Before an AI assistant can qualify a lead or draft a follow-up, it reads your data. If the record is wrong, duplicated, or missing consent status, automation doesn't just underperform — it produces confident mistakes at scale.
The preparation checklist
- Deduplicate with explicit keys (email, phone, address) and a merge policy that preserves history.
- Normalize the fields your rules depend on: sources, stages, statuses. 'FB', 'Facebook', and 'facebook ads' are three sources to a computer.
- Make consent explicit: channel, disclosure text version, timestamp, and source. If you can't say when and how someone opted in, treat them as not opted in.
- Separate facts from notes. AI can summarize notes; it should act on structured fields.
- Define ownership. Every automated action needs an accountable human owner and an audit trail.
Then scope the AI narrowly
Give each AI job a description: what it reads, what it may do, and where a person reviews. A qualification assistant that scores against readable criteria is auditable. A vague 'AI does follow-up' mandate is not.