A routing model
The model hadn't been rebuilt in years. The new one forecasts more than six times tighter and sets the routing every day.
Before a number drives a decision, I check it against the raw data.
For more than ten years I've built analytics at Moody's, Morgan Stanley, HomeServe, and now National Debt Relief, mostly for lead routing, agent performance, and compliance. After hours I build software with AI coding agents, run a small AI automation practice, and keep a deal model ready in case the right small business comes up for sale.
Now: Revenue Analytics Manager, Data and AI, at National Debt Relief
Four employers, newest first. The tools changed at each one. The habit didn't.
2025to present
I'm a manager on the revenue analytics team. I run test-and-learns and analysis on day-to-day operations and the KPIs behind them, and I use ChatGPT Enterprise and Claude Code every day. I built the lead-routing model that sets daily routing configuration and the compliance monitor that checks dial timing in all 50 states.
2024to 2025
I owned forecasting, monthly reporting, and the budget for a subscription business. I built an Excel validation tool with an LLM behind it and cleaned up the Alteryx and Dataiku pipelines that fed Tableau and Power BI.
2020to 2023
I rolled out self-service Tableau reporting, and centralized teams used it in place of one-off requests. I wrote the SQL that automated revenue-cycle analytics for senior leadership.
2015to 2020
I led a Six Sigma project that improved operational efficiency by 50%. I delivered SOX and Dodd-Frank regulatory reporting and ran a firm-wide access program to 100% adherence.
What I work in: SQL (Snowflake), Python, R, Sigma, Tableau, Power BI, Git and MCP.
Four times the summary said one thing and the raw data said another.
The model hadn't been rebuilt in years. The new one forecasts more than six times tighter and sets the routing every day.
So it missed the exact failures it was built to catch. I rebuilt it on the right timestamp, and it now flags potential violations in every state we call.
Contact rates dropped in every group halfway through the test, so the strategy was not what changed. Many of the numbers we dialed were flagged as spam before anyone picked up.
Sellers show you the number they want you to see. I asked for four months of raw statements, and the lease, not the earnings, is why I walked away.
The same habit, after hours. Some of it is live, some is still being built, and one is a deal I walked away from.
A book
A hardcover book about the day a child was baptized, with the child's name in the story.
Workfile software for residential appraisers, built on PostgreSQL and PostGIS to the MISMO 3.6, UAD 3.6, and USPAP standards. It handles comp justification, adjustment rationale, and automated QC flags.
Live and still being built, with inbound interest from appraisers.
propertyappraiserpro.comNYC parcel zoning lookup with links to the source data. Check a lot’s zoning district and development potential before deeper feasibility research.
nyrezone.comAn AI automation and lead generation practice for small businesses in Westchester and NYC, using the same workflow audits I run at larger companies. Its free starter kit, Main Street MCP, lets a business's AI assistant answer from the business's real information.
Early, no clients yet.
Get the free starter kitThe fun one. Upload a golf or baseball swing, and AI roasts it, then gives you the real fix. There's a web app, and a native iPhone app that reads your pose on the phone.
Live on the web, and the iPhone app is in TestFlight.
shankdit.comA YouTube channel of long-form ambient and lofi mixes, produced end to end with an AI music pipeline.
Listen on YouTubeI ran full financial due diligence on a family-owned retail business: a proof of cash against raw statements, normalized seller's discretionary earnings, a twelve-month cash flow model across financing structures, and a lease risk analysis. The lease is why I walked away.
I passed at the terms on the table. The model is ready if better ones come back.
Most of this was built with AI agents. First what I wrote for that work, then the agents themselves and what each one has actually done.
My always-on assistant for research, scheduled jobs, and project tracking, built on the OpenAI, Gemini, and ElevenLabs APIs.
ainsley on GitHubThe structured knowledge base I use to work with LLMs day to day.
neuron on GitHubThe Claude Code skills I use in my own work.
Download the skill libraryA .zip file, no signup.
A directory of MCP servers for legal work, with its own MCP endpoint.
legalaimcp on GitHubAn MCP server that answers subscription revenue and churn questions from RevenueCat data.
revenuecat-mcp on GitHubWhat each one has done for me, with a link wherever the work is public. Together they keep twenty projects running, and the whole point is to be needed less often.
The long readNeeded less oftenHow I run twenty projects with AI agents and keep my own part of the job small. Real numbers from my own records, on a page that checks itself.Read the storyThree rules I keep at the day job and after it.
First I pin down the objective and the output we need. Then I map how the work happens today, down to the person who reads a sticky note and types it into a system, and every point where it stalls. That's where the bugs and the fixes are. Fifteen minutes saved on a daily task is more than 60 hours a year.
Every decision has a constraint someone already knows, like a budget, a floor, or a deadline. It usually comes out halfway through the analysis. I ask for it first, so options that were never available stop taking time.
When the data doesn't support the case, I say so before anyone builds a slide around it. Then I show the closest real example, because a real example holds up when someone checks the work.
I'm a manager on the revenue analytics team at National Debt Relief. After hours I'm building Property Appraiser Pro and the NYClaw starter kit, and helping a friend who runs an electrical business put AI on his phones. It answers calls, books appointments, sends notifications, and handles the common questions about estimates. I'm still looking at small businesses to buy. I still ask for the raw statements, and I still walk away when the terms don't work.
The best way to reach me
gbrussich52@gmail.comI post longer write-ups on acquisitions, analytics, and building with AI on LinkedIn.
If you're hiring for a role, a retainer, or a fixed assignment, the short version is at gianibrussich.com/hire.
Start with the live products: NYRezone for NYC zoning research, Property Appraiser Pro for residential appraisal workfiles, and LegalAIMCP for legal MCP tools. The Built section describes each one, and the Tools section links to the public code. These are my own projects, not client case studies.
For small business AI automation in Westchester and NYC, start with NYClaw. Its free starter kit, Main Street MCP, connects an AI assistant to a business's own information. The practice is early and has no clients yet. To discuss a workflow, email me the task, the tools you use, and what a useful result would look like.
Nine Claude Code skills I use in my own work: single-use data dashboards and autonomous-loop briefs, multi-agent workflows, landing-page conversion audits, AI-search visibility, brand strategy, Supabase error fixes, last-30-days research, diagram drawing, and reverse engineering. The last three are other people's open-source (MIT) skills that I use daily, bundled with credit. It's a .zip file with no signup.
I'm a manager on the revenue analytics team for Data and AI at National Debt Relief, where I rebuilt the lead-routing model and the 50-state compliance monitor described on this page. Before that I worked in revenue management at HomeServe, strategy analytics at Morgan Stanley, and operations at Moody's. Outside work I build Property Appraiser Pro for real estate appraisers and run NYClaw, an AI automation practice for small businesses.
Property Appraiser Pro (workfile software for real estate appraisers), NYRezone (NYC parcel zoning lookup), NYClaw (AI automation for small businesses) and its free Main Street MCP starter kit, LegalAIMCP (a directory of MCP servers for legal work), The Slow Pour (an AI-produced ambient and lofi YouTube channel), Shankdit (an app where AI roasts your golf or baseball swing, then gives you the fix), and Storied & Blessed (a hardcover baptism book with the child's name in the story).
Claude Code, Claude, ChatGPT, Gemini, Google Stitch, OpenAI Codex, Grok Build, Grok Bot, Grok Voice, OpenClaw, Muse, ElevenLabs, and Suno. The Tools section on this page says what I've done with each one and links to the public work.