Applied AI

Software we wrote inside our companies, with AI doing much of the work. Each tool is now run by the people who use it.

We find the problem by working in the business, then build the fix on the systems the company already owns.

01 · Market selection

Florida roofer map

The problem
Where should we grow next, and who already works there?
What we built
A map of every Florida roofer we could find in public data, each sized by its number of reviews and laid over home density and income.
The result
305 Florida roofers have 200 or more reviews. We use the map to choose markets and the companies we talk to.
Show roofers with

Loading map…

02 · Homeowner guidance

Roofus

The problem
Homeowners with storm damage have insurance questions that a roofer has to answer carefully.
What we built
A storm-claims guide for Roof Pros customers. Every answer cites a Florida statute or state guidance, and the compliance rules run in code before the model answers.
The result
Every answer carries its source, and Roofus is built not to promise coverage or tell anyone to file a claim.

How long do I have to tell my insurer about roof damage?

For a new claim, Florida law gives you one year from the date of loss to notify your insurer.

Fla. Stat. § 627.70132

I can't tell you whether your policy covers this. Your insurer or a licensed adjuster can.

Fla. Stat. § 489.147

Example of the answer format.

03 · Demand generation

The lead engine

The problem
Leads lived in four separate systems, and callbacks depended on who checked which one.
What we built
One software stack that connects Opta, AccuLynx, RingCentral and Hatch through their APIs, with dashboards for speed to lead, appointments and sales by source, run by a central call center.
The result
A central call center built in two months, working every lead from every source.
OptaAccuLynxRingCentralHatchOne software stackSpeed to leadSets and confirmationsSales by repSales by source

Four systems feed one software stack; the dashboards read from it.

04 · Back office

Commissions agent

The problem
Commission statements took days of spreadsheet work every pay period.
What we built
An agent that reads closed jobs, job profit and each rep's pay plan, then drafts the statement for a manager to approve. Nothing is paid until a person signs off.
The result
Managers review and approve statements instead of building them by hand.
How a statement is drafted, in percentages
Commission as a share of the contract10.0%

Illustrative pay plan.

05 · Search and reputation

Property pages

The problem
Years of job photos sat in CompanyCam, where homeowners searching for a roofer could never see them.
What we built
We turned 173,000 CompanyCam files into property pages and city hubs, with Claude writing the job write-ups through the batch API.
The result
4,295 property pages, 236 city hubs and 5,162 job write-ups.
City hubs

    Have a problem in your business that software should solve? We'd like to hear about it.

    Start a conversation