Purpose and fit
Use this when several AI ideas are competing for attention. Compare tightly defined tasks, such as drafting an internal report, rather than broad ambitions such as “transform operations”. Agree the comparison criteria before scoring.
Before you start
Start with the initiative notes, backlog items or process observations you already have. Supply the candidate task names, affected users and any available evidence with its source and date. Adapt the comparison fields to your actual records; if effort, frequency or cost is not recorded, leave it unknown rather than creating numbers.
Use only information you are authorised to share with the chosen AI tool. Adjust the prompt to the fields and context you actually have; a specially created dataset is not required.
Required inputs
- A short list of candidate tasks, their current process and affected users
- Available baseline evidence: task frequency, effort, rework, delays or quality problems
- The data and systems each idea would need
- A sponsor, implementation constraints and any existing approval requirements
Practical workflow
- Write one sentence for each candidate: the problem, the affected person and the proposed improvement. Keep the scope narrow enough to test.
- Record the baseline and mark each input as measured, estimated or unknown. Include where it came from.
- Rate value, evidence strength and feasibility separately using the anchors below. Write a reason next to every rating.
- Review exposure separately: sensitive data, external actions, consequential decisions and failure recovery. Mark unresolved requirements as a gate to resolve. A high value rating does not cancel a blocked gate.
- Compare the candidates and choose one small next step: investigate, design a bounded pilot, defer or reject. Record why the sponsor chose it and when to revisit the decision.
Expected output
Expect a draft comparison and questions for a sponsor, not a proven ranking, business case or savings forecast.
One comparison sheet and a short decision log. Suggested fields:
Candidate | problem and beneficiary | baseline/source | expected change | value | evidence | feasibility | risk/dependency gates | next step | owner | review date
Suggested rating anchors, to adapt before use:
- Value: 1 = modest/local benefit; 2 = meaningful recurring benefit; 3 = important operational benefit. Describe the benefit in words alongside the rating.
- Evidence: 1 = assumption; 2 = partial observations; 3 = relevant baseline observations. “3” is not proof the proposed solution works.
- Feasibility: 1 = major unknowns; 2 = manageable dependencies; 3 = a bounded test can be prepared with available skills and authorised access.
Human checks
The affected teams validate the baseline, the system/data owners check the dependencies, and the sponsor approves the next step. Keep unknowns visible. Challenge attractive estimates with high integration, review or support costs. Avoid adding the three ratings into a universal score unless the sponsor has explicitly chosen and explained that method.
Limitations
The sheet supports prioritisation. It cannot establish actual savings, resolve security concerns or substitute for a tested business case. Ratings reflect the chosen context and can change when evidence improves.
Starter prompt
Help me prepare an AI initiative comparison for [team/context]. Use only the candidate notes and approved evidence below: [insert sanitised notes].
For each candidate, state the problem, beneficiary, baseline with source, proposed change, evidence gaps, effort/dependencies and possible failure consequences. Keep value, evidence strength and feasibility separate. Use the supplied rating anchors only where the notes support a rating; otherwise write “unknown”.
List unresolved data, access, approval and recovery requirements as gates. Do not offset a blocked gate with potential value. End with the questions the business sponsor needs answered before selecting a next step. Do not invent savings, costs or approvals, and do not choose on the sponsor’s behalf.
Return one row per supplied candidate, with a source or explicit unknown for every material baseline and a reason for every supported rating. Preserve conflicting estimates with their sources. End with unresolved gates and sponsor questions; leave the final selection to the sponsor.
Before drafting, ask a focused question if a missing detail materially changes the task or acceptance criteria. Otherwise produce a clearly limited draft and list the gaps. Label any non-critical assumptions; never invent facts, figures, sources, approvals or completed actions.
Automatic copying isn’t available. Press Ctrl+C or ⌘C to copy the selected prompt, or use your device’s copy command.
Data and AI output disclaimer
Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.