Your AI Strategy Is Backwards (And Your Team Already Knows It)

Every finance leader I talk to asks the same question first: "What AI tools should we buy?"

That's the wrong question and it's costing you months of progress.

I recently sat down with Angela Liu, founder of GaapSavvy, to compare how we actually use AI day to day. Between my work leading AI Deals Desk at Google Cloud and Angela's experience building product strategy for finance teams, we've stress-tested these tools across deals, revenue recognition, FP&A, and education. What we found surprised us both: the habits that matter most are the opposite of what organizations are being sold.

But before I share what actually works, let me tell you what changed my perspective entirely.

My students rated me in the top 10% of professors last semester. They also told me to stop using AI so much. They essentially said: "Devon's really authentic and he cares about us. We don't like when he uses AI. It feels like he's phoning it in."

That feedback stuck with me. Because it applies far beyond the classroom. Your board, your team, your stakeholders can tell when you're phoning it in too. AI fluency matters, but it's a multiplier on your judgment, not a replacement for it. 

Here's what Angela and I have learned about getting that balance right.

Start with Access, Not Software

The first step isn't evaluating vendors or building a custom solution. It's giving your team direct, secure access to foundational large language models: ChatGPT, Claude, Gemini. This is ground zero for AI capability building.

Many organizations try to build an in-house version on top of a base model. This approach fails because your functionality will be diluted, and your engineers cannot build fast enough to keep pace with how quickly these models evolve. You'll always be behind.

Give your team unfiltered access to the leading LLMs. Let them play, learn, and discover applications you haven't imagined yet. Security concerns that blocked this two years ago have significantly improved. Enterprise-grade solutions from OpenAI and Microsoft now make direct access feasible for companies of all sizes.

Master the base models first. They're your foundation.

Niche Tools Have Their Place (If You Choose Carefully)

This doesn't mean all specialized tools are worthless. Angela and I both use niche applications selectively.

The difference is knowing what to look for. A tool like TABot works because it's built by an ex-Google engineer who understands technical accounting deeply and iterates based on real practitioner feedback. That's different from a startup slapping a chat interface on GPT and calling it an "AI accounting solution."

The filter: Is the tool built by people who understand your work and have deep expertise in engineering, or is it just a wrapper trying to capture a market? If it's the latter, you can probably replicate what it does in the base models yourself.

Base models are the bread. Specialized tools are the butter. Get the order right: you'll look foolish if all you bought was butter and you forgot the bread.

Your Team Is Already Using AI (Whether You Approve or Not)

Here's a reality every finance leader must accept: your team is using public AI tools regardless of company policy. They're pasting data into ChatGPT. They're drafting memos in Claude. It's happening. Your team is curious. They want to upskill. They're trying to stay relevant in a profession being reshaped in real time.

This places responsibility on you to create a secure environment where this activity can happen responsibly. You can either pretend it isn't happening while sensitive financial data gets exposed, or you can lead by providing a sanctioned space for them to learn.

One idea I love is AI Play Days. Instead of lunch and learns or office hours, set aside curated time with an expert (internal or external) to work through live demos and use-cases. This will give your team freedom and permission to explore use cases in a safe environment with expert guidance.

The choice isn't whether your team uses AI. It's whether you create a space that’s safe for them to use it and share what they are learning with others.

Each Model Can Do Everything. But Each Has an Edge.

All three major LLMs can handle most tasks. The difference is which one does specific things better right now.

ChatGPT currently leads on communication. When I need to navigate tone or turn a draft into something professional, ChatGPT handles it best. It has the highest emotional intelligence of the three. Its long history of memory makes it clearly differentiated for personal use cases.

Claude currently leads on analysis. I asked Claude to build a seven-tab valuation model last month: discounted cash flows, terminal value, sensitivity tables, color-coded with proper formatting. That model would have taken me a full day to build from scratch. An intern would need a week. With Claude, I built and validated it in an hour.

Gemini currently leads on research, especially paired with Notebook LM. Notebook LM is the real advantage here. You can upload earnings call transcripts, SEC filings, analyst reports, even YouTube videos you haven't had time to watch. It synthesizes everything and generates presentation-ready slides, summaries, flashcards, and study guides. Research that used to take me five days now takes 30 minutes, with polished materials ready to share.

These edges shift constantly. What's true today won't be true in six months. Staying engaged with all three is how you keep your advantage.

Use AI to Critique, Not Create

This habit changed my workflow more than any other. I stopped asking AI to write things for me. Now I write first, then ask AI to critique what I wrote.

My go-to prompt: "Rate this out of 10 based on completeness, accuracy, and ease of understanding."

Then I make the improvements myself.

This keeps the skill development with me. The AI coaches; I execute. In a profession where judgment is everything, outsourcing the thinking is how you become replaceable. Using AI to sharpen your thinking is how you stay essential.

Curiosity Beats Expertise

In a field changing this fast, the traditional goal of becoming an "expert" is a trap. For finance and accounting, where the culture rewards having all the answers, this mindset is now a liability. The more valuable trait is curiosity. The willingness to experiment and be comfortable not knowing.

Here's what that looks like in practice: during a team share session, employees discovered they could use Google App Scripts to stitch together fixed NetSuite reports. This was a task they assumed required an engineering ticket that would sit in a queue for months. By exploring on their own, they built a functional script and pushed past a boundary they didn't know they could cross. That's the ROI of a curiosity-driven culture. It comes from giving people permission to play.

Leadership Is the Bottleneck

None of this happens without visible support from the top. Creating space in the schedule for AI exploration. Allocating a real, substantial budget for training and premium tool access. Showing up personally and sharing your own experiments, including what didn't work.

I'll share Angela’s example: She was researching SEC filings using one tool, then tried Perplexity Finance, which pulled from analyst reports and Reddit threads she had to sort through with her own professional judgment. Messy. Imperfect. That's the kind of messy exploration I want to see from my own teams too.

It's hypocritical to expect your team to embrace AI if you're not using the tools yourself or putting budget behind it. You cannot ask anything of your people that you don't first ask of yourself.

Authenticity Is the Skill That Won't Be Automated

My students taught me something important: they're not impressed by efficiency. They spot AI-generated content instantly, and it signals that you didn't care enough to show up yourself. What they value is genuine expertise and someone willing to be direct with them.

The goal isn't to produce more deliverables faster. It's to use these tools in ways that free you up for the work only you can do: the judgment calls, the difficult conversations, the strategic decisions that require someone who's been in the room.

That's what separates the finance leaders who will thrive from those who will be replaced.

Want to Go Deeper?

Angela and I are building live training cohorts and a certification process for a select group of finance professionals who want to learn this from the ground up. Not surface-level tips. Real workflows, real applications, and the chance to build world-class use cases together.

If that's you, drop a comment or send me a message. We're keeping the first group small intentionally.

Watch our full conversation here: https://youtu.be/Fxo2bGIlRIY