“Light has changed my perspective on using AI. That is something I now expect from a finance system.”

Consolidation alone took half a week, sometimes a full week. FX rates were entered by hand. AI use in the team amounted to asking ChatGPT the occasional question.

Before Light, Cédric’s AI use was the occasional ChatGPT question. Now he uses AI in his work every day through Light, and builds on it himself. Light scans every incoming invoice and pre-fills the data, so he reviews instead of typing. Receipts dropped into Slack are matched to their transactions. Agents run on a schedule in the background, and when he wants a new one, he describes the process in plain language to the Light Command Interface.

Cédric used to work out which receipts were missing, who they belonged to, and message each person on Slack. Then message them again. It was the task he liked least.

“It was a very repetitive task, so I postponed it until the end of the month, when it was almost too late and I really needed them.”

Now an agent sends a reminder once a receipt has been missing for 3 days. Employees drop the receipt into Slack, and the agent matches it to the transaction. Receipts arrive sooner, and Cédric no longer chases.

With no technical background each agent was built by describing the manual process to the Light Command Interface, which turned it into a brief the agent runs every month. To accompany the agents Oper also built a finance dashboard in Lovable, connected to Light through its APIs and Claude connector. It covers budget versus actuals, cash flow forecast, and management and investor reporting, and refreshes every morning.

Trust was built through testing. The agents were initially scheduled to run a minute later, with their outputs checked against the trial balance. For the first two or three runs, Cédric compared the automated results with the manual process, refining the prompts until the outputs matched consistently. He retains control over decisions that require judgement: for accruals, the agent checks with him before creating a journal entry.

With his audit background, Cédric knew how time-consuming it was to respond to auditors’ sample requests, often spending days retrieving supporting documents one by one. So when Light’s team invited customers to suggest app ideas, he proposed an audit assistant that would do the legwork: paste in a list of transaction IDs and receive a ZIP file containing all the supporting documents.

“I asked whether it could be built, and by the same evening it was already built. I tested it in the morning and it worked perfectly. Now I'm heading into audit season knowing I won't have to dig through all the documents.”

In Cédric's experience, a custom integration used to mean a paid request costing thousands of euros. Light's Personio app, which syncs joiners and leavers, took 5 minutes to set up.

Cédric’s next questions are commercial ones: where Oper makes money, where it can save, and what the numbers say while the month is still running. Audit season is next, and the documents are already a paste and a download away.