2026 Update
This case study remains our most-cited example of "small tool, big impact." In 2026, we've since added AI-powered anomaly detection to this client's system, further reducing manual review to under 2 hours/week.
A Logistics Manager was losing 43 hours/week to manual data entry between incompatible systems. Slickrock.dev built a $15k Python-based validator that automated the process, saving $60k/year and delivering a 3-month ROI payback period.
TL;DR
We automated the "grunt work" of data entry and reconciliation. The result? 43 hours/week freed up for strategy. Small tools yield massive compound leverage.
The Client: A Regional Freight Brokerage
The client was a mid-sized freight brokerage with 45 employees, processing approximately 800 shipments per week. They had been in business for 12 years and had grown organically—which meant their tech stack had grown organically too.
Their Logistics Manager, let's call her Maria, had been with the company for 6 years. She was brilliant at her job: negotiating rates, managing carrier relationships, and solving problems. But she was spending half her week on something that had nothing to do with her actual skills.
The Problem: The Human Bridge
Maria was acting as a human API between two systems that refused to talk to each other:
- The Accounting System (QuickBooks Desktop): Where invoices lived, exported as CSV files with one format.
- The Shipping Provider (proprietary TMS): Where manifests lived, exported as CSV files with a completely different format.
Every week, Maria would export both CSVs, open them side-by-side in Excel, and manually match each invoice to its corresponding manifest. She was looking for discrepancies: wrong weights, wrong addresses, wrong fuel surcharges. If she found one, she had to flag it, investigate it, and fix it before the month closed.
Why Existing Tools Failed
Before coming to us, they had tried three approaches:
- Excel VLOOKUP: Worked for exact matches, but the invoice numbers were formatted differently between systems (trailing zeros, dashes vs. underscores). Failure rate was too high.
- Zapier Integration: Couldn't handle the CSV batch upload pattern. Their TMS didn't have a proper API, just CSV exports. Zapier needs real-time triggers.
- Hiring a Data Entry Clerk: They tried. The clerk quit after 3 months. "Soul-crushing" was the exit interview feedback.
Key Insight
The Real Cost of Manual Work: Maria was a senior employee earning $75k/year. At 43 hours/week on data entry, they were paying $37.50/hour for someone to do work that should cost $0.00037 per operation with software. That's a 100,000x cost inefficiency.
The Solution: The Validator
We built a custom web application we called "The Validator." The tech stack was intentionally simple:
- Backend: Python 3.11 with Flask for the API layer
- Matching Engine: RapidFuzz library for fuzzy string matching
- Frontend: Tailwind CSS with HTMX for interactivity (no React needed)
- Database: SQLite for local storage (graduated to PostgreSQL later)
- Hosting: Fly.io at $5/month
How It Works: The Workflow
Upload CSVs
Maria drags both files onto the web interface. The app normalizes column headers automatically (it learned the mapping during setup).
Fuzzy Matching
The engine runs RapidFuzz on invoice numbers with a 92% similarity threshold. This catches 'INV-00123' vs 'INV123' vs '00123-INV' as the same record.
Discrepancy Flagging
Any row where weight differs by >5%, or amounts differ by >$10, gets flagged for human review.
Exception Queue
Maria reviews only the flagged items (typically 15-20 per week, down from 3,200). She approves, rejects, or escalates.
Export Report
A clean PDF summary generates automatically for the accounting team.
The Edge Cases We Had to Solve
No real-world data is clean. Here's what we hit during the build:
- Date Format Hell: QuickBooks exported MM/DD/YYYY. The TMS exported DD-Mon-YY. We wrote a date normalizer that handled 7 different formats.
- The "Split Shipment" Problem: One invoice sometimes covered multiple manifests. We added a one-to-many matching mode with aggregate validation.
- Character Encoding: The TMS exported in Windows-1252. QuickBooks exported in UTF-8. We auto-detected encoding using chardet.
- The Phantom Zero: Excel was silently stripping leading zeros from invoice numbers when users opened the CSV. We added a warning banner when this was detected.
The Results: Before & After
The first week The Validator went live, it flagged an anomaly Maria hadn't noticed in her manual process: a carrier had been double-billing fuel surcharges for 3 months. Total recovered: $12,400.
""I got my weekend back. And I actually caught a $12,000 billing error the first week we ran it. The app paid for itself before we even got the final invoice from Slickrock."
"
What Maria Does Now
With 41 hours/week freed up, Maria was able to refocus on high-value work:
Verification Checklist
- Renegotiated carrier contracts, saving $180k/year in shipping costs
- Built a new carrier onboarding program that reduced churn by 40%
- Trained two junior staff members who now handle expanded operations
- The company promoted her to Director of Operations
The ROI wasn't just the $60k in salary efficiency. It was unlocking a senior employee's potential that had been trapped in busywork.
The Investment
Build Cost: $15,000 (4-week sprint)
Monthly Hosting: $5/month (Fly.io)
Year 1 Maintenance: $2,000 (mostly adding the LLM anomaly detection in 2026)
Payback Period: 3 months
Key Insight
The Leverage Formula: $15k investment → $60k/year savings + $12k error recovery + $180k contract savings = 16x return in Year 1. This is why we say apps are assets, not expenses.
Lessons Learned
- Start Small: We didn't build an ERP. We built a single-purpose tool that did one job perfectly.
- Embrace "Good Enough": 92% fuzzy match threshold was better than 100% exact match that missed edge cases.
- Design for the Exception: The UI was optimized for the 20 exceptions, not the 3,180 automatic matches.
- Data Quality Reveals Itself: Automating the process exposed data quality issues that had been invisible in the manual workflow.
Conclusion
Look for the boring work. That is where the gold is buried. The tasks that make smart people feel stupid are often the tasks that should have been automated years ago.
If you have a Maria in your organization—someone spending half their week as a human API between incompatible systems—you have a 16x ROI opportunity waiting to be unlocked.
Read This Next
Think of Custom Apps as Assets, Not Expenses
Reframing cost logic—proprietary software adds valuation, subscriptions drain it.
Escape Per‑User Pricing and Own Your Platform
Break free from verified user taxes. Gain control, data ownership, and predictable costs by building your own platform.
When to Fire Your SaaS and Build Your Own
A practical decision framework and cost indicators for making the switch.
