Australian privacy expectations mean minimising personal data in prompts and keeping audit logs.
Measure one metric for 30 days: time-to-first-contact, MQL-to-meeting rate, or rep hours on triage. Expand only if the metric moves. We scope CRM AI work in our hourly bands after discovery, often alongside marketing automation integration.
Published 13 August 2026