Analytics And Advisory Intelligence

Financial_Trends:
  metrics_tracked:
    gross_margin:
      calculation: "(Revenue - Cost of Sales) / Revenue"
      alert_threshold: "Drop of more than 5 percentage points vs same period last year"
      finding_template: "{Client} gross margin has fallen from {X}% to {Y}% — {delta}pp decline over {N} months. Primary driver appears to be {top expense category change}."

    vat_liability_trend:
      calculation: "Net VAT payable per period"
      alert_threshold: "Increase of more than 20% vs prior 3-month average"
      finding_template: "{Client} VAT liability has increased {X}% over {N} months. May reflect higher turnover, a change in customer mix, or a classification issue worth reviewing."

    staff_cost_ratio:
      calculation: "Staff costs (EGLS account 64) / Revenue"
      alert_threshold: "Ratio increase of more than 3 percentage points"

    cash_position:
      calculation: "Cash and bank balances (account 38) trend"
      alert_threshold: "Declining for 3 consecutive months"
      finding_template: "{Client} cash position has declined for {N} consecutive months (from EUR {X} to EUR {Y}). With a tax payment of EUR {Z} due on {date}, this warrants a cash flow conversation."

  minimum_periods_required: 3
  preferred_periods: 12
  confidence_note: "Findings with fewer than 6 data points are marked LOW CONFIDENCE"

Anomaly_Detection:
  methods:
    own_history_comparison:
      description: "Value more than 2 standard deviations from client's own trailing average"
      example: "Electricity costs EUR 4,200 in January vs trailing 12-month average of EUR 1,800"

    sector_peer_comparison:
      description: "Value more than 1.5x or less than 0.5x the sector median for that account"
      note: "Sector medians computed from anonymised aggregates across the portfolio"

    vat_rate_anomaly:
      description: "Transaction classified at incorrect VAT rate based on product/service category"
      example: "Food product invoice at 24% rather than correct 13% reduced rate"
      action: "Flag for accountant review — never auto-correct"

  output_fields:
    - metric: "What was measured"
    - actual_value: "What was found"
    - expected_range: "What is normal for this client / sector"
    - deviation: "How far outside normal"
    - confidence: "HIGH / MEDIUM / LOW"
    - suggested_action: "Plain English recommendation"
    - data_source: "Which files the finding is based on"

Cashflow_Forecast:
  inputs:
    known_inflows:
      - Recurring revenue based on trailing 3-month average
      - Outstanding debtor invoices from registry
    known_outflows:
      - Upcoming tax deadlines from cli-deadline-monitor
      - Recurring supplier payments from banking history
      - EFKA contributions due
      - Payroll from prior period data
    uncertainty_ranges:
      - Revenue: +/- 15% based on historical variance
      - Ad-hoc expenses: +/- 20%

  output:
    - Best case / expected / worst case cash position at each month end
    - Months where cash may go negative (flagged HIGH RISK)
    - Specific upcoming payments that may cause strain

  confidence_degradation:
    1_month: "HIGH confidence"
    2_months: "MEDIUM confidence"
    3_months: "LOW confidence — directional only"

Portfolio_Intelligence:
  sector_benchmarks:
    computed_from: "Anonymised aggregates of all active clients per sector"
    metrics:
      - Median gross margin by sector
      - Median staff cost ratio by sector
      - Median VAT liability as percentage of revenue by sector
    refreshed: "Monthly — after all monthly statements are generated"
    minimum_clients_per_sector: 3

  portfolio_risk_distribution:
    description: "Distribution of risk scores across the full portfolio over time"

  workload_concentration:
    description: "Risk-weighted workload per accountant — alerts if one person carries disproportionate risk"

  common_issues:
    description: "Issues appearing across multiple clients simultaneously"
    examples:
      - "6 retail clients showing declining margins — possible sector-wide trend"
      - "3 clients using the same supplier showing unusual payment patterns"
      - "4 clients with outstanding document requests over 21 days"

MORNING ADVISORY — 19/02/2026
Generated: 19/02/2026 05:47 Athens time

WATCH LIST — 3 clients warrant attention today

1. GAMMA CONSTRUCTIONS AE (EL555444333) — RISK: HIGH (8/10)
   Cash position has declined 3 consecutive months. Current balance EUR 12,400.
   Tax payment of EUR 8,900 due 25/03/2026. Margin is tight.
   Suggest: Review with client before month end.

2. DELTA SERVICES EPE (EL777888999) — RISK: MEDIUM (6/10)
   VAT liability increased 34% vs prior 3-month average. January: EUR 4,100
   vs average of EUR 3,060. Not yet investigated.
   Suggest: Check whether turnover genuinely increased or a classification
   issue exists in the January invoices. Confidence: MEDIUM (4 months data)

3. EPSILON RETAIL OE (EL222333444) — RISK: MEDIUM (5/10)
   Bank statements for January not received. VAT filing due 25/02/2026 in 6 days.
   Suggest: Send document request today.

PORTFOLIO SNAPSHOT
Active clients: 42 | Low risk: 31 | Medium: 8 | High: 3 | Critical: 0
Next high-volume deadline: VAT February 25/02/2026 (6 days, 18 clients affected)

SECTOR NOTE
Retail sector (7 clients): Gross margins down an average of 3.2pp vs Q4 2025.
Pattern appears sector-wide. May be worth raising with retail clients proactively.
Confidence: MEDIUM (7 clients, 3 months of data)

Analytics_File_Structure:
  owns: "/data/reports/analytics/"

  daily_outputs:
    - "{YYYY-MM-DD}_risk-scores.json"
    - "{YYYY-MM-DD}_morning-advisory.json"
    - "{YYYY-MM-DD}_morning-advisory.txt"
    - "{YYYY-MM-DD}_anomalies.json"

  monthly_outputs:
    - "{YYYY-MM}_portfolio-report.pdf"
    - "{YYYY-MM}_sector-benchmarks.json"
    - "{YYYY-MM}_trend-analysis.json"

  on_demand_outputs:
    - "{YYYY-MM-DD}_{AFM}_risk-detail.json"
    - "{YYYY-MM-DD}_{AFM}_forecast.json"
    - "{YYYY-MM-DD}_{AFM}_trends.json"

  reads_from:
    - "/data/clients/*/financial-statements/"
    - "/data/clients/*/compliance/"
    - "/data/clients/*/correspondence/"
    - "/data/banking/reconciliation/"
    - "/data/efka/"
    - "/data/reports/analytics/"   # Prior outputs for trend continuity

  never_writes_to:
    - "/data/clients/"
    - "/data/compliance/"
    - "/data/banking/"
    - "/data/efka/"

Analytics_Schedule:
  nightly_run:
    time: "01:30 Athens time (after backup and integrity check)"
    operations:
      - "Compute risk scores for all active clients"
      - "Run anomaly detection across all clients with new data"
      - "Update trend analysis for clients with new financial statements"
      - "Generate morning advisory"
    duration_estimate: "10-30 minutes depending on portfolio size"

  monthly_run:
    trigger: "After all monthly statements generated (typically 5th-10th of following month)"
    operations:
      - "Recompute sector benchmarks"
      - "Generate portfolio monthly report"
      - "Run 12-month trend analysis for all clients with sufficient history"

  on_demand:
    available_during: "Business hours"
    rate_limit: "Maximum 10 on-demand requests per hour"
    bounded_by: "Pre-computed daily outputs where possible"

Upstream_Skills_Read:
  greek-financial-statements:   "P&L, balance sheet, VAT summary — primary financial data"
  client-data-management:       "Compliance history, document registry, correspondence log"
  greek-banking-integration:    "Reconciliation data, cash position, transaction patterns"
  efka-api-integration:         "Payroll costs, employee counts, contribution history"
  cli-deadline-monitor:         "Upcoming deadlines — feeds cash flow forecast"
  greek-compliance-aade:        "Filing history, penalty records"
  client-communication-engine:  "Correspondence patterns — identifies chronic document requesters"

Downstream_Skills_Feed:
  conversational-ai-assistant:  "Answers analytics queries; morning advisory pulled via chat"
  dashboard-greek-accounting:   "Risk scores, portfolio snapshot, watch list"
  client-communication-engine:  "Advisory findings can prompt draft letters (human initiates)"

Memory_Integration:
  log_episodes: true
  episode_types:
    - morning_advisory_generated
    - anomaly_detected_and_surfaced
    - risk_score_computed
    - forecast_generated

  log_failures: true
  failure_types:
    - insufficient_data_for_analysis
    - sector_benchmark_too_small
    - financial_statements_missing

  useful_patterns_to_detect:
    - "Anomaly types consistently dismissed by accountants — thresholds may be too sensitive"
    - "Risk score bands that consistently predict actual compliance problems — calibration quality"
    - "Clients where forecasts are consistently wrong in one direction — systematic bias"

  rate_limit_group: "analytics_operations"

Error_Responses:

  insufficient_history:
    output: "{Client} has only {N} periods of data — trend analysis requires at least 3. Showing available data with LOW CONFIDENCE marker."
    action: "Produce what is possible, clearly labelled. Do not refuse to show anything."

  sector_too_small:
    output: "Sector benchmark for {sector} cannot be computed — only {N} clients (minimum 3 required). Showing own-history analysis only."

  missing_statements:
    output: "Financial trend analysis for {client} requires Skill 15 outputs for {period}. Statements not yet generated."
    action: "Show analysis based on available periods. Note the gap explicitly."

  on_demand_rate_limit:
    output: "On-demand analysis limit reached (10/hour). Pre-computed morning advisory is available now. Full analysis available after {time}."
    action: "Direct to pre-computed outputs. Never drop the request silently."