Data readiness for AI: how SMEs prepare documents and data

Before an AI workflow can use your data, identify the current documents, remove obsolete versions and decide who maintains each source. This guide gives SME owners and operations leads five checks and a cleanup order for the sources behind one workflow.

Block Consult4 min read
Figure · Data

For most practical AI workflows, existing SharePoint folders, file servers, ERP records, inboxes and spreadsheets provide enough source material to begin. What needs checking is whether the material behind one workflow is in a state a tool can use.

Data readiness in this sense is narrower than a data strategy. It covers the documents and records that one workflow reads, and the work is done when each of them passes the five checks described below.

Identify the inputs for one workflow

Assistants and automations rarely need all company data. They need the sources used in one workflow: the price list for quoting, the SOPs for support answers or the contracts for compliance checks. Assess readiness for that workflow.

Write the list of sources with the people who do the work. For each step, note which document, record or field they look at, where it lives and who updates it. Watch for sources that only appear when someone shows you the task, such as a personal spreadsheet or an old email kept for reference. If the work depends on them, they belong on the list.

Five checks for AI data readiness

Check each source on the list against five criteria.

  • Current: the latest version is identifiable, and the outdated ones are archived.
  • Findable: the material lives in a known place with sensible names.
  • Permissioned: it is clear who may see what, so the tool can respect the same lines.
  • Extractable: text can be read from the files. Scans need OCR and tables need structure.
  • Owned: someone is responsible for keeping the source current.

Each check has a quick test. For currency, ask two colleagues to open the latest price list and see whether they open the same file. For permissions, pick a restricted document and confirm who can open it today. For extraction, copy a paragraph and a table out of a typical file and look at what arrives.

The checks apply to records as well as documents. If the workflow reads fields from an ERP or a CRM, count how many records have those fields filled in and look at when they were last updated. A field that is complete for some customers and empty for others gives uneven results.

A source that fails a check becomes a task. Record the failures in one table with four columns: the source, the failed check, the fix and the person responsible.

Clean the source documents

Start by cleaning the sources for one workflow: archive stale versions, agree on naming, fix the two or three broken export formats and record the source of truth.

Work in that order. Archiving comes first because old policies and duplicate templates produce confused answers, and a tool that finds two price lists cannot tell which one applies. Move stale files to an archive the tool does not read. Then agree a naming pattern that states the document type, the subject and the date or version, and apply it to the current files.

For exports, open a file as the system produces it and check that columns and headers survive. Finally, write down which system or folder is the source of truth for each kind of information, and who owns it.

Decide as well what the tool should not read at all. Drafts, superseded contracts and personal working files can be excluded without cleaning them, and the source owner makes that call.

The common mistake is to widen the cleanup. Once people start archiving, the whole shared drive looks like a candidate. Stay with the list for the one workflow.

Decide what can wait

A central data platform, a company-wide taxonomy and perfect master data are not prerequisites for a first working system. They may become worthwhile later, as more workflows use the data.

The sequence for a first workflow has four steps.

  1. Pick the workflow and list the documents and fields it needs.
  2. Clean and label that slice, and archive what is stale.
  3. Connect those sources to the tool with permissions intact.
  4. Assign an owner who keeps the source current after launch.

The third step needs care with sensitive data. Scope the workflow so the tool only touches what it needs, mirror your existing permissions, and agree processing terms with any external provider before connecting live data. After connecting, test with two accounts that have different access rights and confirm that each one sees only what it could open before. For companies in Switzerland, our article on AI readiness for Swiss SMEs covers data protection and data residency.

Use unanswered questions to find gaps

Wrong or unanswered questions can point to gaps in source material, as described in our guide to knowledge assistants. Review those questions and update the relevant sources.

Make this a routine for the source owner. Collect the questions the tool answered wrongly or could not answer, and sort each one by cause: a missing document, an outdated version or a file the tool could not read. Fix the source, then ask the question again.

Block Consult's AI readiness and workflow audit reviews the documents, data and software behind each candidate workflow and delivers a data and risk checklist. The article on choosing a first AI workflow shows where data readiness sits in the scoring.

Frequently asked questions

Do we need a data warehouse before starting with AI?
No. Most practical AI workflows read from existing documents and systems. Central platforms become relevant later, when several systems need the same governed data.
Our documents are PDFs and scans. Is that a problem?
Usually manageable. Digital PDFs extract well, and scans need OCR. Outdated content is a bigger risk than file format.
How do we handle sensitive data?
Scope the workflow so the tool only touches what it needs, mirror your existing permissions and agree processing terms with any external provider before connecting live data.
How much data cleanup does a first AI workflow need?
Only the sources that one workflow uses. Archive stale versions, agree on naming, fix broken export formats and record which source is the source of truth.
Contact

Let's build something durable.

Tell us about your project. We reply within two business days.