Resources
Can your data answer the question you care about?
Assess whether connected data is ready for AI across six practical dimensions.
The approach
How does it work?
- 01 Choose a problem
Agree on goals and checks first
- 02 Prepare together
Reduce duplicated preparation across projects
- 03 Validate at the site
Use evidence to decide the next investment
- 04 Expand
Reuse the foundation for the next use case
Details
Scope and considerations
1. Define the question and user
Write who uses the data and for which decision. “Maintenance reviews signals and service history before a stop” identifies periods and evidence. Choose an evaluable task before a model type.
2. Verify sources and access
Map sources, owners and access methods. Compare available with required fields and inspect samples. An API does not guarantee access to every required value; gaps may require another source or a narrower scope.
3. Align time and identity
Check event/ingestion time, time zones and synchronization. Map equipment, product, lot and cycle identifiers across systems. If matching is inferred, record its conditions and reviewer separately from verified matches.
4. Check quality and exceptions
Inspect stops, shift changes, changeovers and outages as well as normal runs. Record gaps, duplicates, invalid ranges, latency and unit changes. Keep missing data distinct from a real zero.
5. Prepare evaluation evidence
Confirm labels, inspections and action history. Avoid evaluating on near-duplicate lots or adjacent periods used for training. Define held-out periods, products or equipment and a domain reviewer.
6. Keep data operationally usable
Assign ownership, change notices, quality rules, access and retention. Track downstream impact of sensor replacements and new tags. Mark each checklist item ready, needs work or unknown for a useful consultation baseline.
Start together
Tell us what you want to improve first.
Tell us about your equipment, data and priorities to review a path from connection to operations.