Every clinic, branch, and restaurant throws invoices, claims, statements, and tickets at a central team that codes and reconciles by hand. The bottleneck isn’t people — it’s the layer between them.
The Thesis
The highest-leverage AI work in a multi-site business sits at the points where paper enters the company.
Agents that act on documents at the moment of capture — not chat tools bolted onto the stack. The pattern came out of claims automation for a US dental group, and it has held in every multi-site business since.
The Principle
Deterministic pipelines do the work, AI reads and classifies, and your ERP stays exactly where it is — the team experiences less work, not a new tool.
What It Looks Like
A morning on the command center of a 300-location group — nobody at headquarters touched any of this.
The work of forty back-office staff at a fraction of the cost — every location, every document, all day.
The Playbook
The same funnel pattern, applied where multi-site businesses leak the most.
annualized capture — 16 offices
of claims routed end-to-end
your best collector’s judgment, every interaction
from cancellation to backfilled slot
Collections, Properly
A collector copilot plus four specialist agents, with instant mid-call hand-off.
A filing surfaces mid-call and the copilot switches posture — what can be said, and what happens next.
The rules for all 55 US jurisdictions, applied live on the call — not looked up after.
Each borrower matched against every program they qualify for — the right offer, the first time.
Which account, which channel, which moment — and when a case should reach a human with authority.
Outbound works the book around the clock, with a clean hand-off the moment a conversation warrants a collector.
Each institution gets its own tone and compliance posture — configured, not rebuilt. New rules apply the day they’re written.
Your best collector’s instincts, on every single call.
This is the platform we built with a collections platform serving banks and credit unions.
Margin Intelligence
Per-location P&L outlier briefs that name the location, the driver, and the action.
The old way
A regional manager scrolls forty location tabs hunting for the number that looks wrong. By the time it’s found, the driver has compounded.
The brief
“Location 114’s food cost ran 340bps over the group median — the driver is one supplier’s April price rise applied a month early. Recommended: dispute the invoices and re-point orders to the group contract.”
The same “material-change writer” pattern from our Finance playbook — pointed at the operating P&L.
Hospitality & Front-of-House
POS data, supplier invoices, digital orders — different paper, same funnel.
Prime cost per location in plain language — the number, the driver, one fix-this-week, in a Monday pulse.
Invoices coded and posted as they land; every quiet price rise flagged the week it happens.
Web, app, and kiosk benchmarked against best-in-class — fixes run with your IT team, not around them.
The Franchise & Dealer Layer
Dealer and franchisee portals throw their own document stream at headquarters — same pattern, different paper.
Every partner claim checked against the entitlement matrix — paid when it matches, queried when it doesn’t.
Claims processed at the door, with reserve modeling so finance sees the liability building — not the surprise.
Onboarding and training drafted from your own playbooks — new dealers ramp in weeks, not quarters.
Tier-1 partner questions answered in any language, around the clock — escalating only what needs a person.
The network scales. The central team doesn’t have to.
Worked Example
Vendor quote normalization at a $1.5B industrial distributor — the single skill behind the $6.8M.
Before
every Friday, per office
Every Friday, an analyst in each office re-keys vendor quotes arriving in fourteen formats. By the time it’s done, some have expired.
After
of review, per office
Quotes normalize and rank themselves; the analyst reviews the finished sheet. Across 16 offices, every week, that’s $6.8M a year.
The 90-Day Ops Playbook
Ninety days from first meeting to a measured decision, baseline agreed in writing.
Count every point where paper enters the company — volumes, formats, and the hours each eats, agreed in writing.
Classifiers train on 18–24 months of your own coding, matching, and resolution decisions.
The two highest-volume entry points go live with exception queues. Specialists review exceptions, not volume.
What works at five sites rolls to fifty. Per-site telemetry catches the outliers.
Measured against the week-one baseline. Company OS scoping if earned.
Where It Scales
Everything you build multiplies by the site count.
Built once at one site — every acquisition inherits the layer at close.
A back office that scales with sites, not headcount.
The Ops Dashboard
Trust is a number on a screen, not a feeling in a meeting.
by funnel
The share of each document stream that never needs a human touch.
the trust number
How long exceptions sit before a person clears them — where trust is earned.
by ticket category
Which help-desk categories resolve without a human — and which still shouldn’t.
outlier flags
Which locations lean on the layer, and which route around it.
trending down
Falling as confidence thresholds tighten and more volume clears untouched.
The operations view of our Usage & Telemetry playbook — every skill reports on itself from day one.
The bottom line
The Proof
One agent layer between every clinic and every system — 60% faster close, 94% of claims routed end-to-end.
94% reduction in manual procurement touchpoints, with 87% of exceptions resolved without a human.
Store operations intelligence across the estate — per-location signal instead of forty spreadsheet tabs.
Vendor-quote normalization replicated across 16 offices — $6.8M annualized.
The Foundation
Built into every band from day one — not bolted on after.
The Rest of the Spectrum
Ready to move
Start with one function. We’ll show you the process inventory, the skills we’d ship, and the number we’d be accountable to.
Talk to LightCI