Data readiness before AI: what to clean up first

You don't need a data lake

Enterprise blogs make data readiness sound like a six-month project. For a 5-50 person business, it's usually a one-week sweep across four places: your shared drive, your CRM, your operational tools, and your people's heads.

1. The shared drive (or wiki)

This is where most SMB AI projects get grounded - chat with your docs, your SOPs, your contracts. The work isn't fancy:

• Delete duplicates. If there are three versions of the pricing PDF, AI will quote whichever it grabs first.
• Mark stale documents (anything not updated in 18+ months) and archive or refresh.
• Pick one canonical folder per topic. Move outliers in.
• Strip personally identifiable info from training-grade samples.

2. The CRM

If AI is going to draft follow-ups, suggest next steps, or route leads, it's reading whatever's in your CRM. Most SMB CRMs are 30-50% empty.

• Define the 8-10 fields you actually use. Hide the rest.
• Backfill the must-haves on active accounts only - don't try to fix history.
• Standardize stage names. "Proposal sent" and "Proposal Sent" and "Sent proposal" become three stages to the model.
• Decide what "qualified" means in one sentence. Write it down.

3. The operational tools

Whatever your team lives in day to day - ticketing, scheduling, invoicing, project management. The question is: are the records clean enough to be useful?

• One open record per real thing - close out orphans.
• Consistent naming. AI handles fuzziness better than it used to, but only to a point.
• Make sure timestamps are correct. Many AI workflows reason about recency.

4. The "in someone's head" data

The most valuable data in any SMB is the stuff that lives only in the owner's head. It's also the data AI most needs to be useful. The fix is a one-day documentation sprint per role - record loom walkthroughs of the 5 most common workflows, transcribe them, paste them in.

This is the highest-leverage prep you can do. A 90-minute loom from the owner often becomes the single most important document in the entire knowledge base.

What you do NOT need to do

• Buy a data warehouse.
• Hire a data engineer.
• Migrate to a new CRM "to get AI-ready" (almost always wrong).
• Try to perfect everything before starting - aim for "good enough that the AI doesn't quote the wrong year."

A one-week readiness plan

1. Mon-Tue: Shared-drive cleanup. Delete, archive, dedupe.
2. Wed: CRM field reduction and stage standardization.
3. Thu: Loom walkthroughs of the 5 highest-value workflows.
4. Fri: Pick a single AI project. Point it only at the cleaned subset.

Do that and you're more "AI-ready" than 80% of the companies hiring consultants to make them AI-ready.