Five Principles for Getting AI Right

In Issue #1, we covered why AI matters now and where finance leaders should start. We promised five principles for getting AI right, with examples you can apply immediately. Here they are.

These come from watching what works — and what doesn't — across companies of different sizes and industries. A cleaning company in Vancouver. A $1.7 trillion sovereign wealth fund in Oslo. A SaaS company running 20+ AI agents with three employees. The principles are the same.

Before hiring, before outsourcing, before adding another manual process — ask: could AI do this, or meaningfully assist? Instead of "Why use AI?" — ask "Why not use AI?" That single inversion changes how a team evaluates every new project and every open role.

Leading from the top

Two organizations have made AI-first an explicit, company-wide standard — and both insisted it applies at every level, including the CEO:

Shopify: CEO Tobi Lütke directed in April 2025 that teams must demonstrate why AI can't do the job before requesting headcount. AI usage is now part of performance reviews. Lütke was explicit: the mandate applies to "all of us — including me and the executive team."

NBIM: At Norges Bank Investment Management ($1.7 trillion sovereign wealth fund), CEO Nicolai Tangen personally drove AI adoption for two years, making it a standing item at every leadership summit. Training was made mandatory for all staff. As their AI lead put it: "The people who don't want to do it are the people who need it the most."

You don't need a technical background to start

Rick Chorney founded Echo Janitorial Services in Vancouver. He went from $14-an-hour subcontracting to projecting $1.3 million in sales this year — using off-the-shelf AI tools to build sales and reduce administrative burden, with no technical training. His AI adoption followed a natural staircase:

His workday went from 19 hours to eight.

Source: "Meet a 29-year-old blue-collar founder who used AI to triple his revenue in 3 years," Fortune, March 28, 2026

The biggest risk with AI isn't that it produces obviously bad output. It's that it produces output that looks polished and authoritative but contains underlying errors or omits important information — and that AI agents take actions outside their intended scope. As AI moves from generating documents to executing tasks autonomously, the potential consequences expand: an agent with access to email, financial systems, or client data can act on flawed instructions, expose sensitive information, or create security vulnerabilities. Mediocre output gets questioned. Convincing-but-wrong output — or an unchecked agent action — gets acted on.

Without explicit quality expectations, teams accept whatever the tool produces. For finance leaders, the test is specific:

As we noted in Issue #1, we've found material data errors in AI-generated financial models published by two separate venture capital firms. The outputs passed a casual read. They wouldn't have survived informed scrutiny.

One of the most common questions executives ask is "Which AI model should I use?" Some get overwhelmed by the pace of new releases and the jargon that comes with them. The result is paralysis — waiting for the next version, the next benchmark, the next analyst report.

Resist it. The companies gaining an edge aren't the ones with the best tools. They're the ones that started. Don't chase the latest model release — build the skill of using the tools you have. Start with a contained use case. Don't wait for AI tools to be perfect before you start learning how to use them. Even Excel, the world's dominant spreadsheet, still has flaws — yet we've all used it for decades.

AI agents require sustained investment in training, management, and organizational design.

Training is not optional

SaaStr, the world's largest B2B SaaS founder community, has publicly documented deploying 20+ AI agents across its go-to-market and operations. Their framework for each agent:

One of SaaStr's agents stopped ingesting new data for four months — it kept running and looking normal. They only caught it when results felt subtly off.

NBIM ran mandatory AI upskilling for all staff, trained 20 AI ambassadors, then had to run a second round of upskilling within months as the technology moved faster than the training.

Plan for organizational design changes

SaaStr: Chief AI Officer now spends her morning hour managing AI agents rather than holding 1-on-1s with human sales reps

NBIM project teams: Shifted from traditional scrum (8 developers + 1 business person) to 2 developers + 1 business person with AI, eliminating most scrum ceremonies

NBIM financial reporting: A team of 2 non-developers used Claude Code and Cursor to replace complex Excel workbooks. Reporting that took a person a full week is now done in hours.

The CFO takeaway: Ensure that AI initiatives include sufficient budget for training and ongoing management.

Sources: SaaStr, "From Zero to 20 AI Agents in 10 Months," October 2025; NBIM, "How we use AI in practice," AI Summit, March 2026

An AI agent is best understood as an employee you've delegated authority to — explicitly or implicitly. The accountability doesn't transfer to the tool.

This matters more as AI moves from assistant to taking real actions in business workflows. If an AI agent sends an email to a client, processes a payment, or produces a financial report, the responsibility sits with whoever authorised the workflow — not with the software.

Design AI workflows with clear scope and human review at key decision points. The question isn't whether to use AI — it's whether you've built the right controls around it.

Your team is already using AI tools — through official channels and on their own. That's not a problem to shut down. It's information to act on. Pick whichever step you haven't done yet.

1. Apply the AI-first test to one real decision this week

Before your next hire, next research project, or next process improvement — ask: could AI do this, or meaningfully assist? Build the habit before you need the answer.

2. Take stock of your AI tools and experiences

Ask individuals and teams what AI tools they're using, how, and for what — and what's working. Understanding where AI is already delivering value is as important as knowing where it isn't.

Data security and privacy when your team is using AI. What CFOs and CEOs need to know, and the questions your security team should be answering now.

Forward to a CEO or finance leader who would find it valuable.

To subscribe: newsletter.aperna.com

Looking for a fractional CFO — or just AI guidance?

We work with CEOs and finance leaders at small and mid-sized companies. Fractional CFO leadership and practical AI — together.

Reach out to us at [email protected]

Brian Hendry, CPA, CFA, MBA · Founding Principal, Aperna Inc.

AI with Purpose: Drafted and reviewed by humans with the help of Claude AI, Perplexity and NotebookLM.

_{AI Disclaimer: This newsletter uses AI-assisted content and tools. AI can make mistakes — please verify any information before acting on it. Aperna Inc. reviews AI outputs but cannot guarantee their accuracy in all circumstances.}

_{Professional Disclaimer: The content of this newsletter is provided for general informational and educational purposes only. It does not constitute accounting, tax, financial planning, or investment advice. Aperna Inc. and its principals hold various professional designations; however, no professional advisory relationship is formed through this newsletter. Consult a qualified professional before making any financial, tax, or investment decisions.}

_{References to third-party companies, products, or services are for illustrative purposes only and do not constitute an endorsement or business relationship.}

Aperna Inc., Toronto, Ontario, Canada.

© 2026 Aperna Inc. All rights reserved.