Workflow Optimization Analyst
Shorten workflows without losing the safeguards that matter.
Rank workflow changes by time saved and implementation effort, with controls, risks and verification attached.
Ready-to-use prompt
The prompt
Copy it as-is, then swap the bracketed placeholders for your own details before running it.
Role: You are an operations improvement analyst redesigning an existing workflow.
Context:
- Current workflow, step by step: {{current_steps}}
- Volume per week and current elapsed time: {{volume_and_time}}
- People involved and their available hours: {{capacity}}
- Controls, approvals or compliance requirements that cannot be removed: {{controls}}
- Tools available, including anything already paid for: {{tooling}}
- What good looks like after the change: {{target}}
Task: Return four groups of changes.
1. ELIMINATE: steps that add no value to the output. For each, state what it was protecting against and why that risk is acceptable now.
2. AUTOMATE: steps that are rule based and high volume. Name the tool, the trigger, and the manual fallback when it fails.
3. RESEQUENCE: steps that can run in parallel or earlier, with the dependency that currently forces the order.
4. REASSIGN: steps sitting with the wrong role, and the queue they are stuck behind.
For every change give: estimated elapsed-time saved per run, implementation effort in hours, the control it must preserve, the new failure mode it creates, and the measurement that proves it worked.
Then return:
- A ranked sequence, highest saved-time per implementation hour first.
- Changes that must not be made together, and why.
- Anything that looks like a process problem but is actually a capacity or a tooling problem.
Rules:
- Never remove a control listed as mandatory. Propose a cheaper way to satisfy it instead.
- Do not propose a tool the team does not have unless the saving justifies the purchase, and say so explicitly.
- Reject any change whose saving you cannot express in time, money or error rate.Estimated results
Editor's note
Why this prompt matters
A workflow can contain only an hour of actual work yet take days to finish. Requests wait for an approver, missing information sends them backwards, and staff copy updates between systems because nobody trusts the handoff. Cutting the visible task time will not necessarily shorten delivery; removing the wrong approval can make delivery faster but unsafe.
This prompt turns a current-state workflow into a controlled improvement backlog. It separates work to eliminate, automate, resequence and reassign, then requires each proposal to carry a time estimate, implementation effort, preserved control, new failure mode and verification measure. That makes the output usable for an operations review rather than just a brainstorming session.
Supply {{current_steps}} with owners, handoffs, working time and waiting time. Use {{controls}} to distinguish mandatory safeguards from habits, and {{capacity}} to show actual availability. Without those distinctions, the model may recommend moving work to an equally overloaded person or automating a delay caused by missing information.
Anatomy
Prompt engineering breakdown
Role
Operations analyst, not tool salesperson.
Context
Steps, demand, capacity, controls, tools and target.
Goal
Reduce elapsed time and effort safely.
Constraints
Preserve controls; justify tools; quantify benefits.
Output format
Four groups plus ranking and conflicts.
Why this structure works
Links proposals to safeguards and tests.
What you'll get
Expected output
What you receive
Expect four change groups followed by an implementation ranking, incompatible combinations and a diagnosis of capacity or tooling constraints. Each recommendation should identify a specific step, not simply promise to “streamline approvals.” Savings should refer to end-to-end elapsed time per run, with assumptions visible.
Worked example: internal purchase requests
Consider a hypothetical team handling 30 requests weekly. Requests take two business days end to end, including 40 minutes of handling. The team already has a shared form, request register and workflow automation. Budget approval and a retained decision record are mandatory. The following figures are illustrative planning assumptions, not measured results.
| Group and change | Saved/run; setup | Preserved control; new risk; verification | |---|---|---| | ELIMINATE duplicate status tracker | 15 min; 2 h | Retain decisions in register; missing history; reconcile sampled records. | | AUTOMATE completeness checks on submission using existing workflow tool | 1 h; 4 h | Approval remains; valid requests blocked; track false rejections and cycle time. | | RESEQUENCE supplier-detail collection alongside budget review | 2 h; 4 h | No order before approval; wasted preparation; compare elapsed time and abandoned work. | | REASSIGN intake triage from manager to coordinator | 30 min; 4 h | Manager retains approval; misrouting; audit routing and coordinator backlog. |
For automation, failed or disputed checks enter a visible manual intake queue. Confirm that supplier-detail collection does not genuinely depend on approval and that the coordinator has capacity outside the manager’s review queue.
Rank resequencing first, automation second, then elimination and reassignment tied on saved-time per implementation hour. Do not add savings blindly: overlapping waits may disappear only once.
Do not remove the tracker while changing automated logging until register completeness is verified. Pilot against comparable requests, checking cycle time, handling effort, exceptions and approval evidence. A persistent approval backlog may require capacity, not another redesign.
Under the hood
Why this prompt works
The Role keeps attention on operational redesign rather than generic productivity advice. Context anchors recommendations to the actual sequence, demand, available people, protected controls, existing tools and desired outcome. Those inputs help distinguish an unnecessary handoff from a necessary safeguard with an inefficient implementation.
The Task separates four different interventions. Elimination asks whether work should exist; automation asks whether rules can execute it; resequencing tests dependencies; reassignment tests ownership and queues. Requiring effort and savings makes prioritisation explicit, while failure modes and verification stop the answer at a testable proposal rather than an unsupported promise.
The Rules establish boundaries: mandatory controls survive, new purchases need justification, and unquantifiable benefits cannot support a recommendation. The final ranking is a starting point, not automatic approval. Owners still need to validate estimates, check shared dependencies and authorise changes before rollout.
Model fit
Best AI models for this prompt
ChatGPT
Use ChatGPT to structure the backlog and calculate ranking ratios. Request consistent time units and check overlapping savings.
Claude
Use Claude to scrutinise controls and failure modes. Ask it to distinguish mandatory safeguards from inferred preferences.
Gemini
Use Gemini with supplied documentation for existing tools. Require documentation support for proposed triggers and subscription features.
Grok
Use Grok to challenge sequencing and ownership assumptions. Require workflow evidence rather than general automation claims.
When to use
- Before hiring to clear an approval queue.
- After mapping handoffs and rework.
- Before renewing underused workflow tooling.
- When piloting a shorter intake process.
When not to use
- No reliable current-state steps exist.
- Regulatory sequencing needs authorised compliance review.
- Live incidents require immediate operational judgment.
- Savings estimates lack supporting observations.
Get more from it
Pro tips
- 1
Separate touch time from waiting.
- 2
Define business versus calendar hours.
- 3
Record exception paths in {{current_steps}}.
- 4
Include availability, not just headcount.
- 5
Assign each pilot an owner.
- 6
Set rollback criteria before rollout.
Don't ship this
Common mistakes
✗ Automating redundant work.
Fix — Complete elimination review first.
✗ Treating overlapping savings as additive.
Fix — Recalculate the combined critical path.
✗ Moving a queue elsewhere.
Fix — Check receiving-role capacity.
People also ask
Frequently asked questions
Q.How do I stop it removing a control I need?
List the mandatory controls in the context block. The prompt cannot remove them and must propose a cheaper way to satisfy them instead, with the residual risk stated explicitly.
Q.Why rank by time saved per implementation hour?
Because effort is the constraint, not ideas. A change saving two hours a week for four hours of work beats one saving three hours a week for sixty, and ranking makes that comparison instead of leaving it to enthusiasm.
Q.Do I need a process map first?
It helps considerably. Without a step list the model will invent a plausible workflow and optimise that. A rough current-state map with timings produces recommendations you can actually implement.