THE PRACTICAL GUIDES
AI data readiness in Australia: define the foundation before buying a build
An AI proposal may start with data readiness, but that phrase can describe a short assessment, an agreed set of definitions or a working data foundation. These are different deliverables. Before buying a model or agent, identify the decision it will support, the records it needs and how you will know those records are complete and current.
This guide compares published service scope from two existing Australian providers and develops an original commissioning example. It does not score either provider's delivery or establish a standard market fee. For the later modelling stage, read the forecasting project guide.
Choose the deliverable before the technology
| Work you could commission | Useful output to define | What it does not establish by itself |
|---|---|---|
| A bounded readiness assessment | Named sources, record definitions, tested gaps and a prioritised plan with owners. | A functioning connection, corrected records or a production data flow. |
| An implemented data foundation | Agreed sources connected, definitions applied, checks running and operating responsibility handed over. | That a later AI model is accurate, economical or appropriate for the business decision. |
| An AI application using that foundation | A specified task with evaluation, access rules, monitoring and a fallback. | That every historical source was complete or that later changes require no maintenance. |
Use supplier descriptions to ask a more precise question
Data Army's AI service page puts preparation and structured data foundations alongside AI implementation, workflows and continuing support. Its data-engineering description covers architecture, integration, data platforms and pipelines. That is a reason to separate foundation work from application work in a brief, rather than assume one AI quote includes both.
RUBIX's Data Foundation offer advertises a fixed-price engagement and an eight-to-twelve-week path from kickoff to live. It describes connecting priority systems, agreeing definitions, cleaning and quality checks, and a governed layer for reporting and later use. The reviewed page does not publish a fee amount. Its advertised range is a description of the offer, not a guaranteed completion date for your sources.
Examples of accounting, CRM, commerce and file sources do not establish that every connector, migration or historical repair is included. Ask which named systems, accounts, fields, history and refresh pattern belong to the fixed scope, and what happens when a source is unavailable or behaves differently from the initial assumption.
Start with one business total you can explain
Consider a fictional retailer preparing order data for a later forecasting project. There are two orders: A has an order total of A$100 and B has A$50. Order A contains two item records and B contains one. The original order total is A$150.
If a table places the full order total beside every item, A$100 appears twice and A$50 once. Adding that column gives A$250. No AI is needed to create this discrepancy; the rows describe items while the repeated amount describes an order. PostgreSQL's official explanation of joined tables supports the underlying point: a source row can produce several matched rows.
The example uses invented numbers, not a client dataset or a measured supplier result. Its practical lesson is to define what one row represents and which records may be counted together. An apparently complete connection can still produce a misleading total if the definitions and relationships are wrong.
- Unit: distinguish one order, one order item and one payment; identify which output you need.
- Keys: identify the order and item identifiers, repeated records and unmatched items.
- Amounts: define currency and whether the chosen total includes tax, shipping, refunds or cancellations.
- Boundary: record which source and cutoff produced the comparison total, including the relevant timezone.
Agree an expected reconciliation for a small approved sample, then a repeatable check over the defined scope. Do not quietly discard unmatched or duplicate records to make a dashboard look tidy. Keep their number, reason and owner visible, with an agreed resolution or explicit exclusion.
Freshness needs a timestamp and a response
“Live data” is too vague for an acceptance test. A buyer might instead specify that the daily planning view must include all available completed orders through a stated cutoff, refreshed by a stated time. The exact rule belongs to the business process, not a generic promise that faster is always better.
Ask the supplier to expose the source cutoff, most recent successful update, record counts and failed or delayed sources. Decide what the user sees when yesterday's update fails: an age warning, the last accepted snapshot or a paused recommendation. A successful scheduled job alone does not establish that its contents are current.
Source access also needs a boundary. Name the approved systems and authorised data owners, the access required for each stage and who approves a new field or use. An assessment should not silently expand into unrestricted production access.
Write a handover you can operate after the project
For a readiness assessment, request a source inventory, agreed definitions, evidence for the tested gaps and an implementation plan with exclusions. For an implemented foundation, add working outputs, reconciliation results, refresh checks, failure handling, documentation and a named operating owner. Specify which deliverables must be demonstrated before acceptance.
Keep continuing costs visible: hosting, connector charges, monitoring, support and changes when a source schema moves. Establish who receives a failure alert, who can correct the source and who decides that an unresolved gap prevents a later model from being used. A report handed over and a maintained pipeline running are different completion points.
Sources reviewed 23 October 2026, Melbourne time. Provider descriptions are published claims, not audited outcomes. No assessment form was submitted or system trial performed. Use the quote comparison to record scope and continuing costs, browse implementation providers or return to all AI consulting guides.
Editorial guidance with linked sources. Reviewed 2026-10-23.