I want to be upfront about something before you keep reading: this is not a success story. Or at least, it’s not only a success story. It’s the kind of story where the ending isn’t fully written yet and the middle was a real mess. I did not wake up one morning and decide to become a technology company.
I run a restoration business. We do water, mold, and fire. We deal with TPAs, carriers, adjusters, property managers, commercial building owners and homeowners who are usually having one of the worst weeks of their lives. Restoration is already chaotic enough without adding another shiny object to chase. Technology, at least in my world, has always been a means to an end.
The technology in the field included Mitigate, Docusketch, our chosen CRM and of course the Google suite – Gmail, Maps, etc. Then finalizing the inspection or monitor with a group text that may or may not get lost in the sea of information, especially with the addition and subtraction of staff over a period of time. In the office, you could add QuickBooks, the various TPA portals, Xactimate, XactAnalysis, Cotality with Claims Connect & Estimate. Not sustainable.
So when I started looking seriously at AI agents in early 2026, it wasn’t because I had suddenly become obsessed with the future of work or read some headline claiming AI was about to replace everybody.
It was because I was drowning. Not dramatically. Not the kind of drowning where the business is falling apart and everyone around you can see it. The quieter kind.
The kind where you’re running hard, getting things done, growing the company, and somehow still feeling like the wheels might come off if you don’t get better systems underneath you.
In 2025, we did just over $800,000 in revenue. Good year. Real growth. Year three and doubled from the previous year. The kind of year you’re supposed to feel proud of for a team of five.
But if you had asked me on any random Tuesday exactly what sat in accounts receivable, how old invoices were, which TPA was dragging payment, or whether cash flow was about to get ugly six weeks out, I could not have told you with confidence.
I had instincts. A gut feeling. And if you run a restoration business, you already know gut feelings can carry you surprisingly far. Until they can’t.
We also had real financial pressure. A merchant cash advance was pulling a significant amount every single week draining our cash flow. That money leaves whether jobs are paid or not. Meanwhile, anyone working with TPAs knows the timing game. A TPA can sit on payment for 30 days after a job closes before processing even starts.
So the timeline looked something like this: You complete the work. You submit the invoice. You wait. And wait.
Meanwhile payroll still hits. Vendors still want payment. Equipment still needs maintenance. Any loan payments certainly don’t pause because collections are behind.
I knew collections were the lever. If I could stop operating on feeling and actually see the aging — invoice by invoice, TPA by TPA — I could intervene earlier. Make calls at 30 days instead of wondering at 90 why cash suddenly felt tight.
The problem was simple. There wasn’t a dedicated person handling all of it. There was me.
Doing 10 jobs at once. At the exact same time, operations had become its own bottleneck.
Here’s what a new loss looked like at our company: A new loss comes in from email or phone call. Then I verify claim details. I log the job in our chosen CRM. I build the Google Drive folder. I send out a mass text to our company with details. Call the operations manager to determine which team responds. Notify the field team and coordinate dispatch.
And somewhere in the middle of all of that, my phone rings because an adjuster wants to know if cabinets saved on a job a week ago were left onsite, another adjuster wants to know if a flooring sample was sent off to ITEL. Customer wants a call back wanting to know when we can be onsite.
Every. Single. Loss.
When you’re doing one or two jobs a week, manageable.
When multiple FNOLs start rolling in from AccuServe, Alacrity (now Altimeter), and direct carrier relationships in the same day, the administration starts becoming the job instead of supporting the job.
My claims admin was buried too. Hours disappeared into invoice comparison work — line-by-line reconciliation between estimates, payments, supplements, and TPA deductions. Necessary work, but exhausting work.
That was the moment I started paying attention to AI agents.
And to be clear, I’m not talking about chatbots. Not the FAQ thing on a website. Not “write me a caption for Instagram.” I mean actual agents.
Software that could connect to all of our technology that we implement to meet the growing requests of carriers and TPAs, follow rules, pull information, trigger actions, and ideally remove repetitive administrative work from already overloaded humans.
The vision sounded overwhelming, compelling and not in the least achievable. Until it was.
A finance agent connected to QuickBooks that could monitor aging reports, track open invoices, understand TPA payment cycles, and alert me to problems before they became emergencies.
An operations agent that could manage new losses, create folders, trigger workflows, and help remove the 30-to-45 minute admin sprint that happened every single time a claim came through.
And eventually, a growth agent focused on marketing and business development.
The names came naturally.
Atlas for finance, because finance carries everything.
Freddy for operations — yes, because something about an AI that never sleeps and shows up in your dreams felt fitting for the restoration industry.
Blaze for growth, felt appropriately on brand for an industry built around catastrophe.
The idea was simple: the agents own their lanes, while my business partner and operations manager and I stay responsible for the decisions. Humans decide. Systems surface the data. Easy enough in theory. Reality had other plans.
Here’s the part nobody talks about when they’re posting polished AI success stories online: The demos lie. Or maybe “lie” is too harsh. Let’s say they simplify.
Everything looks clean when someone is screen-sharing.
“Connect QuickBooks.”
“Set your rules.”
“Watch the magic happen.”
Sure. In a demo.
The first platform we tried broke almost immediately. QuickBooks authentication wouldn’t cooperate. Tokens expired. Connections failed. We burned days troubleshooting before finally scrapping the whole thing and rebuilding on another platform.
That part never makes the LinkedIn post. Nobody writes about spending two weeks muttering, “Why won’t this connect?” while also managing active claims.
The first version ran off my laptop. That should tell you something.
The first thing that broke was my memory. Not Freddy’s — mine. I’d written out a job channel SOP one afternoon. By the next morning, Freddy had no idea it existed. It was sitting in the workspace but not in his long-term memory file. That was an early lesson: an AI agent isn’t magic. It’s a system, and systems need to be built correctly or they fall apart in the most mundane ways possible.
The second thing that broke was the laptop. Which is why we moved Freddy onto a cloud server — at midnight, while also handling two brand new jobs and prepping for an adjuster call the next morning.
Not glamorous. Not cinematic. Just messy. But something interesting started happening in the middle of the mess. Atlas began working.
By early May, QuickBooks was connected. Google Drive was connected. Job updates from the field started feeding into a centralized database. Accounts receivable data became visible in a way it hadn’t been before.
And every Monday morning, I started receiving a finance brief in my email. Open invoices. Aging concerns. TPA delays. Things needing attention.
Was it perfect? Absolutely not. But it worked.
And Freddy started producing wins too.
New loss intakes that once took 30 to 45 minutes dropped to around 90 seconds. 90. Seconds.
Every morning in Slack the management team receives a morning brief which includes what each project’s status is, what is going on in the carrier, insurance carrier and TPA news around the internet.
Once I saw that, I couldn’t unsee what was possible.
This is probably the point where a lot of articles would end with some clean lesson about innovation and the future.
But that’s not what this series is. This isn’t a victory lap. At least not yet.
The point of part one is simpler than that: The problem is real.
If you run a restoration company and still manage AR manually, if cash flow feels more like intuition than visibility, if your admin process gets heavier every time volume increases, then you probably already understand exactly what I’m describing.
The question isn’t whether automation is coming.
The question is whether you’re willing to survive the ugly middle long enough to make it useful.
We decided we were.
Related Posts
Joyce Gabriel is the co-founder of Expert Restoration in Washington state. A Navy kid turned teen mom turned business owner, she spent years in the restoration industry before starting her own company around one principle: integrity in the middle of everything. She is currently building AI-powered systems to take the busy work out of restoration—and keep the human work in.