Data readiness guide: what your data answers mean and how to improve them
The Intelecy value calculator asks seven questions about your data and organization. This guide explains what each one measures, what a strong answer looks like, and how to close the gaps — because readiness determines how fast you capture value, not how much value exists.
Why data readiness determines time-to-value
Readiness affects the speed and certainty of results — not the size of the underlying opportunity. A site with weak data can still have a very large value pool; the gap simply moves that value further out the timeline.
- Strong readiness shortens the path from connection to first insight to weeks.
- Weak readiness does not reduce value at stake — it delays it and widens the estimate range.
- Every readiness gap has a concrete remediation step; none are permanent blockers.
Question 1 — How is plant data used today?
This measures your analytical maturity baseline: where on the path from manual records to live production models your site sits today.
- A strong answer: data is stored in a historian and already used in dashboards or analytics.
- A weak answer (manual records or control screens only) means the first phase is a data foundation step, not a modeling step.
- What it affects: your starting point on the pathway and the time to first insight.
Question 2 — What data can you access?
This measures signal coverage. Different use cases need different signals: anomaly detection needs multivariate process data; quality use cases need lab results; reliability use cases need maintenance event records.
- A strong answer includes historical process data, real-time access, and context (quality, maintenance, or batch information).
- Missing maintenance or quality records limits which use cases can start first — the model can predict drift, but verified failure modes need event history.
- What it affects: which starting use cases are viable and the confidence of the estimate.
Question 3 — How much history is available?
Models learn normal behavior from history. Seasonal processes need enough coverage to see a full operating cycle.
- A strong answer: twelve months or more of continuous history.
- Less than three months means a data-collection period is added before training.
- What it affects: time to verified value and backtest depth.
Question 4 — What is the typical data resolution?
Resolution must match the dynamics of the problem. Fast process upsets need seconds-to-minutes data; energy and throughput trends work with hourly data.
- A strong answer: seconds or minutes for the tags that matter to your use case.
- Daily or hourly-only data limits anomaly detection and short-horizon forecasting, but is sufficient for energy baselines and KPI trending.
- Mixed resolution is normal — Intelecy aligns signals on ingestion.
Question 5 — How accessible is the data?
The most common real-world delay is not data quality but the access path: historian exports, IT/OT approval, and fragmented systems.
- A strong answer: data can be accessed now, or with a known IT/OT approval path.
- Fragmented data across several systems adds a consolidation step before modeling.
- What it affects: the length of the data foundation phase more than any other factor.
Question 6 — Can the organization act on the result?
A model nobody acts on produces zero value. This question measures whether a named process owner and an operating team exist to receive alerts and recommendations.
- A strong answer: a named owner and an operating team that can act on insights.
- If the owner exists but the workflow is undefined, defining the alert-to-action workflow is part of the pilot — it is fast when done deliberately.
- What it affects: whether first-year realizable value is captured or only observed.
Question 7 — Can the use case be replicated?
Industrial AI economics improve sharply when a validated model rolls out to similar assets, lines, or sites.
- A strong answer: similar processes across several sites or enterprise-wide potential.
- A unique asset is still worth doing — it simply caps the scaled three-year value.
- What it affects: the three-year scaled value potential, not the pilot.
How to improve data quality — practical steps
None of these require a data-lake project. Most are configuration and process changes on systems you already own.
- Enable historian logging on the tags that matter — start with the constrained unit, not the whole plant.
- Standardize tag naming and units so signals are interpretable without tribal knowledge.
- Close gaps: fix intermittent sensors and log why outages happened.
- Capture context: record maintenance events, quality results, and operating states in a queryable system.
- Establish real-time access through a gateway rather than periodic manual exports.
Readiness vs. value — keep them separate
The Intelecy value model deliberately separates the size of the prize from the speed of capturing it.
- Full annual value potential: the size of the opportunity if fully addressed.
- Expected first-year realizable value: what readiness, scope, and adoption make achievable now.
- Three-year scaled value potential: what replication across assets and sites unlocks.
- Use the value calculator to see all three numbers for your operation.
Want help applying this to your plant?
Bring the unit, outcome, and one historian export — we will show how the Intelecy workflow maps to your numbers.