The Payroll Company That Let a Robot Into Its Books, One Approval a Day

Most people will link their bank account to an app on their phone without a second thought. Almost none of them will let an AI decide what gets posted to the ledger. A finance team in the entertainment payroll business found a version of automation it was willing to trust with QuickBooks, and it started with the most repetitive job in the company.

Every payroll run in film and television produces paperwork. A production hires background talent for a week, the payroll company pays them, and out comes a batch invoice: gross wages, payroll taxes, fringes, workers’ compensation, casting fees, payroll fees, postage. Multiply that by every production on the slate and every week-ending date, and someone in finance is spending their week pulling invoices out of the production platform, saving them to a shared folder, and re-keying seven line items per invoice into QuickBooks, one at a time, then reconciling the cash that comes in against them.

At Everyset, a payroll and casting platform for productions, that job now happens on its own, every day, without a single human approval. Nicole Jordan-Dahdal is Chief Financial Officer of Everyset through Prism Edge, her fractional CFO firm, where bringing AI and automation into client finance functions is a core part of the practice. She runs the function in partnership with Fiscal Founders, the firm that handles Everyset’s day-to-day accounting.

“When I came into Everyset, the finance team was re-keying seven line items per invoice into QuickBooks by hand for over a thousand invoices a month, on every payroll run, for every production on the slate. There was no clean identifier between our payroll processor and our platform, so matching the cash against those invoices was a manual hunt every month,” said Nicole Jordan-Dahdal, Founder of Prism Edge and Chief Financial Officer of Everyset. “Like most startups, the product roadmap was full and finance was behind it. Meanwhile my team was drowning. So, I stopped waiting for the roadmap and went and solved it myself.”

The finance team uses loopfour.ai, a workflow automation platform built for finance, to watch the folder where payroll invoices land, read each one, break it out into its seven line items, route it to the right set of books depending on whether the production is union or non-union, and create the matching customers, invoices, and journal entries in QuickBooks Online. 

Incoming payments are matched against them for reconciliation. The team reviews the exceptions. Everything else runs as code. The scale is not small. On active days the workflow runs a few hundred times, and on one day in late July it ran more than nine hundred, each run a discrete, logged step in a pipeline that fans out across projects in parallel and comes back together for one approval.

Before Nicole talked to loopfour, she had already written the spec for what Everyset needed, and the spec read like a software project. A webhook-driven sync engine on AWS Lambda, a message queue, a state database, retry logic with exponential backoff, a monitoring stack, and four phases of delivery. Seven categories of payroll line items mapped into the chart of accounts, tax rates validated against statutory rates, union fringes checked, payment status sync, daily reconciliation reports. The estimate on the document was six to eight weeks with a dedicated engineering team.

“That is the thing about finance workflows. Every one of them looks bespoke, because every business has its own line items and its own rules,” said Vinay Datta Pinnaka, cofounder and CTO, of loopfour.ai. “The question is whether you build a custom integration for each one, hire someone to do it by hand, or get the rules into a workflow that runs the same way every time. We think the third option is the only one that scales.”

Instead of a middleware project, Everyset got a workflow. Existing invoice data became the starting point, with the necessary business rules built into the process once. From there, the workflow can automatically handle new invoices and production batches, move the relevant information into the accounting system, and generate a reconciliation the accounting team can review against the bank. No engineers on Everyset’s side, and no changes to the production platform.

Here is the tension at the center of this story. Everyone is being told to adopt AI, and finance teams are being told it most loudly. And yet, in conversation after conversation, the same thing comes up: people are happy to let a chatbot draft an email, and deeply uncomfortable letting one anywhere near the books.

“I just got off a call where I asked someone whether they had connected their bank accounts yet, and the answer was no,” Pinnaka said. “That is not irrational. We link our bank to an app on our phone all the time, but the app is not making decisions. The moment you authorize something to act, the mindset changes. You want to know exactly what it is going to do.”

That is the distinction loopfour.ai was built around. A general purpose large language model is probabilistic. Ask the same thing twice and you may get two different answers. That is fine for brainstorming and dangerous for a ledger. If you authorized a chatbot to post journal entries, you would be trusting it to reason correctly every single time, with no way to check its work.

Loopfour takes a different route. Once a workflow is set up, it runs as predefined steps, identical on the first run and the millionth. AI is used where it is useful, for reading a document or classifying a record, but it does not get to decide what the accounting system does. Every action, every approval, every write to QuickBooks is logged, so the team can open a run and see exactly what happened rather than trusting a black box. Each invoice carries an internal ID, so the system knows what it has already posted and does not post it twice.

“I am not handing a chatbot the keys to the general ledger,” Nicole said. “What sold me was that this behaves like software. Once the rules are encoded, run one and run one million are identical, and every write is logged. I can open any run and see exactly what happened. That is the line between a tool a finance lead can sign off on and one they cannot.

None of this means the first month is magic. Getting a real finance process into code means surfacing every rule the team has been applying by hand, including the ones nobody wrote down. At Everyset, that meant working through how workers’ comp should be split between margin and residual, how casting fees should appear on the profit and loss statement for each entity, and how to handle invoices that had been entered manually before the automation went live. Each of those became a rule in the workflow, and each fix applies to every run after it.

“The pipeline picks the invoices up, splits out the line items, routes union and non-union to the right entity, writes to QuickBooks, and hands back a reconciliation. My accounting team only works the exceptions. We are still counting the wins from this launch, but the first one has been traceability. Every figure ties back to the invoice that produced it, on every run. When you are hand-keying seven line items across a thousand invoices a month, that is the one thing you cannot promise. It has also freed our time for higher-level problems, and we are automating more of them. This was a test project for Everyset, and as a result we are implementing AI across the company,” Nicole said. 

“It works because it is not just me,” she said. “Fiscal Founders runs the day-to-day accounting for Everyset, and they were in the build from the start. The rules this workflow encodes are their rules, the ones they had been applying by hand every month. You do not get a finance process into code by handing it to a vendor. You do it with the people who close the books, understand GAAP, and know what does and does not require human intervention.”

The AI conversation in finance has been stuck on a false choice: hand everything to a model and hope, or keep doing it by hand. Everyset is an early example of a third option. Use automation that behaves like software, deterministic and auditable, and use AI only inside those guardrails. It is less exciting than a chatbot that promises to run your books. It is also the version a finance lead can actually sign off on. 

“Finance teams are being told to adopt AI louder than anyone, and they have the most to lose if it goes wrong,” Nicole said. “The answer is not to hand it all to a model or keep doing it by hand. It is deterministic automation with AI inside the guardrails. That is the version I will put in front of a board.”

For more information, and to be kept up to date on new developments, visit: loopfour.ai, prismedge.co and everyset.com.