How an Insurance accounting team got out of the “doing”

A three-person accounting team at an insurance company built over 50 agentic operations, with no engineering involvement. By letting AI handle the repetitive data entry and coordination, the team saved 30 hours a month and transformed their department into a strategic command center.

Starting with “Fake AI”¶

When a new Controller joined an insurance company a year and a half ago, he just spent 9 years at an insurtech startup that didn’t yet allow their accounting team to use AI. While the company encouraged experimentation with AI, early attempts to bring it to the accounting process fell flat.

At first, the team tried using standard chatbots. “ChatGPT answered questions. It didn't do accounting,” the Controller explains. “Give it an Excel file and the output was unusable. Yes, it's AI, but it's what I consider fake AI. You're not revolutionizing accounting processes. You're bypassing some research.”

Upgrading to Cowork and a Claude plugin for Excel made the work feel a bit more real. The Excel output was better, though still hard to checkpoint and iterate with. Having Claude control his Chrome browser allowed AI to draft journal entries in Netsuite, but it took 10 minutes to do a single one rather than a bulk CSV upload doing many entries in one minute.

The Controller described this as moving from "fake AI" to "hand-holding with AI". It felt a little like magic because AI was actually doing the work, but every task was still initiated by hand and took far longer than the manual route. As he put it, they still weren’t operating at the “AI agent level” that he saw his company’s engineers using.

The Turning Point: draft entries without going into NetSuite¶

The breakthrough happened when they stopped trying to make an AI click through a web browser and instead connected Malleable directly to NetSuite.

The Controller didn’t just want AI to automate clicking buttons, he wanted to shift his team’s focus from repetitive work in spreadsheets to leveraging their domain expertise to add strategic value to the business. As a result, the Controller needed a way to teach AI exactly what he wanted it to do, and how his team should be involved in the agentic operation.

This is when the Controller met Malleable’s team. He was pessimistic. He remembered thinking “every tool promises to revolutionize how we work, but it only helps the people who live in GitHub or Snowflake. NetSuite never fits. But I’ll give them a few hours to see if it’s any different.”

He started with NetSuite incrementally. First, a very simple form that his team could visit to enter a basic journal entry. An agent then drafted it in the background, the form asked for his approval, and then it finalized it. This proved the capability was there and the Controller gained confidence in how he could oversee agents.

“The first run was just two journal entries. It didn’t save time, but it proved the hypothesis,” the Controller says. Once they knew it worked, they scaled it up.

“That was the turning point, when we got an agentic operation with NetSuite,” the Controller says. “It took a workbook, ran calculations, and through the right due diligence and approval, got it into NetSuite. My team was then hooked.”

By August, an agentic operation was routinely preparing and posting entries as large as 956 lines.

Building Momentum: 15 Minutes to Automate 30 Seconds¶

When the Controller had the first form that could post to NetSuite, he brought it to his team’s weekly happy hour, a once a week meeting where the accounting team meets not to talk about daily operations but about using AI and it's a place to collaborate and share thoughts and lessons learned with each other. He showed them how he could send an email and see it in NetSuite 2 minutes later. He described it to them as “you tell AI what you want it to do, it draws boxes and lines to show you what it will do, and then it does it.” It felt approachable, rather than abstract, and soon other accountants were building their own workflows too.

Small wins, then big ones¶

The strategy started simple: try spending 15 minutes building an operation that saves 30 seconds a day, then keep going. It was like a muscle they were building together. Some example wins:

• The team used to spend Sunday mornings downloading AP aging reports and manually working through overdue bills. Now, a daily operation runs at 10am to triage the ten oldest bills from Ramp to the right people on Slack.
• The team built watchdogs to automatically flag duplicate payments and prepaid expenses.
• Agentic operations run throughout the day, checking Ramp and NetSuite 50 times a day so the accounting team can focus on other things.

Once they had small things running on autopilot, the bigger targets became obvious. In August, the team set up an operation to automate claim checks from a daily bank feed and recorded almost 400 checks their first week. Next, they updated almost 800 vendor-subsidiary records in NetSuite without any manual data entry.

Building trust in the system¶

Accounting answers to auditors and regulators. The Controller’s biggest fear was a "rogue run", AI thinking it books a single $1,000 entry and accidentally getting 10,000 of them.

Malleable’s handoff system prevents this by keeping humans firmly in the loop. The AI prepares the work, but it is unable to interact with NetSuite until a human accountant reviews it in a secure form. In August alone, the AI paused for a human decision 400 times. Furthermore, while some tools scope access based on users or workflows, granular permission levels are locked based on the stage and progression of a workflow. This means one part of a single workflow may be able to find overdue bills in Ramp, but afterwards that same workflow cannot use Ramp in any other way.

The Command Center¶

The company is steadily moving toward what the Controller describes as an accounting command center. His mandate to the team is simple: get out of the doing, and do the thinking. Instead of manual data entry, AI agents now pull data autonomously, run the initial close steps, and stop at their checkpoints. The accounting team walks in, reviews the work, and spends the rest of the day on actual insights to improve the business.

"We save 30 hours a month, which by itself is dramatic for a small team," the Controller says. "What's more important is that we now have the space to use our actual experience and give leadership guidance on what needs to happen. This lets us finally spend our time on the strategy we were actually hired to do."

"I'm not technical. I don't think of myself as a software builder; I'm an accountant," he adds. "But this is the first time in my career that I can take a vision for how I wish my job worked, and how I wish my team could collaborate, and we can make it happen."