What does CIOB’s AI Playbook mean for your construction business?

01 / THE PROBLEM

Start with a recurring problem

Choose a task your team understands well enough to explain what a good result looks like.

For example, an estimating team might spend hours reviewing tender addenda and identifying changes that need attention. A project team might repeatedly assemble progress reports from information held in several places.

Before introducing AI, establish why the work takes so long.

• Are people searching for documents or working with incomplete information?
• Does the same material get entered into several systems?
• Are responsibilities unclear?

A clearer template, better document control or a straightforward integration might solve part of the problem. Once you understand what remains, you can assess whether AI has a useful role.

02 / A DEFINED TASK

Give it a defined job

“Use AI in estimating” is too broad to evaluate. A more useful proposal would be to compare two approved versions of a tender document and prepare a draft register of possible changes, with references to the relevant passages, for an estimator to review.

That defines the input, the output and the person responsible for checking it. Preparing a comparison register is a different task from deciding the commercial significance of a change or approving the tender submission.

CIOB’s use case template provides a useful structure for recording the business objective, information needed, expected benefits, risks and measures of success.

03 / PROJECT INFORMATION

Check the project information

For a document review task, the starting question is whether the system has the right documents. Confirm which revisions are current, which files have been superseded and whether the information set is complete.

Make the expected handling of missing or conflicting information explicit. If a subcontractor quotation excludes an item, the output should identify that exclusion. If a document is unavailable, the system should flag the gap.

The same principle applies to reporting. A draft progress report should distinguish recorded facts from assumptions and information that still needs confirmation.

You can start with a controlled set of documents for one task. Use that test to identify the information improvements the workflow actually needs.

04 / INFORMATION CONTROLS

Protect commercial and personal information

Tender returns, subcontractor quotations, client correspondence and project records can contain information that needs careful handling. Before uploading it, establish what the selected service does with the information and whether that use is permitted.

Check access permissions, retention, deletion arrangements and whether inputs or outputs may be used to train models. These arrangements can differ between products, account types and settings. Also check relevant client and contractual restrictions.

Where personal information is involved, include data protection in the assessment. The ICO publishes guidance on applying UK data protection principles to AI systems, alongside an AI risk toolkit. Read the ICO’s AI guidance.

For an initial test, use information you are authorised to process in the chosen environment.

Our guide to where your data goes when you use AI sets out the questions to ask your provider.

05 / MEASUREMENT

Measure the complete workflow

A faster first draft only creates value if the finished work is better or takes less effort overall. Include the time spent preparing inputs, checking outputs, correcting mistakes and dealing with exceptions.

Include difficult examples in the trial, such as a poorly scanned document or an important change buried in an appendix. Use the evidence to decide whether to continue, change the approach or stop.

For a fuller commercial assessment, see how to calculate the ROI of AI.

06 / RESPONSIBILITY

Match controls to consequences

The review needed for an internal meeting summary will differ from the review needed for information that influences site work, contractual commitments or safety decisions.

Name the person responsible for checking the output. Give them access to the source information and enough time to review it properly.

Where AI affects workplace health and safety in Great Britain, HSE’s regulatory approach explains that existing health and safety law applies. Relevant uses require risk assessment and appropriate controls, including consideration of cyber security threats.

Northern Ireland has its own health and safety legislation, including employer risk assessment duties. Businesses working there should assess their arrangements against that framework.

In practice, approval should cover the specific task, information and consequences involved.

07 / EVIDENCE TO EXPAND

Expand when the evidence supports it

A successful trial gives you a reason to consider the next step. Document what was tested, the results, the limitations and the conditions under which the workflow can be used.

Train the people who will operate it and those who will review its outputs. Assign an owner who can respond when documents, software or project requirements change.

Keep a workable fallback and review performance as use expands. A process tested on one project may need adjustment for a different client, document set or contract.

This follows our BUILD™ approach: Baseline, Understand, Implement, Learn and Develop. Establish the problem, test a focused improvement and use the evidence to guide what happens next.

One recurring task.
A defined reviewer.
An outcome you can measure.