PinMeTo Location Analytics Reports

Generate professional, board-ready performance reports for enterprise multi-location brands using PinMeTo's analytics data.

Requirements

Requires PinMeTo Location MCP >= 4.0.0 (12 tools). Earlier majors used different tool names, parameters, and response shapes, and this skill's calls will fail or return wrong data against them.

Confirm the server is connected before starting. To verify the tool surface matches what this skill expects:

node scripts/check_mcp_parity.js

Quick Start

1. Confirm the PinMeTo MCP server is connected
2. Parse the user request for period type and date range
3. Calculate from and to dates (see Date Range Calculation)
4. Fetch data using the MCP tools (see MCP Call Rules)
5. Generate PDF and/or PPTX output with PinMeTo branding

---

MCP Call Rules

Read references/workflow-details.md for the full parameter contracts. These are the rules that break reports most often.

Every insights call needs from, to, and a comparison

pinmeto_get_google_insights({
  "from": "2025-10-01",
  "to": "2025-12-31",
  "aggregation": "quarterly",
  "compare_with": "prior_year"
})

from and to are required on every data tool. Both are strings; insights, ratings, reviews, and review-insights tools take YYYY-MM-DD, keywords take YYYY-MM.

Exact spellings that fail silently or loudly

Correct	Wrong	Consequence if wrong
storeId	store_id	Silently returns all locations. A single-store report gets brand-wide data
half-yearly	half_yearly	Call fails with -32602
compare_with	comparison_type	Comparison silently dropped, all changes read "N/A"
YYYY-MM for keywords	YYYY-MM-DD	Call fails validation

Unknown parameter names are silently stripped; invalid values for a real parameter fail loudly. The silent case is the dangerous one.

Parameters that do not exist

• aggregation and compare_with on ratings tools. Ratings take only from, to, storeId, forceRefresh. For a rating delta, call once per period and subtract.
• limit on the keywords tool. It returns the full set; take the top N in the skill.
• filters / status on pinmeto_get_locations. Use permanentlyClosed, type, city, country.
• city / country / region as location fields. Geography is inside address.

Default Comparison Period

Default to Year-over-Year unless the user asks otherwise. YoY accounts for seasonality, which matters for retail and service brands.

User Request	Use compare_with
"Q4 2025 report" (no comparison specified)	"prior_year" (YoY)
"Q4 2025 vs Q4 2024"	"prior_year" (YoY)
"Q4 2025 vs Q3 2025" or "QoQ comparison"	"prior_period" (QoQ)
"Compare to last month" or "MoM"	"prior_period"

Populating both periodChange and yearChange requires two insights calls, one per comparison type. With a single YoY call, set periodChange to "N/A" rather than inventing it.

Read the response, do not assume it

• insights is an array keyed by metric name. Comparison fields (priorValue, delta, deltaPercent) are flat, not nested under comparison.
• deltaPercent is null when the baseline is 0. Render "N/A".
• warningCode: "INCOMPLETE_DATA" means the range hit Google's reporting lag. Surface the warning text in the appendix rather than reimplementing the lag rule.
• Only retry when retryable is true.

---

Period Detection

Parse natural language to determine report type:

Report Type	Keywords
Monthly	"October 2024", "last month", "2024-10"
Quarterly	"Q3 2024", "Q1", "third quarter", "last quarter"
Half-Yearly	"H1 2024", "first half", "H2", "second half"
Yearly	"2024", "annual", "yearly", "year-end"

Always pass the detected type to the generators with --period. It is authoritative: it selects the yearly three-column table layout (value plus YoY, no period-over-period column) and the highlights heading. Without it the generators fall back to inferring the type from the free-text period label, which is a guess.

Page count is driven by the data, not the period. Each platform, keywords, and reviews section is emitted only when its data is present, so a report covering Google alone is far shorter than one covering three platforms plus keywords and sentiment. A full report over all sections runs about 9 pages. Do not promise a page count before seeing what the fetch returns.

Date Range Calculation

Calculate from and to once, then use them for every MCP call in the report.

# Monthly: first to last day of month. October 2024:
from = "2024-10-01";  to = "2024-10-31"

# Quarterly: Q1=Jan-Mar, Q2=Apr-Jun, Q3=Jul-Sep, Q4=Oct-Dec. Q4 2025:
from = "2025-10-01";  to = "2025-12-31"

# Half-Yearly: H1=Jan-Jun, H2=Jul-Dec. H2 2025:
from = "2025-07-01";  to = "2025-12-31"

# Yearly: full calendar year. 2025:
from = "2025-01-01";  to = "2025-12-31"

# Keywords only: same range at month precision. Q4 2025:
from = "2025-10";     to = "2025-12"

Report Generation Workflow

Step 1: Validate Request

• Location scope is clear (all locations vs specific store IDs)
• Output format confirmed (PDF, PPTX, or both)
• End date is not inside Google's ~10-day reporting lag

If the requested period ends within the last 10 days, warn the user that Google metrics may be incomplete and suggest an earlier end date. The API also flags this itself with warningCode: "INCOMPLETE_DATA".

To report on a specific store the user named loosely ("Store #47", "the Stockholm one"), resolve it to a real storeId with pinmeto_search_locations first.

Steps 2-4: Fetch Data

Follow the fetch sequence in the period-specific reference file (see Reference Files), which lists the calls in order with correct parameters.

Available tools:

• Locations: pinmeto_get_locations, pinmeto_get_location, pinmeto_search_locations
• Google: pinmeto_get_google_insights, pinmeto_get_google_ratings, pinmeto_get_google_reviews, pinmeto_get_google_review_insights, pinmeto_get_google_keywords
• Facebook: pinmeto_get_facebook_insights, pinmeto_get_facebook_brandpage_insights, pinmeto_get_facebook_ratings
• Apple Maps: pinmeto_get_apple_insights

Prefer pinmeto_get_google_review_insights over pulling every review when the report needs sentiment: it aggregates server-side and costs far fewer tokens. Fetch raw reviews only for pull quotes and theme identification.

For a scripted batch fetch of a full year:

node scripts/fetch_pinmeto_data.js --year 2025 --output report_data.json

Step 5: Process & Analyze Data

1. Calculate changes: absolute (current - previous) and percent (((current - previous) / previous) * 100). Guard against a zero baseline.
2. Sum the roll-ups: Total Views and Total Actions are sums of component metrics, not API fields. See references/metrics-glossary.md.
3. Classify keywords using references/keyword-classification.md.
4. Aggregate multi-location: sum totals, average ratings weighted by review count, excluding locations with no reviews.
5. Generate insights: top performers, declining metrics, trends.

Step 5.5: Validate Data

python scripts/validate_report_data.py report_data.json

Fix all validation errors before generating. The schema is documented in references/data-schema.md.

Step 6: Generate Report

Read the period-specific reference file for report structure, then run the bundled generator directly from the skill directory. Do not copy the scripts into the working directory or rewrite them: they encode the branding and layout, and a copy will drift.

# Set once to the directory containing this SKILL.md
SKILL_DIR="/path/to/pinmeto-location-reports"

pip install reportlab python-pptx pillow matplotlib
# matplotlib is not optional in practice: generate_pdf.py imports it at module
# level, and generate_pptx.py silently degrades to text-based charts without it.

# PDF
python "$SKILL_DIR/scripts/generate_pdf.py" \
  --data report_data.json --output report.pdf --period quarterly

# PPTX
python "$SKILL_DIR/scripts/generate_pptx.py" \
  --data report_data.json --output report.pptx --period quarterly

--period accepts monthly, quarterly, half-yearly, or yearly, and takes precedence over the period label in the data. Set periodType in the data JSON instead to make it self-describing.

Step 7: Customer Review (Human-in-the-Loop)

Present a draft for review before finalizing.

7.1 Generate a draft with --draft to add a diagonal "DRAFT - PENDING REVIEW" watermark to every page or slide:

python "$SKILL_DIR/scripts/generate_pdf.py" \
  --data report_data.json --output Brand_Q4_Report_DRAFT.pdf --period quarterly --draft

7.2 Present for review with a summary of the key data:

I've generated a draft Q4 2025 report with a DRAFT watermark.

File: Brand_Q4_2025_Report_DRAFT.pdf

Key data included:
- Total Views: 125,432 (+12% YoY)
- Total Actions: 8,234 (+8% YoY)
- Average Rating: 4.6 (+0.2)
- 12 locations analyzed

Note: Google data for the final 10 days of December may be incomplete.

Reply "approved" for the final version without the watermark, or tell me what to correct.

State any dataWarnings here. A platform that returned no data looks identical to a real decline in the finished report.

7.3 Review checklist (full version in references/customer-review.md): executive summary accuracy, KPI values, chart rendering, table completeness, text correctness, branding.

7.4 Handle feedback:

• Approved: regenerate without --draft
• Changes needed: update the data JSON or chart parameters, regenerate with --draft, repeat

Step 8: Quality Check & Delivery

Work through references/qa-checklist.md before delivering.

Report Data Schema

The generators expect an exact JSON structure with camelCase field names. Full schema, field sources, and common failures: references/data-schema.md.

Brand Guidelines

Both generators hardcode the full palette, typeface, and chart tokens, so generating a report needs none of it in context. Read references/branding.md before editing a generator's tokens, adding a chart type, or answering a brand question.

The four rules that get violated most often:

• Montserrat is the only brand typeface. It ships in assets/fonts/, so reports are on-brand on machines that never installed it. PDF embeds it; PPTX cannot, so deliver PDF when typography must be guaranteed.
• Navy is #000050, and it is the body ink. Not #001334.
• Brand hues are not data colours. #3399FF and #FF8854 fall below the 3:1 contrast floor against a white chart surface. Charts use the validated tokens in references/branding.md.
• Direction is never colour alone. Every delta carries a triangle (▲ / ▼ / –) beside the status colour, so red-green colourblind readers can still tell a rise from a fall.

Never set body copy in white on Blue or Orange: white on Orange measures 2.0:1 and fails even the large-text floor. Use Navy on both brand backgrounds.

Reference Files

File	Purpose
references/workflow-details.md	MCP tool contracts, parameters, response shapes
references/data-schema.md	Report data JSON schema and field sources
references/branding.md	Colours, contrast ratios, typography, chart tokens, tone of voice
references/monthly.md	Monthly report structure (8-15 pages)
references/quarterly.md	Quarterly report structure (10-18 pages)
references/half-yearly.md	Half-yearly report structure (12-20 pages)
references/yearly.md	Yearly report structure (15-25 pages)
references/metrics-glossary.md	Metric keys and derived roll-ups
references/keyword-classification.md	Keyword categorization rules
references/customer-review.md	Customer review checklist
references/qa-checklist.md	Quality assurance checklist

Output Formats

PDF Reports

• Multi-page document with headers/footers and the PinMeTo logo on each page
• Charts embedded as images, tables with brand styling
• Executive summary first, details following

PowerPoint Presentations

• 16:9 aspect ratio (720pt x 405pt)
• Title slide with vertical logo, section dividers in brand colors
• Charts rendered as PNG images for Keynote and PowerPoint compatibility
• Recommendations slide with action items

Example Usage

Monthly report:

"Create a monthly report for October 2024"

Quarterly report for a specific location:

"Generate Q3 2024 report for Store #47"

Aggregated half-yearly:

"Create an H1 2024 executive summary for all locations"

Annual board presentation:

"Generate 2024 yearly report as PowerPoint for the board meeting"