Is your information ready for AI? Here are five checks you should do before a pilot

Published

03 November, 2026

AI continues to buzz around the workplace, and it’s no surprise as more and more people uncover its capabilities. Streamlining processes, automating repetitive tasks, data analysis, just to name a few.

We now find that 1 in 3 employers now use AI in the workplace, and we can only expect that number to grow. However, how do you know if your business is ready for AI?

Short answer: Look at your information.

You’ll know that your business is ready for an AI pilot when information is relevant, reliable, lawfully usable, securely controlled and supported by clear accountability.

So, before connecting an AI tool to workplace documents, businesses should check these five areas, define a limited use case and agree how people will verify the results.

TL;DR - preparing your information for AI

Start with one task and a controlled set of information.

Check its quality, legal basis, access permissions, supplier arrangements and human oversight.

A pilot using personal data still has to meet data protection requirements.

 

1) Do you know which information your AI pilot needs?

Choose the task before choosing the documents. A pilot that helps staff find approved procedures needs a different information set from one that summarises customer correspondence.

Define the business problem, intended users and what success looks like. Then map the information required, including shared drives, email, scanned records and cloud platforms.

For each source, record:

  • Who owns it and approves its use.
  • Whether it contains personal, confidential or commercially sensitive information.
  • Why it is needed and which material should stay outside the pilot.
💡 Practical check: Can you explain why every included information source is necessary? Start with a small, approved collection rather than connecting your entire shared drive.

 

2) Is your information accurate, current and understandable?

AI can retrieve an outdated document just as confidently as a current one. Conflicting policies, duplicate files and poor scans can undermine otherwise useful results.

Identify the authoritative version of each document. Check dates, missing pages, inconsistent terminology and whether scanned text is readable. Add useful metadata, such as document owner, version, approval date and review date.

Hero Reclaim Valuable Practice Space

What do these information management terms mean?

  • Metadata: Information describing a record, such as its author, date and classification
  • ROT information: Redundant, obsolete or trivial information that may need review for removal.

Apply retention schedules before deleting anything and preserve records subject to legal holds. Exclude superseded material from the pilot where appropriate, even when you must retain it elsewhere.

💡 Practical check: Ask a subject expert to test representative documents, including tables and scans. If staff cannot identify the correct answer, the collection needs more preparation. Document scanning and indexing can help make paper records searchable.

 

3) Can you lawfully use this information for AI?

Possessing information does not automatically authorise every new use. Where personal data is involved, check the privacy and data protection laws that apply in the countries where you operate. Even in a pilot, these requirements still apply when real personal data is being collected, stored or processed. This will depend on where you are geographically (GDPR for EU, DPDP for India, PDPA for Singapore and so on) .

Document your purpose and lawful basis. Be clear about why the information is being used, whether you have the right to use it for this purpose and whether additional safeguards are needed for sensitive or personal information

Consider whether a privacy or data protection impact assessment is appropriate, particularly where the pilot involves sensitive information or could create risks for individuals. Complete it before processing starts; consult the regulator first if high residual risk cannot be reduced.

Check confidentiality commitments, copyright permissions and relevant sector rules too. For recruitment or employee assessment, assess discrimination risks consider local equality and discrimination laws.

💡 Practical check: Have your privacy or compliance lead approve the proposed data use before uploading live records.

 

4) Are permissions and supplier controls strong enough?

An AI assistant should only reveal information its user is authorised to access. Connecting it to a repository can expose existing permission problems.

Test access using different staff roles, including attempts to retrieve restricted content through summaries or indirect questions. Check the supplier’s terms and actual settings:

Check What to establish
Data use Are prompts, files or outputs used for model training?
Retention How long are uploads, logs and backups kept?
Location Where is data processed, and who can access it remotely?
Responsibilities Who acts as controller or processor, and which subcontractors are involved?
Exit Can you remove pilot data and revoke access?

 

Put the required processor contract in place where applicable. Assess international transfer requirements, including overseas access; local hosting alone does not settle every transfer question.

💡 Practical check: Confirm controls cover search indexes and generated outputs as well as original documents.

 

5) Can people verify results and stop the pilot?

 

Agree your testing and accountability before launch. AI outputs can be inaccurate even when the source information is sound.

  1. Name a business owner and human reviewers.
  2. Test known questions against approved answers and source documents.
  3. Set acceptable error thresholds and check for biased outcomes.
  4. Record problems and define when to pause the pilot.
  5. Agree what happens to inputs, outputs and logs afterwards.

Give reviewers time, training and authority to challenge outputs. A rubber stamp is not meaningful human involvement. Significant solely automated decisions about people require statutory safeguards, including information, representations, human intervention and contestability. Special category data attracts additional restrictions.

Practical check: Can someone trace an answer to its source, correct it and stop an unsafe workflow?

 

Summary

AI can be a useful tool, but it requires your information to be in order first. Making sure your information is free of ROT data, is accurate, and accessible to the right people is a good place to start.

It’s key that piloting AI must also remain compliant regardless of where you are in the world, particularly when handling personal data, and that the right authorisations are in place.

At Crown Information Management, we can support businesses in getting their information in order with our Information Consulting service. With increasing regulatory pressures and ever-expanding digital footprints, businesses must manage information intelligently. Our consulting services help you map your organisation’s complete data, keeping your business compliant.

Talk to a member of our team today to discuss your information today!

No matter what stage of your information journey you're at, we're here to make work uncomplicated.

Contact Us

FAQs - Preparing your information for an AI pilot

Yes. If a pilot uses personal data, relevant privacy and data protection requirements still apply. Calling something a pilot doesn’t remove those responsibilities
No. Start with a defined collection that meets the pilot’s needs. Resolve its quality, permissions and retention issues before expanding.
Not by itself. Check whether individuals could still be identified, including through other available information. Confidentiality, security and quality checks remain relevant even when data is genuinely anonymous.
Before investing in wider AI adoption, establish whether your information can support the task. Explore Crown Information Management’s digital solutions to help organise and digitise the records your business relies on.

Share this article