What is a data image?
A data image is a unified, continuously updated representation of an organization’s data, assembled from many source systems into one consistent model. Where a warehouse load produces a snapshot and a report produces a view, a data image is meant to be a living copy. It is refreshed as the sources change, it maps every source’s fields onto one shared structure, and it keeps the relationships between records intact, so that a customer connects to its orders, the orders to their shipments, and the shipments to their invoices even when each of those lives in a different system.
How a data image works
Connectors pull data from the ERP, CRM, manufacturing, and ticketing systems, by nightly batch where that is enough and by change data capture or event streams where freshness matters. Transforms normalize units, codes, dates, and identifiers into a canonical schema. An entity resolution step, usually part of master data management, decides that the “ACME Corp” in the CRM and customer 10442 in the ERP are the same company. The result lands in a lakehouse or similar store, with the raw copies kept alongside for lineage. Dashboards, machine learning models, and agents using retrieval or tools all read from the image instead of from the sources directly.
Why a data image matters
Each new AI use case otherwise rebuilds its own integration. With an image in place, a margin model, a customer service agent, and a demand forecast are three consumers of one governed dataset, and a fourth use case starts from data that already exists. A specialty distributor that has acquired two competitors and runs three ERPs can present one customer list, one item master, and one order history to every downstream tool while the ERPs keep running.
The image is derived data. The source systems remain the systems of record, and an agent that needs to change an order should write to the ERP through its API, with an approval step, rather than editing the image.
What to watch for
- Freshness varies by source, so publish the refresh time for each part of the image and make agents aware of it before they quote inventory or balances.
- An image exposes bad source data without fixing it, so plan data cleansing and ownership alongside the integration work.
- Mappings break when a source system is upgraded, so assign someone to own each connector.
- If the image lives in a proprietary platform, confirm that you can export the full model in open formats.