Financial Analysis Skill

Overview

Takes a QuickBooks CSV export (Profit & Loss, Transaction List, or Sales by Item) and produces a plain-English financial analysis for a non-technical business owner. The agent identifies cost categories, flags problem areas, and suggests specific actions — all delivered in Telegram as a readable summary, not a spreadsheet.

Two modes:

1. CSV import (works now) — owner exports from QuickBooks, sends to agent
2. QuickBooks API (optional) — agent pulls financials directly (requires quickbooks-online skill to be wired with OAuth tokens)

When to Use

• "Analyze my QuickBooks export" (after sending a CSV)
• "Where am I losing money?"
• "Which jobs had the worst margins last quarter?"
• "What are my biggest cost categories?"
• "Give me a financial summary for the owner meeting"
• Monthly/quarterly financial review

Prerequisites

CSV Mode (current)

• CSV file exported from QuickBooks Online or QuickBooks Desktop
• File placed on the VPS (or content pasted — see formats below)

QuickBooks API Mode (optional, post-hackathon)

• quickbooks-online skill installed + OAuth tokens wired
• Not required for CSV mode

Supported CSV Formats

Format 1: Profit & Loss (Standard QuickBooks Export)

Date,Transaction Type,Num,Name,Memo/Description,Account,Income,Expense
2026-01-15,Invoice,101,Smith Residence,Odor removal service,Service Income,2800,
2026-01-18,Bill,,Home Depot,Chemicals,COGS - Chemicals,,150

Format 2: Transaction List (All Transactions)

Date,Transaction Type,Num,Name,Split,Amount,Status
2026-01-15,Invoice,101,Smith Residence,Service Income,2800,Cleared
2026-01-18,Bill,,Home Depot,COGS - Chemicals,-150,Cleared

Format 3: Sales by Item/Job

Item/Job,Qty,Amount,COGS,Margin,Margin %
Odor Removal - Residential,1,2800,350,2450,87.5%
Odor Removal - Commercial,2,4500,800,3700,82.2%

The agent auto-detects the format by reading the header row and matching column names. If columns are ambiguous, the agent asks the owner to clarify.

The Flow

Step 1: Owner Sends CSV

Owner sends file to agent (via Telegram document upload or places on VPS). Agent confirms receipt and reads the file.

Step 2: Agent Parses + Categorizes

import csv
import json
from collections import defaultdict

def analyze_csv(filepath):
    with open(filepath, 'r') as f:
        reader = csv.DictReader(f)
        rows = list(reader)

    # Auto-detect format from headers
    headers = set(rows[0].keys())

    if 'Income' in headers or 'Expense' in headers:
        return analyze_pnl(rows)
    elif 'Amount' in headers:
        return analyze_transactions(rows)
    elif 'COGS' in headers or 'Margin' in headers:
        return analyze_sales_by_job(rows)
    else:
        return {"error": "Unknown CSV format", "headers": list(headers)}

Step 3: Agent Runs Analysis

A. Revenue Summary

• Total revenue (period)
• Revenue by job/customer
• Revenue by service type (if identifiable)
• Month-over-month trend
• Average job size

B. Cost Analysis

• Total costs (period)
• Cost categories (ranked by size):
  • COGS (chemicals, equipment, supplies)
  • Labor (subcontractors, wages)
  • Travel/fuel
  • Marketing
  • Software/subscriptions
  • Insurance
  • Other
• Cost as % of revenue (overall)
• Cost trend (increasing? decreasing?)

C. Job-Level Profitability

• Each job with revenue, cost, and margin
• Flag: jobs with margin < 30% (or negative margin)
• Flag: jobs where cost > revenue (loss-making)
• Best performing job (highest margin)
• Worst performing job (biggest loss or lowest margin)

D. Opportunity Analysis

• Most profitable service type → "do more of these"
• Least profitable → "consider raising prices or declining"
• Customer concentration risk (if >X% from one source)
• Seasonal patterns (which months are strongest)
• Cost categories growing faster than revenue

Step 4: Agent Delivers Plain-English Report

📊 [BUSINESS_NAME] Financial Analysis — Jan-Jun 2026

TOTALS:
- Revenue: $48,200 across 22 jobs
- Costs: $14,800 (30.7% of revenue)
- Net profit: $33,400 (69.3% margin — healthy)

YOUR BIGGEST COST CATEGORIES:
1. Chemicals & Supplies: $4,200 (28% of costs)
2. Fuel/Travel: $3,100 (21% of costs)
3. Insurance: $2,400 (16% of costs)
4. Software/Subscriptions: $890 (6% of costs)

⚠️ JOBS THAT NEED ATTENTION:
- Job #1018 ([CITY], small apartment): Revenue $650,
  Costs $580 → Margin only 8.6%. This job cost almost
  as much as it made. Consider a minimum job size of
  $1,000 or raise the small-job rate.

💰 YOUR BEST JOBS:
- Job #1003 ([CITY], 3200 sqft residential): Revenue
  $3,800, Costs $290 → Margin 92.4%. Large residential
  jobs are your sweet spot.

📈 OPPORTUNITIES:
- Your [CITY]/[REGION] jobs average 88% margin vs
  72% in other regions. Consider targeting more [REGION]
  partners.
- Your commercial jobs average $4,500 vs $2,400 for
  residential. But you only did 3 commercial jobs vs 19
  residential. More commercial outreach could boost revenue.
- Fuel costs are 21% of total costs. Premium Region trips
  have higher fuel — the Premium Region
  surcharge should be reviewed to ensure it covers actual
  travel cost.

NEXT STEPS:
- Set a minimum job size? (Your smallest jobs have
  the worst margins)
- Raise the Premium Region travel surcharge?
- Target more commercial jobs?
- More marketing to [REGION] partners?

Want me to dig deeper into any of these?

Step 5: Owner Can Ask Follow-Ups

• "Show me all jobs under $1,000"
• "What's my average margin by region?"
• "Break down the chemical costs month by month"
• "Which months had the highest revenue?"
• "Compare my margins last year vs this year"

QuickBooks API Integration (Optional, Post-Hackathon)

When quickbooks-online skill is wired with OAuth tokens, the agent can pull financials directly without CSV export:

# Pseudocode — requires quickbooks-online skill to be wired
from intuitlib.client import AuthClient

# Pull Profit & Loss report
pnl = quickbooks.get_report('ProfitAndLoss', start_date='2026-01-01',
                            end_date='2026-06-30')

# Pull Sales by Customer
sales = quickbooks.get_report('SalesByCustomer', start_date='2026-01-01',
                               end_date='2026-06-30')

# Run same analysis as CSV mode but from live data

Agent Integration Patterns

Pattern 1: Monthly Financial Review

Owner: "Analyze my June financials"
Agent: [reads CSV / pulls from QBO] → produces report → sends in Telegram
Owner: "Why were chemical costs so high in June?"
Agent: [drills into chemical line items] → "3 large commercial jobs in
       June required extra treatments — $1,200 of the $1,800 total"

Pattern 2: Job Profitability Audit

Agent: "I noticed 3 jobs last quarter had margins under 20% — want a
       breakdown?"
Owner: "Yes"
Agent: [lists jobs with revenue, costs, margin] → "All three were
       small apartments under 1,000 sqft. Your minimum job rate of
       $1,500 would have prevented two of these from being unprofitable."

Pattern 3: CRM Cross-Reference

Agent: "Your CRM shows 8 jobs booked from Standard Region partners this
       quarter, averaging $2,100 revenue. But your financials show
       Standard Region jobs average 72% margin vs 88% in Premium Region. Want me
       to draft a targeted Premium Region partner campaign?"

Pattern 4: Estimator Sanity Check

Agent: "Your financials show actual job costs averaging 30% of revenue.
       Your estimator-engine assumes 25% cost ratio. The gap is mostly
       fuel costs on Premium Region jobs. Want me to update the estimator
       to include a fuel surcharge for Premium Region?"

Common Pitfalls

1. CSV format varies — QuickBooks Online, Desktop, and different report types all export differently. The agent must auto-detect and ask if unsure. Don't assume column names.
2. Cost allocation — Some costs (insurance, software) aren't tied to specific jobs. The agent should separate direct job costs from overhead and be clear about the difference.
3. Time periods — Always confirm the date range with the owner before analyzing. "This shows Jan-Jun 2026 — is that right?"
4. Negative numbers — QuickBooks sometimes exports expenses as positive, sometimes as negative. The agent should detect the convention and normalize.
5. Missing data — If COGS isn't broken out by job, the agent should say so rather than guessing. "Chemical costs are lumped — I can't tie them to specific jobs without more detail."
6. Don't give financial advice. The agent identifies patterns and asks questions. The owner makes decisions. The agent is a tool, not a CPA.
7. Privacy — Financial data stays on the VPS. Don't send raw numbers to external services. Summary in Telegram is fine; raw CSV is not.

Verification Checklist

• Can read CSV file from disk
• Auto-detects CSV format from headers
• Parses income/expense transactions correctly
• Identifies cost categories and ranks by size
• Calculates job-level margins
• Flags jobs with margin < 30%
• Flags jobs where cost > revenue
• [ | Produces plain-English summary in Telegram
• Owner can ask follow-up questions
• QuickBooks API integration (optional, post-hackathon)

One-Shot Recipes

"Quick financial health check"

# Reads CSV → returns: total revenue, total costs, margin %,
# top 3 cost categories, worst job, best job, 2-3 recommendations

"Cost breakdown by category"

# Groups all expenses by account/category → ranks by total →
# calculates each as % of revenue → flags any category growing
# faster than revenue

"Job profitability ranking"

# For each job: revenue, direct costs, margin, margin %
# Sorted by margin % ascending (worst first)
# Flags any job with margin < 30% or cost > revenue