Most Finance teams treat AI as something they consult when stuck. A chatbot for the hard questions. A fallback when the formula won't cooperate. A tool to clean up the writing before the deck goes out.

That's backwards.

High-performing Finance teams ask different questions before they open Excel. The default mindset asks "How do I do this?" The AI-first mindset asks "How can AI make this better, faster, or unnecessary?"

It sounds like a small reframe. In practice, it changes which work gets done by hand and which gets built once and run forever. It changes what your senior people spend their week on. Over a quarter, it changes what the team is capable of taking on.

The three questions, asked in this order:

01 — How can AI make this better? Sharper output. Fewer errors. Higher quality at the same hour count.

02 — How can AI make this faster? Compress cycles. Same deliverable, a fraction of the wall-clock time.

03 — How can AI eliminate this entirely? The biggest leverage. Tasks that no longer need to exist on the team's plate.

The third question is the one that matters most but it is also the one teams skip most often. "Better" and "faster" feel safe. They preserve the existing structure of the work. "Eliminate" challenges the assumption that the task should be done at all.

That's where the real gains live.

Beyond the Chatbot: Five Capabilities Most Teams Miss

ChatGPT, Gemini, Microsoft Copilot and even Claude are just the entry point, not the destination. The capabilities that actually move the needle for Finance live one layer deeper, and most teams haven't touched them yet.

Code generation, on demand. As I mentioned on previous newsletters and articles, one of my favourite abilities of AI is that it can write Python for you.

You don't need to learn to code from scratch. You need to learn to ask. The output is reusable, version-controlled, and runs without you next month.

One prompting framework I recommend for this is SNAKE. More on it here.​

And some examples of what Python can do for Finance and FP&A here: Ultimate Step by Step Guide to Python for FP&A by Christian Martinez​

Mini-apps, not just answers. AI tools such as Claude can now generate interactive dashboards, scenario planners, stakeholder Q&A interfaces, and self-serve variance explorers. Analysis stops being a deliverable that gets emailed once and starts being a product the business uses on its own. The same effort that used to produce one slide deck now produces a tool a hundred people can query.

Learn more about Claude artifacts here.

Persistent workflows and skills. AI can be trained on your house style: your revenue definitions, your margin logic, your commentary tone, your reporting templates. Once it's encoded, every analyst delivers consistent output. The methodology lives in the system, not in one senior analyst's head.

Data and system integration. Direct connections to ERP and data warehouses. Pipelines that transform raw data into analysis-ready inputs without manual stitching. Schema mapping, multi-source unification, real-time refreshes. The plumbing that makes everything else possible.

Smarter insights, ready to ship. From raw numbers to executive-ready intelligence. Auto-generated executive summaries. Decision-grade chart generation. Narrative explanations of variances. Anomaly detection at scale. The output layer the CFO actually sees.

One mistake is starting with the tools. AI-first teams start with friction. Where is the team spending the most time? What feels repetitive? Where do the delays show up every cycle? That's where the leverage is.

In Practice: Reinventing Variance Analysis

Take as an example the work every Finance team does every month. Same task. Two operating models.

The traditional path:

• Export the data — about 45 minutes every cycle, hoping nothing breaks.
• Build the model — two hours rebuilding variance formulas and patching whatever broke since last month.
• Write the commentary — three hours scanning rows, drafting narrative, editing in Word, sending for review.
• Build the deck — two more hours copying charts into PowerPoint and reformatting for the CFO's preferences.

Hours of repeated manual work. And it has to be redone next cycle, in full.

The AI-first path:

• Upload the data to a secure AI environment and ask for variance commentary in your house style — about 5 minutes.
• Have the AI generate Python that calculates variances, identifies drivers, segments performance — about 10 minutes.
• Spin up an HTML dashboard with visuals and auto-generated commentary, exported and ready to share — about 10 minutes.
• Ship a stakeholder mini-app where anyone can ask "Why is OPEX above budget?" and get an instant, sourced answer — about 30 minutes to build the first time.

Here is a full example solved (you can download it and make it yours).

Built once, runs every cycle. Stakeholders self-serve instead of queuing requests. Finance focuses on judgment instead of production.

That's not a faster process. That's a different operating model.

This is the output it produces:

Coaching the Team: Four Moves to Embed the Mindset

You can't mandate a mindset. You can build the conditions where it spreads.

Move 01: Normalize experimentation. Make "Can AI do this for me?" a routine first question, not a special-occasion one. The cost of trying is near zero, and imperfect first attempts improve quickly. The mantra: try it before you scope it.

Move 02: Reward leverage, not effort. Stop celebrating who worked hardest. Start celebrating who built the smartest solution: the one that runs again next month without them. Hours saved beats hours logged. If you don't change what gets recognized, you won't change what gets built.

Move 03: Build AI playbooks. Document the prompts that work. Catalogue the use cases. Make reusable workflows the team's shared infrastructure, not tribal knowledge stuck in one analyst's head. If it worked once, write it down, and make sure the next analyst who hits the same problem finds it.

Move 04: Move from tasks to systems. The goal isn't faster analysts. It's fewer manual processes and more automated intelligence that scales decision support across the org. Don't optimize the task. Eliminate it.

These four moves compound. Experimentation generates wins. Reward systems make those wins visible. Playbooks turn wins into repeatable infrastructure. And the move from tasks to systems is what separates a team that's "using AI" from a team that's been redesigned around it.

The Self-Audit: How AI-First Is Your Team?

Five honest questions, answered for this quarter — not what you aspire to:

1. When a new analytical task lands, what's the team's first question?
2. Are reusable AI workflows or playbooks documented and shared, scattered, or nonexistent?
3. When you recognize good work, do you celebrate hours worked or leverage built?
4. Can stakeholders self-serve answers from your financial data, or does every question route through an analyst?
5. Does your team experiment with new AI capabilities weekly, occasionally, or rarely?

Use my Claude artifact to answer this Audit live.

If the honest answers cluster on the low end, your team is still treating AI as a separate tool rather than a default approach. Start with one recurring task, it can be variance commentary, a monthly report, and rebuild it AI-first end-to-end. The mindset spreads from one visible win.

If the answers cluster in the middle, you have pockets of experimentation but no compounding leverage. Document what works. Reward leverage publicly. Convert one-off prompts into reusable workflows the whole team uses.

If the answers cluster on the high end, you're operating as a system rather than a service desk. The frontier from here is stakeholder-facing mini-apps, persistent skills that encode your methodology, and treating analysis as product rather than deliverable. Keep going! Every quarter, ask what manual processes still exist and what would make them obsolete entirely.

Want to learn more Applied AI for Finance?

If this resonated, here are five places to go deeper:

• Join the community — The AI Finance Club: ai-finance.club/join-now​
• Structured program — AI Finance Accelerator​
• YouTube tutorials — AI for Finance channel: youtube.com/@christianmartinezAIforFinance​
• My book — Available on Amazon: amazon.co.uk/dp/1394369166​
• Subscribe by email — Future briefings in your inbox: christianmartinezonline.kit.com​

If this was useful, forward it to a Finance leader who needs it.

Thanks,

Christian Martinez