What You Will Be Able To Do

• Locate the data a use case needs
• Tell structured from unstructured data
• Assess quality on six dimensions
• Explain access and least privilege
• Assign owners and a simple routine

Why Data Decides The Outcome

• A tool is only as useful as what it can read
• No price list means invented prices
• Inconsistent records give confident wrong forecasts
• Effort on information equals effort on tools

Where Company Data Lives

• Systems of record: ERP, CRM, HR
• Work tools: email, chat, drives, wikis
• Shadow layer: personal spreadsheets, exports
• Outside sources and paper records

The Data Inventory Page

• For each source: what it contains
• Who looks after it
• How someone reaches it
• Sensitivity level and quality findings

Source: Supplier invoices (email + ERP)
Contains: date, supplier, amount, PDF
Owner: Head of Accounts Payable
Access: AP team, read-only for audit
Level: confidential

Structured And Unstructured Data

• Structured: rows and columns with fixed meaning
• Unstructured: emails, contracts, calls, images
• Most company information is unstructured, per industry estimates
• Language models are first usable tools for reading it

Retrieval-Augmented Generation In Brief

• Search approved documents first
• Give the best passages to the model
• Model answers from those passages
• Quality of answer equals quality of documents

Question -> search approved documents
         -> pick best passages
         -> model answers from passages
         -> show sources to the user

Data Quality: Six Dimensions

• Accuracy, completeness, consistency
• Timeliness, uniqueness, validity
• Sample 50 records and count problems by type
• Record the share of records affected

Case: A Spacecraft Lost To A Unit Mismatch

• 1999: Mars Climate Orbiter lost on approach
• One team used pound-force seconds, another newton seconds
• Each system worked as designed
• Shared definitions and ownership were missing

Access And Least Privilege

• Each person and system gets only what the job needs
• AI search must respect the asker's permissions
• One broad account can expose everything
• Review access every quarter

Four Sensitivity Levels

• Public: safe to share
• Internal: for staff only
• Confidential: limited groups
• Restricted: personal data, results before release, trade secrets

Ownership Roles

• Owner: accountable, decides who may use data
• Steward: maintains definitions, fixes errors
• Custodian: runs storage systems and security
• Small teams may combine roles, never skip them

Light Governance Routines

• One shared glossary of key terms
• Lineage: where data came from and how it changed
• Catalog once you have more than a handful of datasets
• Start with one use case, one owner, one page of rules

Recap

• Inventory where data lives, who owns it, how to reach it
• Check quality with a small sample
• Tag sensitivity and respect permissions
• Name owners and fix red items with dates

Next Week

• Build, buy or partner
• Vendor types and foundation models
• Open versus closed models
• Total cost of ownership and lock-in

Questions And Exercise

• Exercise: one-page data inventory for your top use case
• Sample 50 records and mark items green, amber, red
• Confirm with named owners
• Questions from the room