A completed remodeling job leaves behind meeting notes, revised selections, delivery updates, and questions someone eventually answered. When the team sits down to discuss what to improve, those records can be hard to reconstruct from memory.

Datum recommends using AI to prepare an evidence-backed agenda for that review. Ask it to assemble the sequence of events and identify questions for the people who did the work. Keep suspected causes separate from documented facts, then let the team choose a change to test on a future job.

Choose one handoff to examine

Start with a specific question, such as why a fixture decision reached purchasing later than expected. Name the job, the review period, and the records available. Include examples of what went smoothly as well as the events that required extra work. Otherwise, the review may describe only the problems people remembered to document.

Anthropic’s context-engineering guidance recommends selecting relevant information for the task rather than filling the model’s context indiscriminately. For this review, Datum’s application is a focused packet: the decision record, relevant messages, the issued order, and the schedule revisions surrounding the handoff. Keep references to the originals so a reviewer can open them.

Use an AI environment approved for those project records. Remove unrelated personal details and give the review team only the material it is authorized to see. A whole company mailbox is a poor starting packet for a question about one purchasing handoff.

Reconstruct the sequence without assigning blame

For each event, ask for the recorded date, what happened, and the source passage. Preserve the difference between when an event occurred and when someone wrote about it. A message entered on Friday may describe a conversation from Wednesday.

  • Documented event: what the supplied record explicitly establishes.
  • Unresolved conflict: where records disagree or a revision is missing.
  • Possible explanation: an interpretation the team still needs to confirm.
  • Review question: what evidence or participant account could settle the uncertainty.

Do not treat a missing message as proof that nobody communicated. State that the supplied records do not show the handoff. Likewise, an event happening before another event does not establish that it caused the delay. Preserve competing explanations when the packet cannot resolve them.

A fixture delay can have several explanations

Consider a hypothetical bathroom project. A meeting note says the client discussed a different faucet finish. A later purchase order uses the original finish, and a replacement order follows. AI might summarize this as purchasing ignoring a client change.

The records described do not establish that conclusion. The discussion may never have become an approved selection. An approval may exist outside the packet. Purchasing may have received a superseded log. A useful review agenda links the note and both orders, then asks which selection was approved and what purchasing received before ordering.

Have the participants confirm or correct that sequence. Record their explanation as a dated review statement, with its author, rather than rewriting the original job history. If the cause remains uncertain, keep it open. The meeting can still identify a record that would make the next handoff easier to verify.

Leave with a change someone can check

If the team confirms that purchasing could not identify the approved selection, a proposed change might be to attach the approval record to the order request. Name who will maintain that record, when the check happens, and how the next project manager will know it was completed. Avoid adding a general instruction to communicate better.

Anthropic’s evaluation guidance distinguishes an AI system’s account of its work from the actual outcome and recommends judging against defined success criteria. Applied here, a readable review draft is only part of the test. Your team must also check whether its event sequence is supported and whether the proposed action addresses the confirmed problem.

Before using this preparation method on a live review, try a completed job whose history the team knows. Ask reviewers to record unsupported causes, missed events, incorrect dates, and time spent correcting the draft. Repeat the trial with a different job before deciding whether it saves useful preparation time.

Keep the approved action with the review record and revisit it after the next relevant handoff. Record whether the change was used and what happened. One successful handoff is evidence to discuss, not proof that every future job will improve.

Bring a recurring handoff problem to a conversation with Datum. We can discuss how your own people could build a review process around the records they already keep.

Sources Read

Next step, if this note maps to a problem on your desk: book a conversation with Adam. Thirty minutes, weekday evenings, Eastern time.

Related Field Notes