There’s a category of data that breaks Excel. Literally — you try to open the file, the machine freezes, and after 15 minutes you give up and close the tab. Excel has a hard row limit just above a million, and some of our transaction reports blow past it without blinking.
For a long time, those files just sat there. We knew they existed. We knew there was useful information inside. We couldn’t get to it.
And that was just the technical barrier. Some data we could open but couldn’t properly interpret — terminology from corners of the payments world that don’t appear in any accounting curriculum. Some data we could access and understand, but only in fragments, because assembling the full picture manually took more time than it was worth.
That’s what AI actually changed for us. The access. Work that was technically or conceptually out of reach is now in reach. What we’ve done with that access has been, quietly, one of the more significant shifts in how we operate.
Data that was not interpretable and technically unmanageable is now clear, visible and controllable.
The dashboard that replaced a spreadsheet
Guy is on our finance team. A few months ago he built a dashboard — using Claude as the engine — that pulls GMV and net revenue per publisher, month-on-month, with AOV and take rate calculated automatically. Before it existed, this analysis lived in a fragile Excel file that someone had to update manually. Updates were slow. The picture was always slightly out of date. And the analytical depth just wasn’t there.
Now it updates with minimal effort. The finance team uses it regularly. Our Business Development team is getting access too.
What it changed in practice: period-over-period comparisons that used to take hours happen in seconds. Spikes and drops in the data surface immediately rather than turning up in a monthly review. And recently, we added geographic filtering. Within days of adding it, we could see our geographic spread clearly — where we’re gaining ground, where we’re not, whether the things we’re doing are actually working. That led directly to decisions we couldn’t have made with the old setup.
The quieter changes
These two examples are the most visible. But AI changed a few smaller things too, and they add up.
I don’t know SQL — the database query language that lets you pull structured data from large datasets. Before, if I needed a specific query written, I waited for someone who did. Now I describe what I need in plain language and Claude writes the query. The dependency is gone.
We also connected Claude into NetSuite, our accounting system, to handle vendor invoice processing. The accountant used to enter each invoice manually — vendor name, amount, service category, cost centre. At the volume we operate, it consumed real time. Now Claude prepares the entries and a human reviews every single one before it’s committed.
That last part matters. NetSuite is the financial record the whole company relies on — investor reports, board updates, everything. We built the approval requirement into NetSuite itself, not into Claude, so it can’t be bypassed. Every action requires sign-off. Claude prepares. A human decides.
Every action requires sign-off. Claude prepares. A human decides
What I’d tell another finance controller
If I’m honest about what made the difference, it comes down to a few things:
1. Start with the data you can’t open. The biggest shift for us didn’t come from making existing processes faster. It came from getting access to data that was technically or conceptually out of reach. If there are files your team can’t open, or reports you receive but don’t fully understand, that’s where to start.
2. Prompt for expertise. Asking Claude to analyse a file gives you a summary. Asking it to analyse a file as a domain expert with 20 years of experience gives you a different class of response. The role you assign it matters.
3. Build the safeguard before you build the workflow. If you’re connecting AI to a system others rely on — payroll, accounting, investor reporting — the human approval layer needs to be structural. We built ours into NetSuite, not into Claude. That’s the right order.
4. Bring your specialists in early. Claude can surface analysis. The people who know the domain are still the ones who turn it into action. Get them working from the same picture as fast as you can.
5. Give it something genuinely hard. The tools are better than most people expect. Find the file that breaks Excel. Throw it in. See what comes back.
--
Elad Geffen is a Financial Controller at Appcharge. Charged is Appcharge’s editorial publication, covering the people and craft inside the company and beyond.
