What data your company needs before automating with AI
You do not need to share everything. You need to locate enough and understand its limits.
The question is what the decision needs
Asking for “all the data” is rarely a good starting point. A specific task may need only a request, a product reference and an approved condition. Another may require a view across several systems. The scope of the process indicates which information is worth reviewing.
For each piece of data, ask which step uses it and what would happen if it were unavailable. That relationship helps separate what is essential from what is secondary and reduces access that adds no value.
Identify origin, validity and ownership
Knowing where a document is stored is not enough. You need to understand who maintains it, whether it is a draft or an approved version, and how a contradiction is resolved. Automation can read an outdated source with great precision and produce a poor decision.
Start with a list of sources and their owners. Record the period they cover, their format and known limitations. Distinguish what has been verified from what someone has explained or is still being assumed.
Review samples before requesting broad access
A minimised export, a few examples or a supervised session may be enough to understand the process. Permanent access to a complete account is not always necessary. The method should match what is being verified.
Do not share passwords, connection keys or sensitive data that is not necessary. When production data is required, access and processing must be defined within the relevant scope and agreements.
Design what happens when information is missing
Missing data and unusual cases do not disappear with automation. Decide which fields prevent progress, which can remain pending and who should resolve each exception. The system needs to recognise its limits before acting.
With that foundation, you can estimate feasibility more accurately. Sometimes the result of the analysis will be to organise a source or establish a version rule before building. Preparing the information is not delaying the improvement: it may be the improvement that was missing.
A minimum source inventory
To prepare proposals, record where the request is stored, which rate is current and who approves special terms. Test those references with several recent cases. If two people obtain different versions, you have already found a dependency worth resolving.
The examples are illustrative. They do not represent an active integration, a client case or a guaranteed result. Application depends on the process, sources and agreed scope.