Cut billing reconciliation time 40% on a ~$1M/month billing system

A ~175-person studio needed invoices it could trust

Cowlar Design Studio is a product-development and consulting studio of about 175 people, the services arm of Cowlar (Y Combinator W17), working across IoT, AI/ML, SaaS and AgTech. It bills fractionally: each client pays for its share of the engineering time and expenses spent across many projects running at once. Close to $1M a month went out through that system.

Leadership needed two things from the data: invoices it could put in front of a paying client, and a view of which projects made money. Both depended on hours and expenses being recorded correctly. A quiet error became a wrong invoice, or billable work that never reached an invoice at all.

Before: inconsistent logs and spreadsheets weeks out of date

When I joined as Billing & Compliance Analyst in May 2024, the data had four problems:

• Revenue leakage. Engineers logged hours inconsistently, and specialised work went unbilled.
• Manual reporting. Financial health came from consolidated spreadsheets that were often a week to a month out of date.
• Data friction. Collecting data by hand interrupted engineers, and data hygiene suffered.
• No profitability view. Resource allocation ran on intuition. Nobody could see which projects consumed the most unbilled time.

The volume was too high to check by hand and the stakes too high to leave unchecked.

How I rebuilt the data layer

One set of SQL datasets replaced per-team spreadsheets

I designed and maintained more than 15 custom SQL datasets between engineering output and finance. Billing, invoicing and reporting all read from them. The alternative was the existing habit: each team consolidating its own spreadsheet and reconciling the differences afterwards. With one set of modelled tables, a disagreement had one place to be settled.

Automated capture worked better than asking engineers to log more carefully

The obvious fix for poor logging was stricter rules for engineers. I rejected it, because the interruption was the cause of the bad data. n8n and Google Apps Script workflows collected time and expense data in the background, so engineers no longer had to stop and enter it. Data quality improved once collection stopped depending on busy people remembering to do it.

A QA/UAT gate caught errors before a client saw them

Between the datasets and the reporting I put a QA/UAT stage: written workflows and SOPs that checked billing data for accuracy and completeness. Correcting an invoice after a client has seen it costs trust and delays payment; catching the error before costs one check. The SOPs also kept the process from depending on my memory, and they left audit-ready records behind.

Power BI replaced the weekly look-back

On top of the checked datasets I built Power BI dashboards for project financials and profitability. They replaced manual spreadsheet consolidation and moved reporting from weekly look-backs to real time, so leadership could see where engineering hours were going while there was still time to act.

Reconciliation time fell 40% and billing accuracy rose about 25%

• Financial reconciliation took 40% less time once the Power BI dashboards replaced manual consolidation.
• Billing accuracy improved by about 25% after the automated capture went live.
• Resource allocation improved by 15% once leadership could see profitability by project.
• Close to $1M a month ran through the system while I owned its data.

These are the figures I reported at the time. The before-and-after measurements live in a former employer’s systems, so I can’t show the underlying counts here.

What I’d tell the next team

Fix the collection step before the reports. Most billing errors here started when an hour was logged, long before anyone opened a dashboard. Automate capture so the people doing the work are not also keeping the record, put a written check between the data and the invoice, and then invest in reporting. A dashboard built on unchecked data only shows the errors faster.

Limitations and what I’d do next

• The gains are reported figures, without a published method. Next time I would record the measurement alongside the result: hours per reconciliation cycle before and after, and what counted as a billing error.
• The data layer leaned on one analyst. The SOPs reduced that risk; a second trained owner would have removed it.
• Apps Script and spreadsheet tooling have ceilings. They suited a process that changed often at this size. At higher volume I would move capture into a monitored pipeline with alerting on failed runs.
• Rule-based QA only catches errors someone wrote a rule for. A periodic sample of invoices traced back to the raw logs would catch what the rules miss.

Team and credits

I was the main contact for billing and compliance questions across finance, HR and project teams. I owned the SQL datasets, the QA/UAT process and SOPs, the Power BI reporting, and the n8n and Apps Script automation. Engineering, finance and leadership supplied the source data and used the outputs. The HR agent I built at the same studio fed worklogs into this pipeline.

The billing figures belong to a former employer, and no client names or client data appear here. This page describes the pipeline instead of charting their numbers.

Technical appendix

Pipeline stages

Stage	Tools	What it did
Capture	n8n, Google Apps Script	Collected project time and expense data without manual entry by engineers
Store	SQL (15+ custom datasets)	Analytics-ready tables shared by billing, invoicing and reporting
Check	QA/UAT workflows, SOPs	Accuracy and completeness checks on billing data; audit-ready records
Report	Power BI	Project financials and profitability, refreshed in real time
Use	Invoicing, leadership reviews	Client invoices; decisions on where engineering hours went

Wider automation. Alongside the billing pipeline I automated HR, accounts and billing workflows with Python, JavaScript and Google Apps Script, Slack, Odoo and Google Workspace, which cut manual data entry across departments.

What is left out, and why. Table schemas, query logic, dashboard screenshots and any invoice-level data stay with the former employer.