Most AI initiatives are decided before the first model is chosen. They are decided by the data the model will draw on: who owns it, how good it is, where it sits and what it may be used for. The five questions below are the ones a data readiness assessment has to answer. Each can be answered with information your organization already has.
Someone owns the data the use case depends on
An AI use case needs an answer to a plain question: who decides what this data means, who may change it, and who is accountable when it is wrong. Where nobody owns the data, quality, consistency and access decisions are made case by case, and the use case inherits every one of those decisions without knowing it.
Ownership does not need a large program to start. It needs the data sets the first use case will touch, a named owner for each, and the classification, access and privacy requirements that apply to them written down. That is the core of a governance operating model, and it can begin with one use case and grow from there.
The quality of that data is known, not assumed
Teams usually know where their data breaks. The gaps live in spreadsheets, in the reconciliations that run before every report, and in the exceptions people have learned to work around. A model does not learn those workarounds. It learns the data as it is.
Before the first use case, measure the quality of the data it will use: completeness, consistency across sources, duplicates, and how current the values are. A data quality assessment shows where quality breaks today. Monitoring keeps that view current once the use case is live, so issues are seen and resolved as they appear rather than discovered in the output of the model.
The data can be reached where the use case needs it
Information that is spread across applications and platforms, and does not connect, cannot feed a model without manual work. The first version of a use case often runs on an extract. The production version needs pipelines that bring the data together on a schedule the business can rely on.
The question is not whether the data exists. It is whether it can be integrated, cleaned and transformed into a form the use case can consume, and whether that path can run without a person in the middle. Data engineering and integration is the work that turns a one-off extract into a dependable supply.
The platform underneath can carry the workload
AI and advanced analytics put different demands on a data platform than reporting does: larger volumes, unstructured content such as documents and messages, and workloads that need to scale up and down. Legacy or fragmented environments are hard to scale or change, and the limits show as soon as a use case leaves the pilot stage.
Look at the platform the use case will run on and ask whether it was designed for this workload. Warehouses, lakes and lakehouses each fit different patterns. The target architecture, and the platform and technology choices behind it, are better settled before anything is built on them.
The rules for using the data in AI are written down
The last question is about permission. Which data may be used to train or ground a model, who may see the output, and what has to stay out of scope because of classification, privacy or contractual limits. Governance sets these requirements. The technical protection that enforces them sits with cybersecurity, so ownership stays clear and the two connect.
Rules that exist on paper are not enough. They have to be applied to the use case at design time and monitored once it runs, so that a change in the data or in the model does not quietly take the use case outside the rules.
A no is useful information
A no to any of the five questions tells you what has to be in place before the use case can succeed, and it usually points at work that will serve every later use case as well. Three steps follow:
- Assess the current state in a structured way, so the gaps are documented rather than felt.
- Order the work by the value it unlocks and by what the first use case needs, not by what a full program would cover.
- Fix what the first use case depends on, and let the roadmap carry the rest.
A data strategy and a maturity assessment give that order. They say where your data is today, where it needs to go, and what the first step is. Organizations differ in maturity and priorities, so the right starting point is the one that matches your situation: an assessment, a single capability, or improving data you already run.