COMPEL Specialization — AITE-VDT: AI Value & Analytics Expert Template 5 of 5

This template produces a portfolio-level AI scorecard with two artifacts: (1) a spreadsheet-form scorecard for human consumption, and (2) a platform-neutral JSON schema for BI-dashboard implementation. The spreadsheet feeds the monthly or quarterly steering-committee meeting; the dashboard supports self-service drill-through for analysts.

Portfolio Scorecard: [Program Name]

Program: [Name, scope] Reporting period: [Quarter / Month] Scorecard date: [Date] Prepared by: [AI value lead] Reviewed by: [FinOps lead, Program office] Scorecard version: [Major.minor]

1. Portfolio overview (one page, top of scorecard)

Status distribution: [🟢 Green × N | 🟡 Yellow × N | 🔴 Red × N]

Top three findings:

1. [Finding 1]
2. [Finding 2]
3. [Finding 3]

Pending decisions this period: [List]

2. Feature scorecard (one row per feature, sorted by status then realized value descending)

Maximum 15 features per scorecard. If the portfolio has more, nest into program-level scorecards.

3. Attribution governance footnote

Portfolio primary attribution model: [Shapley / Linear / Last-touch / etc.]

Features using non-primary attribution:

• [Feature X] uses [Model]. Realized-value reported here is under the feature’s native model; Shapley re-analysis is [in-flight / planned for Q_].
• […]

Aggregate caveat: Aggregate realized-value is the sum across features’ native attribution models. Under uniform Shapley attribution, the aggregate is estimated at [$] ± [$]. See [Reference] for the attribution-harmonization workstream.

4. One-page executive narrative (accompanies scorecard)

Portfolio health

[Two to three sentences summarizing the aggregate: count by status, realized value vs. target, significant movement since last period.]

Reds and pending decisions

For each red feature, one short paragraph covering what’s wrong, what’s the recommended action, and when the decision is expected.

• [Red feature 1]: [Three sentences]
• [Red feature 2]: [Three sentences]

Attribution note

[One paragraph on the mixed-attribution situation and in-flight remediation.]

Looking forward

[Two to three sentences on next-quarter expected decisions, anticipated portfolio shifts, emerging risks.]

5. Platform-neutral JSON schema (for BI implementation)

{
  "portfolio": {
    "name": "[Program name]",
    "period": "[YYYY-Q#]",
    "prepared_by": "[Name]",
    "aggregate": {
      "feature_count": 0,
      "green_count": 0,
      "yellow_count": 0,
      "red_count": 0,
      "realized_value_total": 0,
      "investment_to_date_total": 0,
      "payback_ratio": 0.0,
      "primary_attribution_model": "shapley"
    },
    "features": [
      {
        "id": "F001",
        "name": "[Feature name]",
        "stage": "evaluate",
        "status": "green",
        "realized_value_qtd": 0,
        "realized_value_cumulative": 0,
        "investment_to_date": 0,
        "payback_ratio": 0.0,
        "primary_risk": "[Specific risk statement]",
        "risk_probability": "M",
        "risk_impact": "H",
        "next_decision": "[Decision description]",
        "next_decision_date": "YYYY-MM-DD",
        "owner_role": "[Role]",
        "attribution_model": "shapley",
        "notes": "[Any notes]"
      }
    ],
    "pending_decisions": [
      {
        "feature_id": "F001",
        "decision": "[Description]",
        "date": "YYYY-MM-DD",
        "decision_maker_role": "[Role]"
      }
    ]
  }
}

Implementation guidance per BI platform

• Power BI: Import the JSON via Power Query; create a table visualization with conditional formatting on status; drill-through to feature-level VRRs.
• Tableau: Use JSON data source; build a scorecard view with a shape or color mark on status; action filters for drill-through.
• Looker: Define a LookML model with feature as primary table; dashboard with tile-per-row and conditional formatting.
• Metabase: Import JSON as a question; dashboard with single-cell visualizations per row; dashboard filters for stage and status.
• Superset: Native SQL-backed dashboard with heatmap or table visualization; filter boxes for interactive filtering.
• Grafana: Stat panel per feature with threshold-based color coding; organized into dashboard rows.

The same JSON schema feeds all six platforms; the implementation is BI-tool-specific but the underlying data contract is unified.

6. Scorecard preparation workflow

Week 1 — Data refresh

• Each feature lead submits current realized value, investment, primary risk, next decision to the AI program office by [Date].
• Data must come from the same source as feature-level VRRs.
• Discrepancies are flagged for resolution before aggregation.

Week 2 — Status calibration

• AI program office reviews each submission against status-calibration rules.
• Calibration meetings resolve any disputed status assignments.
• Attribution-model compliance is checked; non-primary usage is flagged.

Week 3 — Aggregation

• Scorecard assembled per the structure above.
• Aggregate totals computed.
• Attribution-model footnote and caveats drafted.
• Narrative drafted.

Week 4 — Review and distribution

• Joint sign-off by AI program office and FinOps lead.
• Pre-brief for steering committee chair.
• Distribution to steering committee with the one-page narrative.

7. Failure-mode checklist

Before distributing, verify the scorecard does not exhibit:

• All-green bias (every feature green; unlikely in a healthy portfolio)
• Feature inflation (more than 15 features; nest if needed)
• Realized-value inconsistency (different attribution models without disclosure)
• Investment-to-date ambiguity (different definitions across features)
• Risk-column dilution (boilerplate risks; rewrite to be specific)
• Decision-column absence (features with no open decision; question their presence)

Appendix A — Status-calibration rules

Appendix B — Linkage to other artifacts

• Individual feature VRRs: [References]
• Attribution-harmonization workstream: [Reference]
• Compute-budget portfolio view: [Reference]
• Board-grade quarterly summary: [Reference]