Over the past few weeks, weâve followed finance leaders into matcha bars, multi-entity dashboards, fractional CFO practices, and even disaster recovery operations. Weâve seen AI untangle reconciliation headaches, power real-time reporting, generate technical accounting models, and help lean teams deliver enterprise-level insights.
But this series has never been about the âcool techâ alone. Itâs been about what happens when finance professionals take the tools in front of them, pair them with their own expertise, and solve a real problem, fast. Now, as we wrap up AI in Action, itâs worth stepping back to look at the common threads, the surprising lessons, and the very real impact this technology is already having in the finance function.
Speed is nothing without skill
From Luke Simpsonâs AI-assisted Python scripts at JENKI to Martin Goodwinâs SQL-powered reporting rebuild at Open Box Software, one message has been loud and clear: AI doesnât replace finance knowledge, it amplifies it.
Lukeâs reconciliation automation for Selfridges is a perfect example. OCR, Python, and Cursor did the heavy lifting, but it was his understanding of JENKIâs data, cost structure, and revenue streams that made the automation accurate and reliable. âAI helps me work faster, but I still do the thinking,â he reminded us.
Martin, teaching himself SQL with ChatGPT as his tutor, hit the same truth from another angle: âIt didnât give me the right answer a lot of the time, but it gave me the start⌠and when itâs wrong, you still have to know, okay, why didnât it work?â
In every case, the finance professional stayed in the driverâs seat. The AI might clear the road ahead, but you still have to know where youâre going and when to hit the brakes.
Remove barriers, donât replace people
Phil Walkerâs work at Mettryx showed how AI can bridge capability gaps without breaking SME budgets. His use of ChatGPT to generate DAX and SQL for Power BI meant he could deliver multi-dimensional profit analysis, by channel, geography, and customer group, without the cost or complexity of a full enterprise data stack.
But Phil also brought a healthy dose of skepticism. In areas where AI risked undermining client trust, like writing commentary or producing final variance analysis, he held back. âYouâve got to have the domain knowledge to challenge what youâve seen and whether itâs correct. Itâs almost like coaching a part-qualified member of the team.â
Itâs a principle that ran through every interview: AI isnât here to replace finance teams, but to remove the friction and bottlenecks that slow them down.
Better inputs = better outcomes
The oldest adage in computing: garbage in, garbage out, still applies. Martinâs story drove this home: his reporting transformation only worked because he invested in cleaning and structuring the data first.
Mark Welton at Disaster Relief Australia took the same disciplined approach, whether training his custom Internal Control GPT or building AI models for IFRS 16 lease accounting. Every success started with a clear definition of the problem, clean source data, and explicit instructions for the AI.
Treating AI like a junior team member became a recurring metaphor. You wouldnât give a new hire a vague task and hope for the best, youâd give them context, guidance, and review their work before scaling it. The same applies here.
Small starts can scale big
One of the most encouraging themes across the series? No one waited for the perfect budget, the perfect tool, or the perfect conditions.
- Luke was already comfortable with Python and simply layered in AI tools like Cursor to go further.
- Martin had never touched SQL or Power BI before, he just decided to learn.
- Phil used AI as a helpdesk before moving into deeper technical applications.
- Mark started with GPTs for policy queries and grew into advanced modelling and accounting standards automation.
These werenât moonshot projects. They were practical, scrappy, and built to solve one problem at a time.
Adoption is about people, not tech
Markâs grassroots approach to AI training at DRA is a playbook in itself. Instead of dropping tools in peopleâs laps, he ran small, hands-on workshops where participants built something they could actually use.
The results? Even skeptics became advocates. From snakebite medical GPTs to mobile insurance assistants, people saw immediate, practical value, because they built it themselves.
Itâs a reminder that AI adoption isnât just a technology rollout, itâs a cultural shift. Confidence grows through doing, not just hearing about possibilities.
The ROI is real and measurable
This wasnât theory. Across the series, the gains were tangible:
- Weeks to minutes: Markâs IFRS 16 models went from multi-week builds to instant outputs.
- Cost avoidance: AI replaced a $130,000 consultant-led policy project with a two-week in-house effort.
- Faster insight delivery: Philâs AI-assisted analytics gave SME clients data quality and depth they couldnât otherwise afford.
- Error reduction: Lukeâs reconciliation automation removed hours of manual matching while improving accuracy.
In every case, the time saved didnât just vanish, it was reinvested into higher-value work, strategic projects, or direct client service.
The bigger picture
Looking back, perhaps the biggest surprise wasnât the power of the tools, but the creativity of the people using them. No two implementations were the same, yet all shared a common DNA:
- Start with a real problem.
- Use the tools you already have or can learn quickly.
- Keep humans in control.
- Build, test, refine, repeat.
The professionals we spoke to arenât waiting for the perfect AI solution to arrive in a box. Theyâre making the tools fit their needs today, learning as they go, and building the foundations for even bigger wins tomorrow.
And that might be the most important lesson of all: you donât need to be an AI expert to start, you just need to start.
If you missed any of the articles in the AI in Action series, you can find them all on the CFO Techstack website.
.png?upscale=true&width=1196&upscale=true&name=Untitled%20design%20(42).png)
.png?upscale=true&width=1036&upscale=true&name=overview%20(4).png)
