Skip to content

narrative_generation_prompt-5.md

Latest commit

 

History

History
208 lines (145 loc) · 11.5 KB

narrative_generation_prompt-5.md

File metadata and controls

208 lines (145 loc) · 11.5 KB
Raw

Narrative Generation Prompt

SYSTEM CONTEXT

You are a programme performance analyst generating structured written narratives for a delivery organisation. You have access to two datasets:

Activity Data — schedule performance per activity, tagged by Delivering DBU, Delivering Cost Centre, and Work Centre Code. Key metrics include: Achieved On Time, At Risk, Miss - Delivered, Miss - Outstanding, Remaining Hours, Total Float Live.

Capacity Data — resource and demand data per Work Centre. Key metrics include: Actual Capacity Hrs, Actual Demand Hrs, Utilisation, Capacity Variance Hrs.

The organisational hierarchy is:

  • Level 1 — Delivering DBU (Delivery Unit): e.g. Delivery Unit 1 through Delivery Unit 5
  • Level 2 — Delivering Cost Centre: e.g. Cost Centre 1.1, Cost Centre 1.2 (each belongs to one Delivery Unit)
  • Level 3 — Work Centre Code: e.g. Work Centre 1.1.1, Work Centre 1.1.2 (each belongs to one Cost Centre)

There is also a blank roll-up at each level (i.e. all Delivery Units, all Cost Centres within a DBU, all Work Centres within a Cost Centre), represented by leaving the relevant field(s) blank.

TASK

For every node in the hierarchy — including blank roll-ups — generate three narratives, each written to support a specific named dashboard view.

Use the latest available Snapshot AP period in the data for all calculations.

Filter rows where Deleted != 1 before aggregating activity data.

METRICS TO CALCULATE

Aggregate the activity and capacity data for the unit being described. For blank roll-up rows, aggregate across all constituent units at that scope.

Calculate the following before writing each narrative:

Metric Formula
On-time rate (achieved) sum(Achieved On Time) / count(activities) × 100
At risk count sum(At Risk)
Miss - Delivered count sum(Miss - Delivered)
Miss - Outstanding count sum(Miss - Outstanding)
Total Remaining Hours sum(Remaining Hours)
Average Total Float Live mean(Total Float Live)
Utilisation sum(Actual Demand Hrs) / sum(Actual Capacity Hrs) × 100
Capacity Variance Hrs sum(Capacity Variance Hrs)
Actual Capacity Hrs sum(Actual Capacity Hrs)
Actual Demand Hrs sum(Actual Demand Hrs)

For blank roll-up rows, also identify the highest and lowest performing constituent unit on at least two dimensions.

RAG THRESHOLDS

Assign a RAG status to each node using these thresholds:

Status Condition
RED Achieved on-time rate < 25% OR utilisation > 125%
AMBER Achieved on-time rate < 50% OR at_risk > 2 OR utilisation > 105%
GREEN All other cases

THREE NARRATIVE TYPES

Each narrative type corresponds directly to a named dashboard view. Write as if the narrative will appear in that view's Narrative panel, complementing the visualisations a user can already see on screen.

1. exec_summary — Summary Dashboard

Dashboard context: The Summary view shows four KPI tiles (Remaining Hours, Capacity Variance, Total Float, Avg Utilisation), a Delivery Status by Period stacked bar chart (Achieved On Time / Forecast On Time / At Risk / Miss Delivered / Miss Outstanding), an Average Float trend line, and a Remaining Hours Burn Down bar chart.

Audience: Senior leadership / executive. No technical jargon.

Length: 2–3 sentences maximum.

Required content focus:

  • Lead with RAG status and the headline on-time rate (n of total).
  • Reference the capacity signal: quote Avg Utilisation and Capacity Variance Hrs from the KPI tiles.
  • Close with the most critical risk: cite the Miss Outstanding count and/or At Risk count as the action item.

Tone: Confident, direct, board-ready.

2. programme_detail — Programme Detail Dashboard

Dashboard context: The Programme Detail view shows a Total Remaining Hours Burndown line chart (by Delivering DBU over time), a Delivery Status table (Achieved On Time / Forecast On Time / At Risk / Miss Delivered / Miss Outstanding split by month and quarter), a Programme Performance Summary stacked bar (At Risk counts by month), and a Schedule Float by Programme Over Time line chart.

Audience: Programme managers and PMO. Moderate technical detail.

Length: 4–6 sentences.

Required content focus:

  • Open with on-time rate and the split of Achieved On Time vs Forecast On Time vs Miss Outstanding from the period table.
  • Reference the at-risk trend visible in the Programme Performance Summary chart — call out whether at-risk count is rising, stable, or declining.
  • Cite Average Total Float Live and describe its trajectory (improving, stable, or deteriorating) as seen in the Schedule Float chart.
  • Reference total Remaining Hours and whether the burndown trajectory appears on track or lagging.
  • For roll-up nodes, call out the best and worst constituent performers by on-time rate.

Tone: Analytical, factual, structured.

3. delivery_unit — Delivery Unit Dashboard

Dashboard context: The Delivery Unit view shows a capacity/demand table (Actual Capacity Hrs, Actual Demand Hrs, Capacity Variance Hrs, Utilisation by month), a Forecast vs Actual: Capacity and Demand by Delivery Unit grouped bar chart showing variance values per period, a second Forecast vs Actual bar chart by Delivering DBU, and a Utilisation by Delivery Unit Over Time line chart.

Audience: Delivery Unit leads, Cost Centre managers, Work Centre supervisors.

Length: 5–8 sentences.

Required content focus:

  • Open with current period Utilisation % and Actual Capacity vs Actual Demand Hrs (i.e. the figures in the capacity table header row).
  • Cite the Capacity Variance Hrs for the current period and characterise whether it represents a shortfall or surplus.
  • Reference the Forecast vs Actual variance chart: describe whether the gap between forecast and actual capacity/demand is widening, stable, or narrowing.
  • Reference the Utilisation Over Time trend: is utilisation pressure increasing, easing, or stable for this unit?
  • Bring in the activity performance context: on-time rate, miss-outstanding count, at-risk count, and remaining hours.
  • For roll-up nodes, call out which constituent unit has the highest utilisation and which has the lowest, and name the on-time delivery range across constituents.
  • Close with a concrete, action-oriented recommendation tied to the data (e.g. rebalance demand, redirect surplus capacity, escalate float recovery).

Tone: Operational, specific, action-oriented.

NARRATIVE WRITING RULES

  • Ground every narrative in the data. Reference actual calculated figures. Do not be vague (write "73% of activities achieved on time", not "most activities were delivered on time").
  • Highlight the most important signal first. Lead with the most significant performance indicator.
  • Maintain hierarchy awareness. At Work Centre level, reference the parent Cost Centre. At Cost Centre level, reference the parent Delivery Unit.
  • Use consistent performance language:
    • Green / on track: "achieved on time", "within forecast", "capacity aligned to demand"
    • Amber / at risk: "at risk of slippage", "demand exceeding capacity", "negative float trending"
    • Red / missed: "missed baseline", "overrun", "unrecovered float"
  • For blank roll-ups, synthesise across constituent units — identify best and worst performers and note the range.
  • Do not fabricate data. If a metric is not available, acknowledge it rather than inventing a figure.
  • Each of the three narratives for the same unit must be meaningfully different in content, not just length. The exec_summary speaks to KPI tiles and headline risk; the programme_detail speaks to burndown, float trends, and period tables; the delivery_unit speaks to capacity variance, utilisation trends, and demand-supply gaps.

OUTPUT FORMAT

Write the output as a single JSON file — no other files, no wrapper text.

The JSON must be an array where every element has exactly these six keys:

Field Description
delivery_unit Full value of Delivering DBU (e.g. "Delivery Unit 1"), or "" for org-wide roll-up
cost_centre Full value of Delivering Cost Centre (e.g. "Cost Centre 1.1"), or "" if not at Cost Centre level
work_centre Full value of Work Centre Code (e.g. "Work Centre 1.1.1"), or "" if not at Work Centre level
narrative_type One of: "exec_summary", "programme_detail", "delivery_unit"
narrative The full written narrative text
rag_status "Red", "Amber", or "Green" — the same value for all three rows belonging to the same node

Important: Field values must exactly match the source data strings. rag_status must be exactly "Red", "Amber", or "Green" — no other casing or values are permitted.

How the three columns populate at each level:

Scope delivery_unit cost_centre work_centre
Org-wide roll-up "" "" ""
Delivery Unit 1 "Delivery Unit 1" "" ""
Cost Centre 1.1 "Delivery Unit 1" "Cost Centre 1.1" ""
Work Centre 1.1.1 "Delivery Unit 1" "Cost Centre 1.1" "Work Centre 1.1.1"

Example output rows:

[
  {
    "delivery_unit": "",
    "cost_centre": "",
    "work_centre": "",
    "narrative_type": "exec_summary",
    "narrative": "The organisation is rated AMBER: 35.3% of activities achieved on time (24 of 68). Average utilisation stands at 113.4% with a capacity variance of -1,943 hours, indicating sustained demand pressure. Immediate action is required on 26 outstanding missed activities and 6 at-risk activities to prevent further programme slippage.",
    "rag_status": "Amber"
  },
  {
    "delivery_unit": "Delivery Unit 1",
    "cost_centre": "Cost Centre 1.1",
    "work_centre": "Work Centre 1.1.1",
    "narrative_type": "delivery_unit",
    "narrative": "Work Centre 1.1.1 within Cost Centre 1.1 is operating at 89.7% utilisation this period, with actual demand of 751 hours against actual capacity of 837 hours — a surplus of 86 hours. The capacity variance is positive and has been widening over recent periods, suggesting demand is easing relative to available resource. Utilisation has trended downward from a peak earlier in the year, indicating reducing pressure on this work centre. Activity performance stands at 50.0% on time across 4 activities; one miss-outstanding activity requires recovery and one activity is at risk of slippage, with 29 remaining hours in scope. The available 86-hour capacity surplus should be redirected to support float recovery on the at-risk activity and assist adjacent overloaded work centres.",
    "rag_status": "Green"
  }
]

PROCESSING INSTRUCTIONS

  1. Read the activity CSV. Filter Deleted != 1. Use the latest Snapshot AP period only.
  2. Read the capacity CSV. Use the latest Snapshot AP period only.
  3. Walk the full hierarchy (org → DBU → Cost Centre → Work Centre). For each node, aggregate activity and capacity rows by filtering on the relevant dimension(s).
  4. Calculate all metrics listed in the METRICS TO CALCULATE section.
  5. Assign RAG status per the thresholds above.
  6. Write all three narrative types per node, ensuring each narrative is anchored to the metrics its dashboard view displays.
  7. Append three rows to the output array (one per narrative type), each carrying the same rag_status value for that node.
  8. After processing all nodes, serialise the complete array to a single .json file.
  9. Do not produce any other output — no print statements, no summary, no separate files.

End of prompt.