All insights
3 min read

Before adding AI, get the data it depends on ready

A practical checklist for sources of truth, data meaning, document search, integrations and access rules before an AI feature goes into daily use.

AI starts with your data — AtmosGate Insights editorial cover

When a business asks for an AI feature, the first useful conversation is usually about data: where it lives, who owns it and whether it can be trusted. The quality of an assistant or an automated workflow is largely decided before anyone writes a prompt.

Models change every few months. The records, documents and rules a business keeps are the part that stays specific to it. Here is the preparation we recommend before a first AI feature is planned.

Recognise the usual obstacles

Most organisations do not lack data. They lack data that can be combined and relied on. Sales records sit in one system, finance in another and operational detail in spreadsheets and inboxes. Each tool works on its own, but together they cannot answer a simple question that crosses departments.

An AI feature connected to that situation inherits every inconsistency. It may also answer confidently from an outdated document or a duplicated record.

  • The same customer stored several times with different spellings.
  • Fields such as “status” whose values nobody can explain.
  • Contracts, procedures and price lists that exist only as scanned files.
  • No clear rule for who may see which records.

Choose sources of truth and document their meaning

For each core entity, such as customers, products or orders, decide which system is authoritative and synchronise the others from it. Clean the data the first AI use case depends on before attempting everything else.

The UK Government Data Quality Framework describes quality in terms of dimensions such as accuracy, completeness, consistency, timeliness, uniqueness and validity, and treats quality as something to assess against a specific use. That is a practical lens here: judge the data against the task the AI feature will perform.

Keep a short data dictionary that explains what each important field means and how it is used. It helps staff, and it helps an AI system interpret records correctly.

Sources and release notes: [1]

Connect systems and set access before search

Fresh data needs reliable integrations between systems rather than monthly spreadsheet exports. Key documents should be converted to searchable text, kept current and organised so that retrieval returns the latest approved version.

Design permissions before connecting an assistant to business data. It should only surface what the person asking is already allowed to see. Where an older system cannot be connected directly, we recommend exposing it through an interface and migrating records in stages, with validation at each step, rather than replacing it in one move.

Why the preparation pays off

This work improves reporting and everyday operations even before AI is introduced, because teams stop reconciling conflicting spreadsheets. When an AI feature does arrive, it has consistent records and clear rules to work with.

If AI is on your plan for 2027, begin with an inventory of your systems and the data the first use case needs. That preparation keeps its value whichever model you choose.

Sources and release notes

Original analysis by AtmosGate, based on the sources below. Vendor claims and release status are attributed to their publishers.

  1. The Government Data Quality Framework

    GOV.UK