How to Use AI at Work: Real Examples for Finance and Accounting

I am a Finance Program Manager at Google, and I use AI almost every day. Not to do my job for me, but to do it faster and better. That distinction matters, so I want to be clear up front about how I think about this. AI is an accelerator, not a replacement. It can get you about 80 percent of the way on most tasks. The last 20 percent, the part that makes the work yours, is still on you. I call it the last mile, and it is where your judgment, your context, and your voice live.

Here is the trap I see finance and accounting professionals fall into. They paste a request into ChatGPT, copy the answer, and send it. If everyone does that without iterating, we all start to sound like ChatGPT. The output is fine, but fine is forgettable, and in our field forgettable can quietly cost you credibility. So let me walk through the specific ways I put AI to work, and the guardrails I keep on at all times.

Build process flows that expose your bottlenecks

One of my favorite uses is turning a messy process into a clean diagram. Take the month-end close. I will ask a tool like NotebookLM to create a one-slide process flow for the close using a clear L1, L2, and L3 hierarchy in a professional finance style, showing the end-to-end cycle in logical phases. In seconds I get a structured map: collect financial information, verify and reconcile data, adjust for accruals and prepayments, prepare statements, conduct the final review.

The real value shows up when you layer in pain points. When I map a close and mark which steps are manual versus automated, the bottlenecks jump off the page. One team spends the first three days of the month manually pulling transactions from ERPs, spreadsheets, and bank portals, so by the time the data is compiled it is already stale. Another spends 40 percent of its time matching ledgers to credit card statements by hand. AI did not find those problems for me. It gave me a clear picture fast so I could find them myself.

Manage projects with a dedicated AI thought partner

For every project I own, I set up what I think of as a project agent. I feed it the artifacts that already exist: the business requirements doc, the product requirements doc, weekly meeting notes, design documents, and prior slide decks. Then I use it to draft leadership updates, weekly progress notes, and first-pass decks.

The benefit is simple. Instead of starting from a blank page every week, I iterate on something. That frees me up for the higher-priority work that actually needs my brain. I also lean on Gemini to handle transcription and meeting notes so I can focus on driving the conversation across teams instead of scrambling to write everything down.

• Pull your project documents into one place as the source of truth.
• Use the agent to draft recurring updates and summaries.
• Always edit the draft before it leaves your hands, and put your own voice back in.

Turn a wall of data into an executive summary

Executives do not want the full analysis. They want the headline, the impact, and the recommendation. AI is excellent at the first compression pass. Say you have to tell your VP that subscription revenue dropped. I will hand the tool the underlying numbers and ask for an executive summary based on the project artifacts.

It might come back with something like this: in 2025 we saw a 300 million dollar decrease, down 24 percent, driven by roughly 500,000 customers unsubscribing from Premium, TV, and Music after a Q4 price increase. Premium alone drove 175 million of that loss and 60 percent of the churn, concentrated in the U.S. mid-income segment earning 75,000 to 100,000 dollars. That is a strong skeleton. But I do not stop there. I tighten it into a clean impact table, sharpen the recommendation to validate root causes and prioritize Premium retention, and make sure the framing matches how my VP actually thinks. The 80 percent was AI. The last 20 percent, the part the room remembers, was me.

Brainstorm and structure slide decks

I use AI as a thought partner to accelerate the brainstorming phase of deck building. It is great for getting a rough structure, suggesting an order of arguments, and pushing past the blank slide. From there I influence by storytelling, which is a human skill no model has. The deck still has to carry my point of view, not a generic one.

Prompt with structure: CO-STAR

The quality of your output depends entirely on the quality of your prompt. The framework I rely on is CO-STAR, which Sheila Tao used to win Singapore's GPT-4 prompt engineering competition. You give the model Context, an Objective, a Style, a Tone, the Audience, and the Response format you want. A request built that way returns something usable. A lazy prompt returns AI slop you have to throw away.

The authenticity guardrails I never drop

Here is the part most people skip, and it is the most important. The moment your work reads like it came from a machine, you lose trust. Researchers at Berkeley describe a formula for authenticity built on three things: credibility, transparency, and reputation. One line that sounds artificial can defeat the credibility of an entire document. People decide whether to believe you in the first few sentences, and one robotic phrase tells them to stop trusting the rest.

So I hunt for the stylistic footprints AI leaves behind. The biggest tell is the em-dash. Its use jumped from roughly 3 per 1,000 words to about 10 per 1,000 words after these tools went mainstream, and some researchers banned it from their tests because it had become a reliable AI fingerprint. I write with commas and hyphens instead. I also strip out the buzzwords models lean on: delve, leverage, orchestrated, significant, critical, foster. And I cut the cliches: paradigm shift, harnessing the power, seamless. If a sentence could have been written about any company by anyone, it is not done yet.

• Read every draft out loud. If it does not sound like you, rewrite it.
• Delete the em-dashes, the buzzwords, and the cliches before anything ships.
• Add the specific detail only you know: the team, the number, the real constraint.

The bottom line

AI lets me move faster on flowcharts, decks, project updates, and analysis than I ever could alone. But the work that earns trust at Google, or at any Big 4 or Big Tech firm, is the work that still sounds like a person who knows what they are talking about. Let AI carry the first 80 percent. Own the last 20 percent in your own voice. That is the whole game.

I teach this live and for free. If you want to see these workflows in action, schedule a session at summitresume.com/resources.

Want the complete roadmap? Read The Complete Guide to Breaking Into Big Tech Finance.