A brittle no-code billing stack: rescued, then rebuilt.
An industrial staffing firm ran weekly and monthly timesheets, manager approvals, QuickBooks invoicing, and PDF billing packets through a chain of five disconnected tools, until retries started cutting duplicate invoices. I stopped the bleeding, reverse-engineered every rule, and designed the purpose-built app to replace it.
Anonymized. Specifics under NDA, framing and lessons in the open.
The firm's weekly and monthly cycle (collect timesheets, route manager approvals, invoice the end clients in QuickBooks, assemble PDF billing packets, run payroll) ran across five disconnected tools, with a spreadsheet as the only system of record. Nobody could see status without opening spreadsheets. Approval routing was hard-coded per person. Collection was poll-based with brittle workarounds, and the invoicing automation had grown into a 200-step monster with dozens of retry handlers. Then the retries started creating duplicate invoices in QuickBooks. A live billing crisis.
First, I stopped the bleeding: diagnosed the retry-induced duplicates, separated the real invoices from the bad ones, cleaned out 150+ stale failed runs, and restored a clean billing week. Then I reverse-engineered ~100,000 lines of no-code logic into a clear spec and a ~25-entity data model, preserving every business rule (overtime tiers, per-diem and lodging variants, client-specific billing adjustments, even a literal "stop-OCR" separator page a client's scanner needed). And I designed and prototyped the replacement: one operations console with event-driven collection, magic-link approvals (no logins for managers or contractors), native invoicing and PDF packets, an audit trail on every change, and a Paychex payroll handoff.
The immediate fire is out: duplicate-invoice risk removed, the broken billing week recovered, the pipeline back to a clean queue. And the firm now has a clear path off the brittle stack: a designed, prototyped replacement that collapses five tools into one console, scoped as a short fixed-fee build, with every hidden rule documented and editable instead of buried in no-code branches. (The rebuild is scoped and prototyped; the incident recovery and the spec are delivered.)
[ specifics under NDA ]
What the engagement delivers.
Two things are already delivered (the incident recovery and a full reverse-engineered spec), and the purpose-built replacement is designed and prototyped. Each card below is labeled with where it stands.
Incident rescue
Found and killed the duplicate invoices, then recovered the broken billing week.
Reverse-engineering
~100k lines of no-code logic turned into a clean spec and a ~25-entity data model.
One operations console
Collect → approve → invoice → packet → payroll → close, all in a single system.
Magic-link approvals
Managers and contractors act straight from a link, no logins to manage.
Native invoicing + packets
QuickBooks invoices and native PDF packets. Drops the third-party PDF service.
Audit trail + editable rules
Every correction logged; client-specific rules made explicit, not buried in branches.
The not-sexy work that mattered most.
Anonymized, but real. The details that made the rescue safe and the rebuild trustworthy.
Killing duplicate invoices
- Retry storms had created duplicate QuickBooks invoices: a live billing crisis.
- I built the logic to tell real invoices from duplicates and safely void only the bad ones.
- Then removed the retry trap that was creating them in the first place.
Preserving the weird-but-load-bearing rules
- Per-client overtime tiers, per-diem and lodging variants: captured, not lost in the rewrite.
- Even a literal separator page inserted so an end client's scanner wouldn't misread the billing packets, the rebuild keeps it.
From polling to event-driven
- Out: fragile form-pagination polling and the brittle workarounds it needed.
- In: a real, event-driven status model the team can see at a glance.
Trustworthy by default
- Secrets in a vault. The old stack had an API key sitting in plain text.
- An audit log on every state change, plus infrastructure-as-code.
[ anonymized: specifics under NDA ]
Is your automation one retry away from a billing crisis?
If a brittle no-code chain is invoicing your customers, I'll stop the bleeding first, document every hidden rule, and scope the rebuild (and I'll be clear about what's fixed today versus what's worth building next). I reply within 48 hours on weekdays.