Cross-Team Reuse Detector

Overview

Identify reusable AI patterns across business units by extracting common intents, shared prompt primitives, and platform candidates from multiple workflow decompositions.

Core principle: Teams describe workflows in different terminology, but underlying intents are often identical. Find the intent, and you find the reuse opportunity.

The Analysis Process

digraph reuse_detection {
    rankdir=TB;
    node [shape=box];

    input [label="1. Normalize Workflows\n(same format)"];
    intent [label="2. Extract Intents\n(WHY, not WHAT)"];
    cluster [label="3. Cluster Similar Intents\n(across teams)"];
    prompt [label="4. Identify Prompt Primitives\n(reusable atoms)"];
    classify [label="5. Classify Reuse Level\n(platform/config/unique)"];
    negative [label="6. Flag Negative Patterns\n(what NOT to share)"];
    prioritize [label="7. Prioritize by Value\n(ROI ranking)"];

    input -> intent -> cluster -> prompt -> classify -> negative -> prioritize;
}

Step 1: Normalize Workflows

Ensure all inputs use consistent format. Key fields to normalize:

• •Step name
• •Step type (decision/action/validation/human_task)
• •Inputs/outputs
• •Judgment criteria (if any)

Step 2: Extract Intents

For each step, ask: WHY does this exist?

Intent is more abstract than action. Same intent can have different implementations across teams.

Step 3: Cluster Similar Intents

Group intents from different workflows that serve the same purpose:

INTENT: "Detect discrepancies before they propagate"
├── Equity: "Match confirmation to internal record"
├── Fixed Income: "Categorize exception type"
└── Derivatives: "Compare internal vs counterparty calculations"

INTENT: "Prevent items from aging without attention"
├── Equity: "Escalate aged breaks >24h or >$1M"
├── Fixed Income: "Escalate SLA breach >4h or >$5M"
└── Derivatives: "Escalate material disputes >3d or >$500K"

Step 4: Identify Prompt Primitives

For each intent cluster, extract the actual prompts that would power it:

Prompt primitives are the reusable atoms. Same primitive, different variables per team.

Step 5: Classify Reuse Level

For each pattern, determine reuse level:

If a pattern requires CUSTOMIZED or FORKED level, question whether sharing is worth it.

Step 6: Flag Negative Patterns

Explicitly identify what should NOT be shared:

Sharing is not always good. Some duplication is healthy.

Step 7: Prioritize by Reuse Value

Score each pattern:

Prioritization formula:

Reuse Value = (Teams × Frequency × Pain) / (Complexity + Maintenance)

Output Format

analysis:
  workflows_analyzed:
    - name: [Workflow 1]
      business_unit: [Unit]
    - name: [Workflow 2]
      business_unit: [Unit]

intent_clusters:
  - intent: "[Abstract intent statement]"
    occurrences:
      - workflow: [Name]
        step: [Step that manifests this intent]
      - workflow: [Name]
        step: [Step that manifests this intent]

prompt_primitives:
  - id: PRIM-001
    intent: "[Which intent this serves]"
    prompt_template: "[Actual prompt with {variables}]"
    variables:
      - name: [var_name]
        description: "[What this variable represents]"
        team_specific: [yes/no]

reuse_candidates:
  - pattern: "[Pattern name]"
    reuse_level: [IDENTICAL|CONFIGURED|PARAMETERIZED|CUSTOMIZED|UNIQUE]
    teams_benefited: [count]
    prompt_primitives: [PRIM-001, PRIM-002]
    implementation_notes: "[How to build]"

negative_patterns:
  - pattern: "[Pattern name]"
    reason_not_to_share: "[Why sharing is bad here]"
    recommendation: "[What to do instead]"

prioritized_recommendations:
  - rank: 1
    pattern: "[Pattern name]"
    reuse_value_score: [number]
    rationale: "[Why this ranks highest]"
    quick_win: [yes/no]

Common Mistakes

Red Flags in Your Analysis

If your output has these, you're not done:

• •Intent clusters without prompt primitives
• •All patterns marked as "CONFIGURED" (too optimistic)
• •No negative patterns identified
• •No prioritization or all ranked equally
• •"70% reusable" without methodology
• •"UNKNOWN criteria" in multiple workflows treated as ready to automate

Special note on UNKNOWN criteria: When multiple workflows have human judgment steps with undocumented decision criteria, this is a knowledge capture opportunity, not an automation opportunity. Flag these for SME interviews before designing AI solutions.

Governance Considerations

For each platform candidate, note:

• •Owner: Which team maintains the shared component?
• •Change process: How do teams request changes?
• •SLA: What's the availability/latency commitment?
• •Escape hatch: Can a team fork if shared doesn't meet needs?

Financial Services Context

Enterprise mandates to share use cases across teams require:

1. •A catalog of reusable patterns (this skill's output)
2. •A platform team to build/maintain shared components
3. •Clear ownership so shared doesn't mean abandoned
4. •Measurable reuse to justify platform investment

The output of this analysis directly feeds the business case for enterprise AI platform investment.