Skip the Blank Page. Transform Builds From Last Month's Close.

This is part of a series of posts on the new capabilities FloQast announced at TakeControl 2026. For the full picture — including journal entries reviewed by the AI Assistant before they post, Detect, Operational Audits, and the COSO AI governance module — check out the complete TakeControl 2026 announcement here.

‍

Most AI tools that automate accounting work start with the same ask: describe your process.

Describing it works, and Transform will build from a description. It just shouldn't be the only way in. Ask a controller to document an intercompany allocation that runs across seven steps, three entities, and a workbook the same person has maintained for four years (and who's currently on parental leave), and you've given them a new job before the automation helps with the one they already have. Same for the accounting manager whose close depends on conditional logic buried twelve tabs deep in a spreadsheet someone built in 2019. The process exists. It runs every month. It's just never been written down anywhere a machine can read.

That's not a motivation problem. It's the nature of close work. The real expertise in any accounting team isn't in their procedures manual — it's encoded in the workbooks they've built and refined, in the judgment calls that happen automatically because they've seen this before, in the output their team has reviewed and signed off on every close cycle. 

So, Transform takes either input. Describe the process, or show it last month's close.

What’s New in Transform

Transform can now build the agent for you, from the accounting work your team has already completed. It runs your team's workflows, and now it builds them too.

The short version: Transform reconstructs your processes as governed, deployable AI agents — complete with living SOPs, built-in controls, and audit trails that writes themselves. No IT involvement, no code, no starting from scratch. Your accounting team builds it, tests it against their own numbers, and signs off before it goes live.

How FloQast Transform Builds

Hand Transform the input files and the reviewed output from a close your team has already completed, and it works backward. It reverse-engineers the logic, the steps, and the checks, then turns them into a governed, deployable AI agent with a plain-language playbook and a built-in audit trail. The documentation your team never wrote writes itself, derived from the work they actually did.

It takes whatever you already have. Last month's source files and the finished output your team signed off on. An existing SOP. A conversation where you describe what you need, if that's easier. Any of those is a starting point, and none of them is a blank page.

There are two other ways in if you'd rather not start there: build from a prebuilt agent for the work every close has, or build it yourself, step by step, which some teams prefer. However you start, you land in the same place. One governed workflow, verified against your own close before it’s live.

‍

The Part Controllers Actually Care About

Here's where Transform is meaningfully different from what you'd get with a general-purpose AI tool.

Before any workflow goes live, it runs against your own prior-period data and verifies its output against work your team has already reviewed and signed off on. Not a simulation. Not a tolerance check. A column-by-column tie-out, entity by entity. If it doesn't match, it doesn't save.

While the agent is being constructed, your team reviews it in real time. Transform shows its work as it builds, every step, every check, every judgment call, so your team can stop it if something looks off. When it finds a decision buried in your data, it stops and asks: what's in scope, whether a standing exception is intentional, which tab it reads versus builds. Each answer automatically becomes part of the audit trail, without anyone writing it down separately. The review and the evidence happen at the same time.

Once it's live, the agent runs on its own between the review gates your team sets. Nothing deploys without an accountant's sign-off, and you decide where a human step belongs inside the workflow.

A Governed SOP That Writes Itself

Most accounting automation delivers the workflow and leaves the documentation to you.

Transform delivers both. Every agent ships with a living playbook: every input, step, and check, written in plain language from top to bottom, with version history, segregation of duties, and approval controls. You don't have to write it by hand. It's generated from the work itself, so it reads like the SOP your team should have had but never had time to create.

When the process changes — a new account, an updated allocation method, an entity added mid-year — the person who owns it describes the change in plain language. A new step slides in, the workflow re-sequences around it, and version history captures it. No ticket, no rebuild, no waiting for IT to get back to them. That’s the part that matters in month four, when the close has moved, and the automation has to move with it.

That's the maintenance answer general-purpose AI can't give.

What Teams Are Doing With It

Liquid AI's accounting team built eight custom agents themselves. Hazeldenes built five — each with a designated human owner and a separate reviewer — and automated a seven-step layered intercompany allocation process that previously required manual effort every close cycle. Their team absorbed natural headcount attrition without backfilling, maintained steady reporting quality, and saved approximately 600 hours per year by preserving human review for every agent. Across FloQast, 400+ agents are live at 125+ accounting teams.

The important detail in both cases isn't the headline number. It's who built the agents. Accountants. No IT queue, no code, no external project. The person who owns the process built the automation and kept ownership of it.

Where This Fits

Building agents from finished work is one of five capabilities FloQast announced at TakeControl 2026, and it pairs directly with the AI Assistant, which brings that same intelligence into FloQast's existing workflows — starting with journal entry review inside Journal Entry Management. Transform is how your team creates automation from work they've already done. The Assistant is how that intelligence shows up in the day-to-day work itself.

The through line across everything announced at TakeControl is that accounting AI has to clear a higher bar than AI in most other parts of the business. Every output needs to be explainable. Every decision needs to be reconstructible for an auditor. Transform’s approach to that bar is to start from work that already passed your review — your data, your numbers, your team's sign-off — and work backward from there.

The next posts in this series will go deeper into journal entry review inside the AI Assistant, how Detect's per-account learning model catches anomalies before month-end, and what FloQast's COSO AI governance module looks like in practice.

‍

‍