Preparing Business Data for AI: 6 Things to Do Before You Build

1. Set one question; do not ingest the whole archive

When a team hears “build internal AI”, the usual reflex is to connect every SharePoint folder, Drive and mailbox. That feels fast in week one, but creates a harder problem: the system cannot know which document is current, who may see it or which answers require care.

Start with a scoped question: where does sales find the warranty terms for product line A, or which document contains the current returns policy? Then list each source, its owner and its update date.

If nobody owns an error or an expiry date, that document is not ready to become a customer-facing source.

Good AI data is not the largest collection. It is data that fits the job, remains current and has an owner.

2. Complete six steps before connecting data

Classify sources by job, identify the current version, set access, standardise names and dates, exclude sensitive information, then prepare ten to twenty real user questions. An assistant is useful only if it can answer routine questions with evidence.

Recent reporting on Vietnamese business AI also points to gaps in data, process and delivery capability, not merely tool choice. Data preparation is not a side task; it keeps a pilot on target.

3. Give every answer a safe exit

AI should not invent an answer to fill a gap. When sources are missing or confidence is low, it should say so, offer the closest current document or hand the work to the right person. That matters for prices, policy, legal content and customer data.

During the pilot, retain the question, sources used, reviewer feedback and refusals. The log improves data and provides evidence for a scaling decision.