Experts weigh in: what it actually takes to implement AI in a finance team

AI is transforming finance; that is no longer up for debate. The much harder issue to resolve, and the one that most content on the topic skips over, is how to actually make it work inside a real finance team, with a functioning tech stack, timely constraints, and a team of people who may or may not be on board with this change.

At Forge Connect Oslo in May 2026, nearly 30 finance leaders came together to spend the day working through exactly that issue with three speakers who brought three very different angles to work through it. Niels de Kind tackled the process problem. Nicolas Boucher tackled the confidence problem. Sebastien Marchon tackled the governance problem. None of them referenced the same framework twice. All three arrived at the same underlying diagnosis: when AI fails in finance, it isn't the technology that broke. It's everything around it.

Bringing together those expert perspectives, here we aim to leave you with actionable steps you could implement in your team from tomorrow. As did the Forge Connect event.

AI fails on the process nobody mapped, not on the technology

Niels de Kind, digital transformation expert at BMW Group and founder of Novendo Consulting, opened his session in Oslo with a show of hands: how many people had started an AI project in the last 12 months. Nearly every hand went up. He then asked: how many had one that was live and adding measurable value. Over half the hands went down.

That gap, he argued, comes from a single mistake: building AI on top of a process nobody has actually documented.

He took the room through a case from his own work: a dealer contract-modification request that looks simple on paper but actually runs through three systems, eight manual checks, and a five-minute phone call to a colleague in legal who's seen the edge case before. It happens 40 times a day. Hand that to a vendor who's only ever seen the data, and six months later the project collapses. Nobody ever asked what that phone call to legal actually contained.

The most valuable AI foundation you can build is not only a data set, but a well-documented, human-validated process map.

– Niels de Kind

Niels de Kind’s fix is a three-step sequence that can't be reordered:

1. Map how the process actually runs today.
2. Design it from scratch with AI in the room.
3. Define use cases against what he calls the IMPACT filter: impact, measurability, process clarity, data availability, complexity, and technical fit.

Fail on more than one of those filters, and the use case isn't ready, no matter how good the vendor demo looked.

Hold on the company-wide roll-out: team adoption starts with one person's fluency

Nicolas Boucher, the AI-in-finance expert educator who has now trained over 10,000 finance professionals, opened with a story involving his daughter. She wanted to keep playing a paid iPad game; he said no. His wife suggested he build the free version himself. In a day, with his eight-year-old designing the features, he did exactly that.

If my daughter can build an app in a day, why do we need six-month transformation plans?

– Nicolas Boucher

His answer for finance teams is what he calls the 30-day playbook, and it starts with the leader building their own fluency first. Week one is all about picking one tool matched to what the company already runs on (Copilot inside a Microsoft stack, Gemini inside Google Workspace, Claude or ChatGPT if neither applies) and building one undeniable use case, a "banger” if you will, that proves AI can do helpful and efficient finance work, not just draft emails.

Boucher has watched the tool landscape shift fast inside his own community: Claude usage went from near zero at the end of last year to 37%, while ChatGPT's share in the same group fell from 80% to 30%. Reason enough, he says, to stop assuming last year's default tool is still the right one. Weeks two and three move from proving it works to making it repeatable: pushing AI into the tools the team already opens every day, then turning any workflow worth repeating into a documented skill instead of a one-off chat. Only once that muscle exists does it make sense to talk about a team-wide programme.

Without governance, adoption goes underground.

Sebastien Marchon, Rydoo's CEO, picked up the governance question with his own company as the case study. Rydoo had used AI inside its product for years, he said, but until early 2026 it had barely touched finance, legal, or HR. A survey to gauge appetite came back with a result that surprised him: the team wasn't worried about AI, it was excited, and what it actually wanted was a framework to implement and use it effectively. This included what tools were allowed, and what data could go through them.

Two weeks after that survey, Rydoo had an AI policy, a chosen toolset (namely Claude and Notion AI), and an AI guild of 12 people drawn from across the business who test tools and spread what works (a guild whose members rotate on a monthly basis to stay fresh). On top of that sits a maturity system modelled on martial arts belts, from white to black, scored across six dimensions including judgment, workflow design, and business impact. Of 208 employees who completed the assessment, 15 sat at white and 76 at green, with finance running slightly behind product on the same scale, a gap Marchon described as unsurprising, but not alarming.

AI is now part of the normal skills. There is no success if there is no measurement.

– Sebastien Marchon

The point of the belt system isn't to embarrass anyone at white. It's to make sure nobody is still at white, unmeasured, 12 months from now. That is one way to ensure an AI-native organisation is built.

What the room agreed on: where everyone could start tomorrow

The conversations that spilled into the afternoon's roundtables echoed all three sessions at once. Finance leaders admitted to approving a dozen AI tools and being able to count active users on one hand. Several flagged that the real constraint isn't appetite, it's time. Teams already running lean don't feel they have room to experiment, let alone map a process properly. More than one person raised the workforce question directly: what happens to the people whose jobs are the process being mapped.

These are all valid concerns to be addressed. However, none of these conversations undo the case Niels, Nicolas, and Sebastien each made. To get started with AI implementation in finance teams, you have to do three things:

1. Map the process before touching a tool,
2. Build personal fluency before asking a team to trust yours,
3. Then, put governance in place before unsanctioned use becomes the default.

The technology, as every speaker at Oslo pointed out in a different way, was never really the hard part. AI is an optimiser and it can do a hell of a lot to increase efficiency and productivity in teams. What it can’t do is be the human thinker that delivers on this implementation plan before rolling out team-wide. That’s what needs you and your mind.