01 / The question
Data as
an asset.
Can we deliver tomorrow’s work safely, with the right people and equipment, at an acceptable margin? The answer crosses five systems. This is a journey from their separate records to one dependable decision, and the responsibility that remains with people.
One answer worth trusting.
A fictional business · synthetic records · no connected systems
02 / Separate perspectives
Useful alone.
Incomplete together.
The asset register knows the machine. HR knows the person. Finance knows the costs. Safety and field tools capture what happened. Each system has authority over its own facts; none can answer the whole business question alone.
- A-017Equipment available?
- P-042Worker eligible?
- J-204Work worth doing?
Joining data starts with agreeing identities, meanings and ownership.
03 / The pressure of growth
More sites.
More coordination.
One team can bridge gaps through conversation. Across sites, exports, handovers and workbook copies multiply. Someone must reconcile the differences before a plan can be trusted. Late costs hide margin pressure; mismatched identities obscure availability; key-person knowledge becomes a continuity risk.
- Source exports
- 5
- Workbook copies
- 1
- Possible pair checks
- 0
Illustration: five exports per site per cycle. Comparing every copy with every other gives n(n − 1) / 2 pairs, not measured staff hours.
04 / When Excel becomes the process
A useful tool.
A fragile handover.
Excel is excellent for exploration and modelling. Trouble starts when files also own collection, joins, rules, approvals and distribution. Copies drift; local formulas diverge; a person becomes the integration layer. Co-authoring helps, but does not establish cross-system authority or tested business rules.
- ExportA snapshot of the source.
- AmendA local rule changes the answer.
- ReconcileWhich copy was approved?
Keep Excel as a consumer. Give shared rules and repeatable workflows an owner.
05 / A shared foundation
Collect evidence.
Keep its context.
A data lake retains tables, events, files and documents for reuse. Batch ingestion, captured changes and streams serve different needs. Storage alone creates no shared meaning: a catalogue, access controls, lineage and owned processing make the evidence discoverable, usable and accountable.
Sources retain operational authority. A lakehouse adds managed tables; a warehouse may be sufficient for structured reporting.
06 / Bronze · preserve
What did the
source say?
Retain the original evidence, its source, event time and arrival time. Do not silently overwrite strange identifiers or dates. This is the starting point for replay, investigation and correction. Follow the highlighted record into silver and gold; its identity stays connected to this source.
source_id: "EX-17" service_date: "23/09/2026"
Asset register · original values retained
07 / Silver · reconcile
Same evidence.
Clearer identity.
Parse dates, standardise units, reconcile identities and remove genuine duplicates. Preserve provenance and correction history. Domain rules already matter here: an expired credential is a valid fact, while an ambiguous machine ID needs a steward. Neither should be silently turned into a positive answer.
asset_id: "A-017" serviced_at: "2026-09-23" identity_map: v2
EX-17 resolves to one machine.
08 / Gold · give it purpose
Facts combine.
An answer emerges.
Join relevant silver datasets and apply versioned domain logic. Produce equipment availability, workforce readiness or project contribution. Each product needs an owner, grain, definition, freshness target and permitted audience. Gold is prepared for a purpose. It is not a promise of perfection, security or authority to act.
A-017 + inspection S-091 available_for_review: true
Equipment availability · one asset per snapshot · maintenance planner
09 / A view for each decision
Shared facts.
Different questions.
A supervisor needs eligibility. Finance needs contribution. HR needs credential evidence. Shared definitions let them reason consistently without giving everyone every record. Semantic models provide common relationships and measures; role-scoped products expose the detail each decision actually needs.
Tomorrow’s readiness
Verify the worker, machine and inspection.
Eligibility and restrictions; private HR details excluded.
Check at assignment and before work.
10 / The age of an answer
Fresh on screen.
Old at source?
An event happens, someone records it, a platform receives it, then processing makes it usable. A fast dashboard cannot fix a late submission. Real-time describes delivery; medallions describe preparation. A live incident count still looks backward. Choose freshness and reporting periods to match the decision.
- Event
- 15:52
- Recorded
- 15:53
- Received
- 15:54
- Usable
- 15:55
J-204 · 16.7% provisional contribution
Revenue plan $12,000 less $8,500 posted and unposted committed costs and a $1,500 labour allowance.
11 / When evidence changes
Show uncertainty.
Give it an owner.
A missing timesheet is not zero labour. An ambiguous identity is not availability. An expired credential can confidently mean “not eligible”. Inject a fault, then resolve it with new evidence. The answer changes; operational approval remains a separate step.
Ready for planning review.
Current evidence meets the illustrative planning checks.
No corrective evidence recorded.
12 / From insight to action
Inform.
Then authorise.
Evidence supports a recommendation. A named reviewer and an authorised operational workflow make the change. Scoped, read-only gold products also give AI more consistent evidence with less direct production access. They do not prevent hallucination or confer permission to act.
Review the proposed assignment.
No assignment change recorded.
Execution needs an authorised API, validation and audit trail. Evidence changes invalidate approval. No real service is connected.
13 / Start with one decision
Make it dependable.
Then reusable.
Choose one valuable decision. Give its product an owner, tested rules, lineage, access boundaries and a recovery path. Measure delay, reconciliation effort and outcomes before expanding. Retention, cost and adoption need ownership too: a shared platform can amplify an incorrect definition as quickly as a good one.
Then another useful connection.
Choose the smallest fit: better source processes, governed exports, APIs, a focused warehouse, or a lake/lakehouse. These can coexist.
Concept references: Data lakes · Medallion layers · Excel collaboration
Synthetic scenario · 24 September 2026, AWST · Use Next / Back or arrow keys. On small screens, scroll to read.