148 reports. 35 minutes. And no manual emailing.
Kimberley had 148 monthly financial and cost-centre reports to produce.
Each report contained five tabs, comparing actuals against budget, forecast and the previous year.
She used Claude to create all 148 reports in 35 minutes.
But what was super smart is what Kimberley did after. She connected the output to Power Automate, which handled emailing the reports to the right budget owners.
Previously, some of those people hadn't received reports because distributing so many files was, in her words, a "logistical nightmare".
As well as making the reporting faster, she also got information into the hands of people who needed it.
Removing hours of repetitive work from her FP&A team's month-end list.
Kimberley is now a finance engineer. Orchestrating her own AI workflows, so she can spend more time on the business.
My prediction for finance in 2027
Imagine a factory before electricity.
People did the work by hand. Then electricity arrived, but at first, production was still organized around the old way of working.
The real transformation happened when electricity became part of the production process. Machines and conveyor belts moved the work forward. And people now didn't have to keep carrying themselves everything.
AI is at this same turning point.
And this is one of my big predictions for finance in 2027.
The biggest productivity gains in finance next year will come from removing the human who moves information between systems.
A future where AI becomes your finance conveyor belt, in the exact same way Kimberley used it.
Think about a normal finance process.
You open your accounting system, download a file, prepare an Excel analysis, create a report and email the result. Then somebody takes that information and uses it for the next task. Maybe a journal entry. Maybe a forecast update.
When I think about my years working in finance, I would say, like 80% of the work followed this pattern. We spent a lot of time getting information from one place to another.
But something big is happening.
Until recently, if your finance system didn't have an API or an approved connector, somebody often had to operate it manually. Now AI tools can see what's on a screen, navigate websites, click buttons and complete tasks through the same interface you use.
And a new generation of AI assistants has its own computer in the cloud. They can keep working after you close your laptop, come back with a finished result and ask you for approval when they need it.
The technical boundaries between your finance tools will keep disappearing.
If you can operate a system through a screen, AI will be able to operate it too. We won't have to wait for every software vendor to build an integration before we can connect one part of our work to another.
That doesn't mean giving AI unlimited access to your company's money or data. The permissions and controls are still super important. But it changes what is possible.
And it changes how you should work.
Instead of being the person who operates every tool and moves every file, you need to become the person who orchestrates the process.
Let me show you how to start…
How to stop being the human connector: 4 steps
1. Find the work between the work
Imagine you prepare a weekly budget-versus-actual report. You download the actuals from your accounting system, open the budget workbook, calculate variances, write commentary, save the report and email the budget owners.
Some of that is important finance work. You need to understand the numbers, investigate unexpected movements and decide which explanations make sense.
But downloading, uploading, copying, pasting and attaching files have no value.
So, your first job is to find every place where somebody is acting as the connector.
What to do: Pick one recurring process and write down its full journey, from the source data to the person who uses the final result. Mark the steps where somebody moves information between tools, plus the steps that follow fixed rules and the steps that require human judgment.
AI tip: Ask AI to find the handoffs you could eliminate.
Give it your process description, including the applications you use. Then ask:
Act as a finance process automation expert. Here is my current workflow: [paste the steps and systems].Map the flow of information from the original source to the final recipient. For each step, identify whether it involves moving data, processing data, making a judgment or approving an action.Find the manual handoffs that could be removed using AI or automation. Rank them by frequency, time spent and difficulty to automate. Do not suggest removing financial controls or approval steps.
This is where you can find a lot of opportunities.
Maybe the biggest waste isn't building the analysis. Maybe it's collecting the files from five different systems every Monday. Or somebody downloading invoices from supplier portals. Or sending 148 separate reports to 148 budget owners ;).
Until you map the complete process, you may be automating the wrong part.
2. Let AI find the best way to connect your systems
Once you've found a handoff, the next question is how to remove it.
Until recently, the answer depended a lot on if your systems could talk with each other. If there was no API or connector, you needed somebody to download the information manually.
But now there are a lot of ways to get data from one system to another.
- Option 1: Connect directly. Use an approved connector, API or MCP connection to let AI access the information at its source. For example, you might connect Claude to your documents or an approved finance data source, giving it mainly read-only access, and only more access when it needs to transfer or change data.
- Option 2: Automate the file movement. If a system can generate a scheduled export, use Power Automate or a similar tool to place the file in a dedicated folder. Your AI workflow can then work with the latest data without somebody having to upload it.
- Option 3: Let AI operate the screen. This is the new option. Claude in Chrome, Claude built-in browser extension and ChatGPT's browser-control tools can navigate websites, click buttons and work through tasks using the interface a human would use.
Either way, imagine an old ERP with no useful API.
Every month, somebody logs in, selects the company and reporting period, opens the General Ledger report and downloads an Excel file.
An approved browser agent may be able to complete those same steps automatically.
AI tip: Don't choose the technology yourself. Get AI to compare the options.
This is the prompt I would use before building any integration:
I want to automate this finance process: [describe the workflow].These are the tools involved: [list the systems].Recommend the best way to connect them and move or process the data automatically.Compare three approaches:1. Direct connectors, APIs or MCP.2. Scheduled exports and workflow automation.3. AI browser or computer control for systems without suitable connectors.For each approach, explain the setup, permissions, reliability, likely maintenance and risks. Recommend the simplest reliable option that would work with our existing tools.Identify where AI should process information and where fixed calculations or standard automation would be more reliable. Include what should happen if a source is unavailable, a file is missing or authentication fails.Do not assume a connector or feature exists. Tell me what I need to verify with IT or the software vendor.
Now AI is helping you design the connection, rather than simply doing another analysis.
This is super important, because it means a missing connector is no longer automatically the end of the automation discussion.
In my AI Finance Club, a member called Jason has been exploring exactly this problem with vendor portals that don't offer APIs. He is getting AI that has access to it’s own computer to login and download invoices (more in the next step).
The only issue is MFA (Multi-Factor Authentication) which requires your input to keep your access secure. So make sure to be careful before you hand all of your browser based tasks over to an AI.
But think about what this means.
As browser agents improve, more systems that previously required manual interaction you will be able to automate with a bot!
3. Give the recurring work to an AI employee
So, let’s say you've connected the systems and tested the workflow successfully.
Do you really want somebody to open Claude or ChatGPT every month and tell it to begin?
Here's where the next big change is happening.
Until recently, even a well-designed AI workflow often needed you to open an application and start it. Or it depended on a computer you had to leave running.
Now some AI assistants have their own computers.
ChatGPT dots, Claude Duet and Grok Bot are examples of this new generation. They have cloud-based working environments that include their own browser and files. You can give them work, and they can continue while your own laptop is off. Grok Bot and ChatGPT Dots are available now (check your licensing for ChatGPT) and Claude Duet is on a waitlist you can apply for.
You can now have an AI assistant responsible for collecting the information your finance team needs every Monday. It checks whether the source data is ready, uses approved connectors or browser actions to retrieve it, places the files in the correct folders and tells you when something needs attention.
You don't need to sit there operating each tool.
AI tip: Give your AI employee responsibility for the handoff, not just a scheduled prompt.
For example:
Every Monday, check whether the latest approved sales, expense and accounting exports are available.If they are, collect them using the authorized connections or browser workflow and place them in the agreed input folders. Check that the files belong to the correct reporting period and that none are missing or duplicated.If a system is unavailable, authentication is required or the files don't match the expected format, stop and tell me exactly what needs attention.Keep a dated record of what you collected, where it came from and whether the run succeeded. Do not change any source data.
And if you don't want to write every instruction, just show AI how you do the job.
Claude has Record a Skill, ChatGPT Work has Record & Replay, and Copilot in Excel has its own skill recorder. Depending on the product, you can demonstrate a task and turn the recorded actions into a reusable workflow.
You still need to review what it learned, test it and give it the right permissions. But it's getting much easier to teach AI a process without having to write code, or write super long prompts.
The important part is defining what a successful run looks like and when the AI must stop and ask you for help.
Because an assistant that works while you're asleep is useful only if you can trust what it did when you wake up!
4. Connect the next action, but keep the decision
What happens next?
In many finance departments, somebody still opens the file, attaches it to an email and sends it to the right people. Or copies the results into another workbook. Or prepares a journal entry from them.
This is the final handoff I want you to challenge.
Remember Kimberley's example?
She used Claude for report creation and Power Automate for distribution. That combination allowed her team to give budget owners reports they previously couldn't distribute at that scale.
Your workflow doesn't have to use an AI tool for everything. Normal automation is sometimes better for predictable tasks, like sending an approved file to a known person.
AI is useful when the process requires reading information, interpreting it or adapting to different situations.
What to do: Define what should happen after the information has been processed. Decide whether the result should be reviewed, sent to an owner, placed in a shared folder or used to prepare another task.
AI tip: Get AI to design the next handoff, with an approval gate.
For example, imagine you have a spreadsheet containing cost centres, budget owners and approved report file names.
You could ask:
Design a workflow that takes approved files from this folder and sends each one to the correct recipient using the cost-centre owner table.Check that no file is missing, no recipient is duplicated and nobody receives another department's confidential information.Start with a dry run that sends the proposed distribution list to the reviewer. Send the files only after approval.Keep a log of successful and failed deliveries.
The same idea could work for other finance processes. Your overdue accounts workflow might prepare follow-up emails for review. Your month-end process might prepare draft journal entries with supporting evidence. Your forecasting process could place an updated file in the finance manager's review folder.
The technical tools may change, but the principle stays the same. AI moves the information, processes what it can and prepares the next action. You remain responsible for the financial decisions and controls.
The One Thing to Remember
I think the way we work in finance is going to change more in 2027 than a lot of people expect.
For years, we've been the people connecting systems that couldn't communicate with each other. We downloaded the information, moved it into Excel, prepared the analysis and passed it to somebody else.
Now AI can operate those systems through connectors, browsers and its own computers.
The technical boundaries are disappearing faster than the old finance processes built around them.
And I think that means we need to make a change in how we work.
We need to stop being operators and become orchestrators.
An operator opens the tools, downloads the files and moves the information.
An orchestrator defines the job, gives AI the right access, sets the checks and reviews what comes back.
Your judgment doesn't disappear. But your time is no longer spent carrying the work from one system to another.
And that, for me, is where the biggest productivity gains will come from.
Best,
Your AI Finance Expert,
– Nicolas
P.S. The best way to learn from me directly is to join my weekly masterclasses. 60 mins + Q&A where I'll get all your AI in Finance questions answered. Join me here.
P.P.S. Want a quick way to start orchestrating instead of operating? Here is a video on 4 Claude plugins that give Claude ready-made tools for finance, data, operations and legal work: 4 Claude Plugins Every CFO Must Have
