Questions about AI at work
18 articles drawn from what comes up in our work with teams. Each one explains a way of thinking about the problem. All are free to read, with no registration.
Training packs and exercises for your own team are supplied as part of an agreed engagement. Talk to us.
Showing all questions
Claude or Copilot: which one for this job?
The answer depends on the task, where the information lives and what your organisation has approved. Compare the tools on a real job, not on a feature list.
- Claude or Microsoft 365 Copilot? Choose by the work you need doneCompare document analysis, presentations, Excel and company knowledge before adding another subscription.
- Which AI should you use to analyse business documents?Claude and Copilot Chat both spotted an unapproved invoice charge in our short document-analysis test. See the results, wording that needed checking and limits.
- Compare AI tools using the same job and review standardUse the same permitted task pack and acceptance checks for each tool. Record the actual setup and the effort needed to produce a checked working output, then describe the limits of what the small test establishes.
Why does my Copilot presentation look generic?
A generic result can reflect an unclear brief, missing source material or the available layouts and features. These articles explain how to identify the problem.
- Why does my Copilot presentation look so generic?Separate a weak brief, a poor source document and a template problem before blaming the whole product.
- Copilot training: what should change in the working day?A useful Copilot programme should leave someone able to complete a specific job, check the result and repeat it without the trainer. Agree that outcome before choosing the course or buying more licences.
- Copilot gave you a formula. Did the workbook actually change?A suggested formula is advice. A changed cell is an edit. A reliable result needs another step: check that the edit is correct, that dependent figures still work and that the saved file contains the change.
- Check AI meeting minutes before a suggestion becomes a decisionCompare each recorded decision and action with the meeting evidence. Preserve proposals, conditions and unresolved questions as such, then ask the meeting owner to resolve anything the record cannot establish.
What does AI need to know about our work?
General knowledge does not tell an assistant how your organisation works. Give it relevant background, current sources and clear instructions.
- What context does AI need to do useful work?Instructions, files, memory and search each supply something different. Here is how to tell what your assistant actually knows about the task.
- Revisit an AI answer when a new attachment arrivesTreat new information as a reason to review the earlier conclusion. Identify which assumptions it changes, check the affected claims against their sources, and update every working output that depends on those claims.
- Give an AI assistant working context your team can maintainKeep reusable background in a dated note with an owner and a review trigger. Supply current task evidence separately, and test whether the assistant applies the context without turning old examples into current facts.
Why can't it find the right company information?
Access, search coverage and conflicting versions can each affect the answer. Understand what information the assistant can use and which source should govern.
- Why can't Copilot find the right company information?A missing answer can come from access, search scope, indexing or the information itself. Diagnose the failure before buying another tool.
- How much knowledge management do you need before using AI?Start with the information needed for a worthwhile job. Resolve ownership, current versions and missing know-how without waiting for a perfect intranet.
- AI in construction: check the evidence before drafting the tenderUse AI to organise requirements and locate relevant evidence before asking it to draft a tender response. The useful first output is a record of what you can support, what is missing and who must resolve it.
- AI document review: can you trace the answer back to the files?To check an AI document review, trace each material conclusion to the exact source passage and test whether that passage supports it. Check missing and conflicting material too: an accurate summary of the wrong set of files can still mislead.
- Keep unreadable information visible when reviewing scanned recordsInspect the source image before relying on extracted text. Record unreadable fields and missing pages explicitly, and keep conclusions that depend on them open until a person resolves the evidence gap.
How do people keep using it after the training day?
Attendance is the start. What matters is whether someone can do a real job again next week without the trainer in the room.
- Prepare a training session when AI access differs across the teamPrepare around what participants can actually open and complete. Check the target task in their approved setup, identify access gaps before the session and provide a fallback that preserves the learning objective.
- Take an AI draft into the place where the work happensDefine the destination and acceptance check before drafting. After moving the content, open the actual saved output, inspect what changed and confirm that the next person can find the correct version with the right approval status.
- Review what happens on the second independent attemptReview the next task completed without live coaching. Compare the accepted output with the intended result, record where help was needed and revise the working instructions around the specific exception encountered.
These articles explain our approach. They do not reproduce client documents, complete training exercises or reports of measured client outcomes.
For recording AI use, responsibilities and review decisions, see AI Register, also from Acuity.