Event Recaps

CFO Connect Summit 2026: How to use AI in finance without losing control

At CFO Connect Summit 2026, 1000+ finance leaders spent three days examining how the finance function is changing as AI becomes part of everyday operations.

Speakers from companies including NetSuite, Spendesk, DeepL, and Amazon shared examples and learnings across finance technology, reporting workflows, agents, hiring, FP&A, international growth, tax, and treasury.

The sessions covered different problems, but speakers kept returning to a practical question: can finance trust the number, understand how it was produced, and use it to make a decision?

This article follows that thread, exploring what the Summit revealed about building a finance operating system for the AI era.

Catch up on all of the Summit sessions here.

Five practical takeaways from the Summit

Choose one recurring workflow before overhauling your entire stack.
Validate the data and the process before automating the output.
Give agents responsibility in stages, with clear points for human review.
Test how technology behaves when the data is incomplete, the model changes, or the workflow fails.
Use the time saved for deeper analysis and better business decisions.

How is AI changing the finance stack?

AI is making the finance stack more connected and placing greater emphasis on how systems, workflows, controls, and people fit together.

CFO Connect research found that finance teams are running on relatively lean stacks, with 90% now operating across six tools or fewer. As AI is layered into existing processes, the focus is shifting towards how systems work together and support the decisions the business needs to make.

Spendesk Founder Rodolphe Ardant described this shift as moving towards:

“The CFO stack as a system rather than a collection of tools. Several tools, but one coherent way of working.”

That puts a broader responsibility on CFOs. The job involves more than selecting software. Finance leaders need to understand which workflows technology should support, where controls should sit, and where people need to retain responsibility.

Where should finance teams start with AI?

While the temptation may be to overhaul existing workflows, speakers across the Summit recommended starting with one recurring process that is painful, measurable, and close to the team’s daily work.

The first step is deciding which process is most ripe for automation.

Which finance workflows are ready for agents?

Agents are most useful when work is structured, repetitive, and governed by clear rules.

The panel on finance workflows, agents, and automation discussed invoice matching, information extraction, transaction classification, and reconciliation as potential starting points. These processes involve high volumes and have relatively clear expected outcomes.

Work involving unusual exceptions, material decisions, or significant business judgment needs more care.

As Drey Kenfack, Founder and CEO of Crediflow AI, put it: “Behind every number, there is a story.”

An agent might recognise that a transaction is unusual without understanding why. The anomaly could reflect an error, a new market, or a deliberate business decision.

Before automating a workflow, ask:

Is the process clearly documented?
Are the inputs structured and consistent?
How often do exceptions occur?
What's the impact if the output is wrong?
Where must a person review or approve the result?
What would improvement look like after 30 or 90 days?

A staged approach allows agents to begin with tasks such as extracting information, classifying transactions, or preparing drafts. Teams can then review exceptions and handle decisions that require context.

What does this look like in practice?

Nicholas Boucher, founder of the AI Finance Club, used reporting to demonstrate how a repeatable AI workflow could work. His process involved four steps:

Connect the relevant source data.
Create the reporting workflow.
Check the output and its underlying calculations.
Turn the tested process into something reusable.

The workflow could produce dashboards, answer live questions, analyse cash and working capital, and run scenarios. Before building any of that, Nicholas insisted on a basic check:

“Do not do any analysis if your data is not good.”

For a finance team, that means agreeing definitions, tracing figures back to their source systems, and assigning clear ownership before asking AI to interpret the results.

The team must then review the output rather than accept it at face value. People should be able to trace the figures, check that the calculations reconcile, and understand how the dashboard or scenario was produced.

“You cannot blame AI for a mistake. It is like blaming Excel because you wrote the wrong formula,” Nicholas Boucher, The AI Finance Club.

How should CFOs evaluate AI vendors?

A software demonstration shows what a workflow can do under the right conditions. The Summit’s governance discussions encouraged CFOs to test the failure path too.

Vendor reviews should cover data storage, subprocessors, model training, change management, audit logs, incident response, and known limitations. Finance leaders should also ask whether material model changes will be communicated and tested before they affect live workflows.

Useful questions include:

What does the tool do when it does not know the answer?
How does it handle incomplete or contradictory data?
Can the team test a new model before it reaches production?
Which actions require human approval?
What records are retained for audit?
What happens if the vendor experiences an outage?
Which use cases does the vendor advise customers not to automate?

The aim is to evaluate each vendor with the same care applied to any other system that affects the numbers.

The same discipline applies beyond AI tools. Finance teams also need reliable controls when transaction data feeds compliance processes or when cash visibility informs critical decisions.

Why are e-invoicing and treasury now finance operating issues?

The tax and e-invoicing session showed how compliance is moving closer to the transaction itself.

An invoice is becoming structured data that can affect payment, shipment, revenue recognition, reporting, and audit readiness. Implementation therefore requires accurate master data, clear ownership, process mapping, testing, and a way to manage exceptions.

A tax engine cannot compensate for missing or inaccurate legal entity details, VAT numbers, addresses, or tax codes. Finance, procurement, IT, accounts payable, and commercial teams all influence the quality of the transaction.

The treasury discussion raised a similar issue around visibility. Forecast error was identified as a major source of risk because finance teams may not know where cash will be in 90 days, even when they have access to plenty of financial data.

A 13-week rolling cash forecast, daily cash visibility, segregation of duties, and stress testing can give decision-makers enough warning to respond while they still have options.

How can finance teams move from experimentation to implementation?

Day 3 focused on putting ideas into practice.

Veronica von Heissen Rotenberg, CFO and Managing Director at Everphone, delivered a talk on moving from strategy to execution. In the session, she described an approach based on volunteer pioneers, cross-functional support, hackathons, and focused internal tools.

“Start with one concrete problem, solve it well, and build from there.”

Here’s how to do it over a 90-day period.

A practical 90-day plan for finance leaders

Days 1 to 15: Choose the workflow

Select one recurring process or decision that causes friction.

Define:

The current workflow
The time it takes
The people involved
The data required
The risks if something goes wrong
The measure that will show improvement

Days 16 to 45: Build and verify

Simplify the process and connect the relevant data. Decide where AI can assist and where a person must review or approve the result.

Create a verification checklist covering:

Data quality
Calculation accuracy
Exceptions
Permissions
Human review
Audit evidence

Days 46 to 90: Standardise and learn

Document the workflow, train backup owners, measure the result, and decide whether the use case should be expanded.

The outcome may be a successful pilot, or evidence that the process needs more preparation before it can be automated. Either way, the team has learned what to do next.

A successful pilot will give finance a workflow it can trust, with clearer visibility into the numbers, and a path to faster decision-making.

Get more from CFO Connect

Finance leaders rarely have the time to test every new tool or redesign every process themselves. CFO Connect helps them learn from peers who are already making decisions about systems, controls, teams, and growth.

Through events, research, and the CFO Connect network, finance leaders can compare approaches, share what is working, and stay close to the changes shaping the function.

Apply to join CFO Connect