Process Optimization Skill
Use this skill when a process feels slow, when you are investigating inefficiencies, or when preparing a process-improvement initiative. It maps the current state, surfaces bottlenecks and waste, and delivers measurable, prioritised improvements.
---name: process-optimization-skilldescription: Analyse processes for bottlenecks, waste, and improvement opportunities with throughput measurement and prioritised recommendations. Use when a process feels slow, investigating inefficiencies, or preparing a process improvement initiative. Trigger with "optimise this process", "find bottlenecks in", "how can we improve".---
# Process Optimization Skill
Analyse processes for bottlenecks and waste, then recommend measurable improvements with prioritised roadmaps. Process data comes from you, a connected MCP source, or companyRAG collections.
## Where the Data Comes From
| Source | What It Adds || --- | --- || **Project / issue tracker via MCP** (e.g. Jira, Asana, Monday) | Processing and elapsed times, status dwell times, volumes, send-backs || **Workflow / CRM system via MCP** (e.g. HubSpot, Salesforce, ServiceNow) | Approval chains, handoffs, SLA data, error and rework rates || **companyRAG collections / file upload** | Process documentation, runbooks, audit logs, time and cost exports |
> **No connected source?** Provide the data in chat or upload the relevant files — the skill works the same way.
## Process Analysis Methodology
### Current State Mapping
Before optimising, map the process as it actually operates — not as it is documented or intended to operate:
1. Walk the process end-to-end: Document every step, decision point, handoff, wait state, and rework loop as they actually occur.2. Identify participants: Who touches the process at each stage — roles, not just departments.3. Capture timing: For each step, record processing time (active work) and elapsed time (calendar time including waits). The gap between the two reveals queue time and delays.4. Note volumes: How many items flow through the process per period — total and by variant.5. Mark pain points: Where do participants report frustration, errors, or workarounds? These are optimisation signals.
### Throughput Measurement
Establish baseline metrics before recommending changes:
```PROCESS BASELINE: Process: [name] Measurement period: [date range] Volume: [items processed per period] Cycle time: [average end-to-end elapsed time] Processing time: [average active work time per item] Process efficiency: [processing time ÷ cycle time × 100]% Error/rework rate: [% of items requiring rework or correction] First-pass yield: [% of items completing correctly on first pass] Cost per item: [total process cost ÷ volume, if data available]```
Process efficiency ratio reveals how much time is spent on value-adding work versus waiting. A process with 2 hours of active work time and 5 days of cycle time has a process efficiency of approximately 5% — most time is waiting, not working.
## Waste Identification
### Eight Wastes Framework
Adapted from lean methodology for knowledge work and business processes:
| Waste Type | Definition | Process Indicators || --- | --- | --- || **Waiting** | Idle time between steps — queues, approvals, information requests | Items sitting in inboxes, approval backlogs, "waiting for [person]" status || **Overprocessing** | Doing more than needed — unnecessary detail, redundant checks, gold-plating | Reports nobody reads, approvals that never reject, data collected but unused || **Rework** | Correcting errors from previous steps | High rejection rates, frequent revision cycles, "send back" patterns || **Handoff friction** | Information lost or degraded during transfers between people/teams | Repeated questions, context re-gathering, "I didn't know that" moments || **Motion** | Unnecessary movement between tools, systems, or formats | Copy-pasting between systems, manual data entry, format conversions || **Overproduction** | Producing more output or more frequently than the downstream consumer needs | Batch sizes larger than needed, reports generated more frequently than consumed || **Inventory** | Work items accumulating between steps — work-in-progress buildup | Growing backlogs, aging items, items started but not finished || **Unused talent** | People performing work below their capability, or expertise not leveraged | Senior staff on routine tasks, specialist knowledge not consulted |
### Waste Assessment
For each identified waste:
```WASTE ITEM: Type: [from eight wastes] Location: [which step or between which steps] Description: [what specifically is happening] Frequency: [how often — every item, occasionally, specific conditions] Impact: [time, cost, or quality effect per occurrence] Root cause: [why this waste exists — not just what, but why] Evidence: [data supporting the assessment — times, volumes, error rates]```
## Bottleneck Detection
### Identification Methods
A bottleneck is the step that constrains the throughput of the entire process. The process can only move as fast as its slowest step.
**Method 1 — Queue observation**: The step with the longest queue upstream is likely the bottleneck. Work accumulates before the constraint.
**Method 2 — Utilisation analysis**: The step where resources are most heavily utilised (closest to 100%) is the constraint. Other steps have idle capacity.
**Method 3 — Flow rate comparison**: Measure throughput at each step independently. The step with the lowest throughput rate limits the process.
### Bottleneck Classification
| Type | Cause | Example | Resolution Approach || --- | --- | --- | --- || **Resource bottleneck** | Not enough people, tools, or capacity at the step | Single approver for all requests | Add capacity, redistribute, or reduce demand || **Policy bottleneck** | Rules or requirements that constrain flow unnecessarily | Approval required for items below a reasonable threshold | Challenge the policy — does it still serve its purpose? || **Dependency bottleneck** | Step blocked by external input or upstream delivery | Waiting for third-party data before proceeding | Decouple, parallelize, or pre-fetch the dependency || **Batching bottleneck** | Work held until a batch accumulates before processing | Monthly invoice processing when weekly would flow better | Reduce batch size or process continuously || **Knowledge bottleneck** | Only specific people have the expertise to perform the step | Single subject-matter expert required for every review | Cross-train, document, or redesign to reduce specialisation requirement |
## Improvement Identification
### Improvement Categories
| Category | Description | Typical Impact | Typical Effort || --- | --- | --- | --- || **Eliminate** | Remove the step entirely — it adds no value | High | Low (often just a decision) || **Automate** | Replace manual work with system-driven execution | High | Medium–High || **Simplify** | Reduce complexity — fewer fields, simpler rules, shorter forms | Medium | Low–Medium || **Parallelize** | Run sequential steps simultaneously where no true dependency exists | Medium | Low–Medium || **Reduce batch size** | Process smaller batches more frequently for faster flow | Medium | Low || **Standardise** | Align variations into a single consistent approach | Medium | Medium || **Relocate** | Move the step earlier or later in the process for better flow | Low–Medium | Low |
### Five Whys Root Cause Analysis
For each significant waste or bottleneck, apply root cause analysis before recommending solutions:
```PROBLEM: [symptom observed] Why 1: [first-level cause] Why 2: [cause of the first-level cause] Why 3: [deeper cause] Why 4: [even deeper cause] Why 5: [root cause — the systemic issue to address] Root cause: [summary — this is what needs to change]```
Stop when you reach a cause that is actionable and systemic. Not every chain goes to five levels; stop when you reach the actionable root.
## Impact Estimation
### Before/After Projection
For each recommended improvement, project the impact:
```IMPROVEMENT PROJECTION: Improvement: [specific change] Addresses: [which waste/bottleneck] Current metric: [baseline from throughput measurement] Projected metric: [expected value after improvement] Improvement: [delta and percentage change] Confidence: [High / Medium / Low — based on data quality] Assumptions: [what must be true for this projection to hold] Measurement method: [how you will verify the improvement after implementation]```
Mark all projections as [AI projection — verify with process data]. Never present projections as certainties.
### Prioritisation Matrix
Rank improvements using impact vs. effort:
| | High Impact | Medium Impact | Low Impact || --- | --- | --- | --- || **Low Effort** | **Do first** — quick wins | **Do second** — easy value | **Consider** — marginal gains || **Medium Effort** | **Plan** — significant projects | **Evaluate** — weigh cost/benefit | **Defer** — limited return || **High Effort** | **Strategise** — major investment | **Deprioritise** — poor ratio | **Avoid** — minimal return for high cost |
## Output Template
```# Process Optimisation Report — [Process Name]
## Executive Summary- Current performance: [key baseline metrics]- Primary bottleneck: [the main constraint]- Top waste category: [biggest source of waste]- Recommended improvements: [count and projected total impact]- Quick wins: [improvements achievable within 2 weeks]
## Current State Analysis- Process map: [step-by-step current state with timing]- Baseline metrics: [throughput measurement results]- Process efficiency: [ratio]
## Waste Assessment[Waste items with type, location, frequency, impact, root cause]
## Bottleneck Analysis[Identified bottlenecks with classification and evidence]
## Improvement Recommendations[Prioritised list with impact projections]
## Implementation Roadmap- Phase 1 (Quick wins): [low-effort improvements to implement immediately]- Phase 2 (Planned): [medium-effort improvements with timeline]- Phase 3 (Strategic): [high-effort improvements requiring investment decision]
## Success Metrics[How to measure whether the optimisation achieved its goals — re-measure baseline metrics post-implementation]
## Assumptions and Limitations[What data was available, what was estimated, what requires validation]```
## Guardrails
- Never generate cycle times, throughput rates, or cost figures from training data. All metrics come from the user's process data.- Never claim benchmarks. "Industry benchmark" and "best-in-class" are meaningless without the user's specific context.- Never present projections as guaranteed outcomes. All projections are estimates requiring post-implementation measurement.- Label generated content: `[From process data]`, `[Framework methodology]`, `[AI projection — verify with process data]`.
> **Tip:** Ask for DOCX, XLSX, or Markdown output via companyFILES to get a formatted optimisation report ready to share.