Tackling your board's next big question

Last week, in the Boardroom Brief, we weighed the arguments for - and against - building your own finance tools versus buying them. This week, we hear directly from CFOs who’ve actually rolled up their sleeves and done it themselves. What risks and best practices came out of their AI coding adventures? 

Your FP&A function has blind spots. This test finds them. 

Tight reporting. Clean models. Fast close. You're probably strong in some areas, but which ones are quietly holding you back? 

Aleph's 5-minute FP&A Fitness Test shows you exactly where you stand relative to peers at your size and stage—and what to fix first. 

Get your FP&A Fitness Score and training plan.

“You can put a cat in the oven, but that doesn’t make it a biscuit,” according to a line from White Men Can’t Jump.

Similarly, you can put a finance pro in front of a Claude coding window but that doesn’t mean they’re a software engineer. Or that they understand the risks.

AI-powered coding applications make it possible for anyone to build their own software- or at least attempt to. Naturally, this has turned heads in finance teams long frustrated with tech tools that fall short. 

But CFOs say vibe coding your own software that passes the “but does it always work how you need it to” test in a business is full of bear traps. And unless you understand code, a self-built app that appears polished on the surface could be a Frankenstein’s monster underneath. It could be carrying bugs; compliance or security issues; or incorrect calculations you just can’t see.

Research into AI coding in the Stack Overflow Developer Survey from July 2025 - released just as vibe coding was becoming a thing - shows AI tool adoption is surging as companies chase productivity gains. But sentiment towards the tech has dropped. More developers distrust its accuracy (46%) than trust it (33%), and 66% are frustrated by the need to remove bugs.

Vibe coding experiments can create tools and models that look amazing but soon fall apart when they’re let loose in the wild. A recent Baytech Consulting report provided a sobering view of the possible risks of AI coding such as its potential to cause maintenance issues; security vulnerabilities; and the illusion of competence among staff who don’t fully understand what they’ve built, leading to fragile, stitched together code bases. 

To keep on top of these risks, some experienced CFOs are taking a disciplined approach to long-term costing of AI workflows with rigorous testing, and a zoned governance model dictating where and how AI can be used in a project.

Building an DIY model can be a good entry point for anyone considering homegrown enterprise tools. But these come with warning tags too. Andy Levinson, CFO at hospital solutions provider Bluesight, said he generally believes in buying tried and tested tools, and then building in-house only where the goal is unique to a company, and  so needs customization.

But, tempted by a YouTube video claiming that AI-coding could build three-statement models in hours, he found his test builds riddled with significant errors in the results.

“They were way too general and didn’t work. For example, the balance sheet didn’t balance,” he said.

Back of the class, YouTube gurus.

However, with practice, Levinson has learnt you can teach AI to construct a working operating model if you take a less rushed approach. 

“If you take the time to break it into parts – such as accounting processes, transactions and labor costs – you can build reusable modules that you can then tie together very quickly,” said Levinson. “So yes, AI is still helping you build a model faster than you could before.”

As to who conducts the quality assurance on a DIY tool or model, that depends on the task. If you’re just using a model in Excel, IT shouldn’t need to be involved with that. But if you’re migrating it into your EPM, it would, said Levinson. 

Another evolving use of AI coding is to generate tools that quickly connect data in disparate systems. But, spoiler alert,  it’s not going to replace your ERP any time soon. 

“I see vibe coding filling the gaps between systems, for example, tying together information from Salesforce and SAP; or Stripe and the general ledger. That accelerates the time to insight, turning more existing information into analytics – it’s a dream for finance people,” said Glenn Hopper, managing director, head of AI at VAI Consulting.

Saul Mateos, CFO and head of marketing, technology and HR at healthcare service platform Gain, is a self-taught enthusiast “experiencing significant productivity gains with vibe working.”

His self-built solutions include performance dashboards, reporting systems, data analytics, and automated workflows and investor presentations. He’s also working on a tool to automate daily reconciliations, which “removes a boring painful job for someone.”

But, Gain’s core finance and accounting systems are still bought. 

“I wouldn’t try to vibe code the ERP,” said Mateos. But, if he needs to replace a bought system, he chooses providers without legacy infrastructure to ensure data is accessible and connected within a structured data lake. That way, he or his team could build their own apps on top.

CFOs creating tools themselves makes sense because of their deep knowledge of the business, he said. 

“It’s the most fun I’ve had in my career,” added Mateos. “The more I automate, the more time I can spend on strategy.”

But for CFOs rightly concerned about safety, how does Mateos assure the homemade apps can be trusted? 

“For lower risk activities like internal reporting or automating non-time-sensitive processes, I build myself and push them to production once I feel they help the organization,” said Mateos. 

For higher risk, client-facing work, he builds the prototype, then hands it to his technology team to ensure it's safe, secure, compliant, and robust enough for daily operations.

“Before anything goes live, technology checks security. They trace where the data goes, who has access, and whether anything's leaving our network that shouldn't be. They add access controls and an audit trail,” said Mateos.

“Second, reliability. I'm a vibe coder, not an engineer. I know enough to build something that works, but my prototypes can break, or be slow or inconsistent under real load” he added.  “Technology optimizes all that and makes sure the thing stays up, not just the day I demo it.”

Other ways to ensure safety and reliability could include:

“Some providers are investing in agents, within FP&A tools, who guide users to build additional features. Some even have agents who can do the building for you. That’s the sweet spot – the speed and flexibility of building with the governance and structure of buying,” said Rob Konferowicz, founder at advisory firm CFO Shortlist.

Vibecoding does create a world of possibilities for finance teams to solve long held system problems. 

The big question is whether CFOs and their teams should be using AI to self solve their tech debt, or whether they should trust the software market to evolve and provide better, more robust solutions. 

Most agree that finance teams should leave the ‘hard-tech’ to the pros; no-one should be vibe-coding a custom ERP. But there is a much stronger case to build custom business specific applications on top and around the core finance systems, to solve integration and data fluidity problems.

If you are embracing vibe coding in your finance function… expect these questions from your board:

Want more? Check out this month’s Playbook where The Secret CFO is breaking down how to crack the technical debt in your finance function. Read the last part on data quality here.

And don’t miss the next Boardroom Brief, we’ll be going weekly every Thursday from next week.

If you found this helpful, please forward it to your fellow finance leaders (and maybe even your Board). If this was forwarded to you, you can make sure you receive the next edition by subscribing here.