TITAN

Labor from 50% to 30% of revenue. Cost per hire from $1,000 to $150.

Titan runs the reporting, quoting, invoicing, collections and recruiting for two home-service businesses. Pink's Windows' franchisor licensed it for all 105 locations.

What it does

I run two Pink's Windows locations and my wife runs a The Designery franchise. Titan is what we actually use to run them day to day: the dashboard we check each morning, the report that lands in the inbox twice a day so nobody has to go looking, the customized agents the teams use to improve their work ad hoc, and the workflows that massively cut our recruiting, invoicing and collections costs.

The whole system is built and accelerated by AI coding agents. It can be stood up as an MVP for another operating brand in a week, and made operational within a month once blockers are cleared.

Titan Pulse dashboard showing quotes sent, conversion rate, invoiced and collected totals, gross margin and dollars per van per day for Pink's Hudson Valley.
Pulse. Every number that runs the day: quotes sent, conversion, invoiced, collected, gross margin, dollars per van per day. Dollar figures are blurred here; everything else is live.

The 6pm email. The same numbers come to us — every visit from that day, and which ones still need invoicing or still need paying. This is the lever that shortened our cash cycle.

SqueeGPT answering a question about removing oily residue from a window, citing Pink's own sales and operations training.
Agents. The crew asks a question and gets our answer, not a generic one. It's built from Pink's training, our SOPs and call transcripts, and rebuilt nightly based on uploaded inputs. GMs use the same tool to pull data from our P&L, bank transactions, CRM and marketing channels.
Titan dispatch view showing the best days to book a job in the Kingston territory with open slot counts.
Dispatch. We can find a spot to schedule a job near another job for routing efficiency, or alternatively pull forward a future job to a spot that just opened up.
Text-in candidate — screening complete Advance
Friendliness★★★
Hustle★★
Quality★★★

No red flags

Summary: Strong candidate in the right town, with years of relevant outdoor work. Clear communication and a well-thought-out answer to the customer-service scenario. All requirements met, no concerns identified.

Assistant Thanks for reaching out. A few quick screening questions before our GM sets up a time to meet — where are you located?

Candidate I'm about ten minutes from the yard.

★★★ — right by the van parking spot

Recruiting. Someone texts the hiring line, an agent screens them on the spot and scores them against the three things we actually hire for, then hands the GM a candidate and a transcript. This is what took cost per hire from about $1,000 to $100–150. Example output — candidate details replaced.
Workflows
Recruiting, quote follow-up, review requests, and overnight reconciliation that fixes data discrepancies before anyone sees them.

What it changed

Labor as a share of revenue50% 30%
Invoiced-to-paid~100%
Bad debt recovered~$20K
Cost per hire$1,000 $100–150
Average ticket, Hudson Valley$1,384
System average ticket$600
Hudson Valley, first year~$760K
Locations licensed105

The franchisor licensed it

Pink's licensed what we built and rolled it out across the network, with individual locations able to subscribe to additional reports and workflows on top. It's the same system in every location, which is the part that matters: these results are reproducible, not a one-off.

How it's built

Every piece of work travels the same circuit — asked for, planned, or turned up by the overnight run. Scroll to run it.

One piece of work

In at the top. Back to whoever needed it.

Three ways in: a request in Titan, a plan on the roadmap, or something the nightly run turned up. The deploy reads the commit and closes the item itself.
Many at once

Agents work in parallel. Merges don't.

37 isolated git worktrees, each its own Claude Code session in tmux. claim → status → resume → release: a session can die, the work can't. One label-driven queue to main.
Staying in the loop

One screen for all of it. They email when they're stuck.

A web terminal and status.sh over Tailscale — every session from a phone. Agents email on a blocked gate or a green PR, with the whole decision in the body.
Overnight

It checks itself, and fixes what it can.

Windmill, 3am: Jobber vs BigQuery, drift backfilled on the spot. 4am: error scan, Claude writes the analysis. 5am: the day's plan, waiting by breakfast.
Next

A coordinator keeps it moving. You come in for the calls that matter.

A coordinator agent drives the sessions, clears the small bugs itself, and ships behind verify-and-auto-rollback — surfacing only where a human decision is genuinely required. Dashed = built, not yet switched on.
Asked for team · Missive Planned roadmap · KK Discovered Pulse · nightly recon >_ >_ >_ >_ resumed 37 worktrees claim · status · resume · release Safety Arch Tests KK Intake one queue, three sources Claim worktree + tmux Build Claude Code · beads Review 4 lenses, before the PR CI GitHub Actions · live preview Queue one at a time Deploy ~2 min to live Landed asker emailed, item closed Dashboard web terminal · status.sh every session, from your phone “PR #786 is green — merge?” Agent email blocked · decision · PR green the whole call, in the body 3am Reconcile 4am Health audit 5am Tomorrow's plan Jobber ↔ BigQuery Coordinator keeps it moving Asks only here Auto-rollback verify, or revert Reply → session answer the email, the agent resumes Files its own PR 4am scan → branch

What the circuit changes

The dev team shrank to one. The work didn't.

A conventional build puts a team between the person who understands the business and the code that runs it. This one doesn't. The work runs in parallel and the operator drops in periodically — which moves velocity, cost and review in the same direction at once.

Who's involved Shipped per week Review per change Your attention Cost per change Then you Seven people, and you outside them A dozen, split across the team One pass, if there's time Spent explaining, not building A sprint Now you agents you drive One person, and the fleet they run 23 merged, by one person Four lenses, then 11 CI checks A handful of check-ins An afternoon Next you asks a tree that runs itself Nobody driving, someone deciding The same, plus the night shift Same four, plus the coordinator's Only where a call is needed Unattended
One dot, one merged change. Then is a conventional seven-person build, indicative only. Now is measured: 181 pull requests merged over the last eight weeks — ~23 a week, by one person. Next is built and not yet switched on.

Velocity usually costs you review. Here it buys more of it — the lenses run in parallel too, so every one of those 752 merged pull requests got a harder look than a rushed team would give it.

Overnight three jobs run on their own: reconcile our data against Jobber and BigQuery and backfill whatever's missing, scan for errors, and draft the next day's plan. The next step is a coordinator that keeps the sessions moving without me, and only surfaces when a decision actually needs a human.

Getting it running

The question I get most: what would it take to point this at a different business?

MVP for another operating brandone week
Operational, blockers clearedwithin a month
What I need from youone conversation

That conversation is about finding the highest-leverage number in your business. For Pink's it was labor as a share of margin — we knew we were overspending on labor per job, so that's what we built toward first. Yours will be something else, and it's worth an hour to find out what.

Two ways to run it

One business

Build fee + usage-based license

A custom implementation pointed at your operation, your data sources and your numbers. A build fee to stand it up, then a license scaled to the functionality you turn on.

A franchise network

Build fee + per-location monthly

The Pink's model. One build, deployed multi-tenant, then a monthly per-location license tiered by functionality — with ongoing improvements and new workflows as they earn their place.

We start with a scoped proposal, or a working demo on your own data, so you can judge it before committing to anything. If you'd rather just see it, the base tier is free — connect your field-service system and the dashboard and the daily reports run on your own numbers.