A Municipal Tax Software Firm Won Back 100+ Staff Hours/Mo
~75%
2–4 months
Not disclosed
A software platform that helps municipalities manage properties in tax arrears was running its entire operation on a legacy Microsoft Access dashboard its new private-equity-backed leadership considered unscalable. Every property in arrears is a tracked task: legal letters, lawyer-assisted notices, door-knocking, and home-sale flows, each generating documents that staff handled manually alongside Excel-based side processes. Growth meant headcount, because the system could not absorb more work.
Lazer rebuilt the product from scratch over roughly six months of design and engineering: a standalone CRM with separate experiences for internal staff and a read-only municipality view, so city users can check status without touching the internal toolset. Design and engineering ran with deliberate overlap so feasibility was checked in real time on the workflow-heavy flows that had previously been explained through videos and deep-dive sessions.
The AI layer transformed intake: LLM-based OCR parses incoming documents, automated intake emails feed the pipeline, extracted text is grammar-corrected, addresses are validated against Google Maps, and new case files are created automatically. Document output proved one of the hardest problems, where both speed and formatting accuracy mattered; the team pivoted from an initial e-signature-platform approach to LibreOffice-based template generation, alongside invoice generation. The system shipped with staging and production environments and exceeded the original scope.
Lazer rebuilt the product from scratch over roughly six months. Rather than patch the legacy Microsoft Access dashboard, the team reverse-engineered workflows that lived partly in software and partly in staff habit, running design and engineering with deliberate overlap so feasibility could be checked in real time on the workflow-heavy flows previously explained through videos and deep-dive sessions. They structured the application as a standalone CRM with separate experiences for internal staff and a read-only municipality view. For intake, they layered LLM-based OCR over automated intake emails, added grammar correction on extracted text, and validated addresses against Google Maps before auto-creating case files. Document output proved one of the hardest problems, since both speed and formatting accuracy mattered; the team pivoted from an initial e-signature-platform approach to LibreOffice-based template generation, and added invoice generation alongside it. To stay resilient when individual model reliability wavered under load, they routed OCR across multiple models. The build shipped with staging and production environments and exceeded the original scope.
Small and mid-size B2B software companies, especially PE-backed ones, running on legacy systems where document-heavy workflows cap growth at current headcount.
100+ Employee-Hours Saved Per Month
Hundreds of Documents Auto-Ingested
More Clients, Same Headcount
Aanikh Kler
Lazer replaced the company's manual document handling with automated intake: LLM-based OCR parses incoming documents, corrects grammar in extracted text, validates addresses, and auto-creates case files. This recovered more than 100 staff hours per month.
The intake pipeline used LLM-based OCR with multi-model routing via a LiteLLM-based router across OpenAI and Anthropic models, a React/Next.js frontend, a Python FastAPI and Celery backend, Supabase, and Google Maps for address validation.
The platform saved more than 100 staff hours a month, auto-ingested hundreds of documents from email into parsed, validated, grammar-corrected case files, and let the firm take on more municipal clients without proportional hiring.
Lazer rebuilt the product from scratch over roughly six months of design and engineering, with results landing in about four to six months.
It fits small and mid-size B2B software companies, especially private equity-backed ones, running on legacy systems where document-heavy workflows cap growth at current headcount.
Have a similar challenge?
