Engagements
Operators, not advisors, means the work has receipts. Six engagements, told the same way we'd tell them to a partner: what was broken, what we built, what changed.
"Pat and his team quite literally solved a $100M bottleneck for us. Hands-on, and a game changer."
SVP Operations · Grubhub
"Brought them in to implement automation across the org, specifically the sales ecosystem. Revenue 2x after the 3rd month."
A.F., Founder & CEO · PVG
A regional/OTR carrier: 80 power units, roughly 9,800 loads a year, mostly brokered freight. Two invisible leaks, money earned but never billed, and loads assigned that drivers couldn't legally finish. Two agents, one on each leak. The common thread: no human can watch 9,800 loads' worth of clocks in real time. An agent can.
The leak: drivers sat at docks for hours, but detention almost never got billed. Notes died in text threads, BOLs went unsigned, and brokers denied claims for lack of proof. Roughly $138K a year in detention was owed; about 25% was collected. With untracked TONU, layover and lumper charges, ~$143K a year leaked.
The agent: sits on the telematics feed and the TMS. Geofence timestamps mark arrival and departure, the rate con sets the free time, and the meter runs itself. The moment billable detention closes, it builds the invoice with a timestamped evidence packet: GPS pings, dwell duration, map snapshot, appointment reference. The packet kills the broker's denial.
Detection went to 100% of events. Capture rate from ~25% to ~80%. ~$150K a year in accessorials now collected, roughly +$110K that used to be written off, with near-zero added labor.
The leak: dispatchers assigned loads off gut and a map, not the driver's remaining legal clock. ~2% of loads ended in a feasibility failure: HOS violations (~$80K/yr) and late deliveries (~$62K/yr). On-time delivery sat at 94% against a 95% contractual floor, putting a major lane at risk.
The agent: checks every assignment before it's tendered. Pulls live HOS clocks from the ELD, runs the route through a truck-routing engine, adds mandatory breaks and real dwell, and returns a green, yellow or red flag right on the dispatch board, with the fix suggested when it's red: team, relay, or move the appointment.
HOS events down ~80%. Feasibility-rooted lates down ~70%. ~$105K a year in hard savings, and OTD climbed from 94% to ~98%, clearing the contract floor that actually mattered.
Residential and light commercial cleaning plus a wash-dry-fold laundry pickup and delivery arm. About 45 cleaners across 15 crews, a laundry plant, 6 vans, ~1,800 recurring cleaning accounts, roughly $7M in revenue. The whole business runs on the recurring book, and the book was leaking quietly. Two agents, one on each leak.
The leak: around 6% of recurring charges failed — expired cards, insufficient funds, hard declines — and only ~20% ever got recovered. The rest silently fell off the book. Add the customers quietly skipping cleans or downgrading with nobody watching the signals, and roughly $400K of run-rate revenue was walking out every year over fixable problems.
The agent: the second a charge fails, it classifies why and acts. Expired cards refresh automatically. Insufficient funds goes on a smart retry schedule timed to paydays instead of hammering a dead card. Hard declines trigger a text and email with a one-tap payment-update link before the next clean gets skipped, and anything unresolved escalates to a human with full context. It also scores every account for churn risk — skips, cadence drift, downgrades, complaints — and drops a retention task with a suggested play while the customer is still saveable.
Failed-card recovery from ~20% to ~70%, about $300K of run-rate revenue saved. Early intervention kept another ~45 at-risk accounts, ~$126K. Combined: $400K+ a year retained, recurring revenue that gets bought at a multiple when the business sells.
The leak: routes were built in a spreadsheet by zip code and gut feel. Cleaning crews crisscrossed the metro with 22% of paid hours burned on travel, badly sequenced days spilled into overtime, laundry vans missed their promised windows 15% of the time, and everything hit the plant in one late-afternoon slug that blew next-day turnaround on 12% of orders.
The agent: solves each day as a routing problem. It clusters jobs by geography and time window, sequences every crew and van to minimize drive time, respects skill and equipment requirements, and re-sequences the rest of the day on the fly when a job cancels or a crew runs long. The constraint most route tools ignore: it treats the plant's hourly throughput as a hard limit, spreading laundry intake across the day so the plant gets a steady feed instead of a 4pm avalanche.
Jobs per crew per day from 4.2 to 4.8 with travel time down to ~15% — roughly 2,250 extra job slots a year, zero new hires. Vans from 24 to 28 stops a day, window hit rate 85% to 97%, turnaround misses 12% to 3%, overtime down ~30%.
A $60M firm: ~110 fee-earning timekeepers, ~2,500 matters a year, half the revenue governed by institutional clients' billing rules. Every law firm's money story is one equation — hours worked → hours captured → hours billed → dollars collected — and money leaks at every arrow. No human can watch 110 timekeepers and every invoice line in real time. Three agents, one per arrow. One rule throughout: the agent drafts, the lawyer attests. Nothing bills, clears, or sends itself.
The leak: attorneys reconstructed their time from memory days later. The 9-minute client call, the 20-minute document review, the three quick emails — real billable work, never written down. Roughly 7,260 billable hours a year were worked and never recorded, ~$2.4M that never made it onto an invoice.
The agent: quietly rebuilds each timekeeper's day from email, calendar, documents, calls and the docket, maps every activity to the right client and matter, and drafts compliant entries with specific narratives. The attorney does a 5-minute review at end of day instead of reconstructing a week from scratch.
~5,100 hours a year confirmed and captured. ~$1.7M in newly collected revenue, near-zero added labor.
The leak: institutional clients enforce billing guidelines through e-billing platforms that quietly shave invoices — lumped entries, vague narratives, unapproved timekeepers, rate caps. About 5% of $30M in governed billings was getting written down: ~$1.5M a year, plus the hours burned fighting rejections.
The agent: checks every line of every draft invoice against that specific client's rules before it leaves the building, flags what would bounce, and offers a compliant rewrite. The invoice sails through instead of coming back shaved.
Write-downs from ~5% to ~1.5%: ~$1.05M a year recovered, and invoices approve faster, pulling cash in sooner.
The leak: ~$4.8M billed but never collected, a real chunk of it simply under-chased — partners hate dunning their own clients, so invoices aged and cash sat locked up ~95 days.
The agent: watches AR aging, segments by payment history and relationship sensitivity, and drafts a tone-matched follow-up — gentle for good clients, firmer for chronic late payers. Sensitive accounts go to the partner as a ready-to-send draft with full context; routine ones go to finance.
Collection realization from ~92% to ~95%: ~$1.5M+ a year recovered, with AR days down 12 to 15 — cash freed, distributions sooner.
The same engagement also stood up instant conflicts screening at intake, AI-assisted document review, and matter-budget overrun alerts — the long tail behind the three above.
PVG develops padel facilities and advises on the capital behind them. Small senior team, every lead high-value, every dropped follow-up expensive. Their goal was simple: grow pipeline without growing headcount. We rebuilt their sales ecosystem around a production AI agent that lives in the team's chat.
Lead flow looked healthy at the top, but follow-up lived in heads, text threads and a spreadsheet. Leads went quiet and nobody noticed. The books never matched reality, and the team was working volume while quality leaked out the side.
Audited every step from lead capture to close. Built an agent wired into the CRM, email, website forms, analytics and team chat: it captures every inbound lead in seconds, assigns an owner, posts a daily follow-up brief, researches prospects, and logs every touch. The team talks to it in plain English. Humans take over the moment a prospect replies.
Revenue 2x by month three. $14.5M in qualified pipeline built in the agent's first 3.5 weeks. Not one lead has gone quiet without a follow-up since. Runs for under $900 a month.
Brought in on a project basis to solve a margin-bleed problem across several delivery markets. Average basket value per driver inside the assigned mile-radius wasn't covering the cost to deliver, eating 7 to 10% of margin and roughly $100M off the bottom line.
The unit economics on delivery weren't holding: basket value per driver, given the radius they were running, didn't justify the cost of the trip. Standard playbook fixes, driver pay tweaks and fee adjustments, weren't moving the needle.
Reframed the problem from "driver economics" to "radius logic." Redesigned how delivery radius was defined per market and how driver orders were batched and assigned, accounting for basket density, drive-time and zone overlap instead of treating radius as a fixed input.
Margin bleed stopped. Affected markets moved from negative to +1.8% on margin, a ~12-point swing on the line that mattered.
A small industrial equipment business (XRF analyzers: sales, rentals, trade-ins) with deep expertise and almost no digital footprint. Buyers searching for exactly what they sell couldn't find them. Engaged through the Diagnose week, now in Build.
No analytics, no lead capture, no way to know what was working. The website didn't match how buyers actually search, and growth depended entirely on relationships and luck.
Diagnose week surfaced the gaps and ranked the fixes. Rebuilt the website around buyer intent (applications, brands, rent vs buy), instrumented everything with analytics and search tracking, and stood up structured lead capture with an assessment funnel.
Engagement in progress. Site live and fully measurable, built to convert search intent into pipeline. Results get reported as they land, that's the deal we make with every client.