AI is both the most over-hyped and under-hyped technology of the last few years. I get it. Your data team, RevOps team, and accounting team are still spending hours reconciling numbers across disconnected systems. Meanwhile, everyone is burning time on ChatGPT, Gemini, Grok, etc., doing personal analysis that nobody else understands.
I can relate. When Gen AI first came out, everyone on my team implemented it everywhere, all at once. It was exciting. Addictive, even. Always one prompt away from perfection.
Instead, we got buckets of mindless content.
One month in, the danger became clear:
Notion pages were flooded with AI-generated words that sounded right but meant little. There was content, but not clarity, much less action. It asked the question, did the person even think or believe what was on the page?
Analyses became irreconcilable. Several cohort analyses had nice insights, but no one could reproduce them, or even tell if it was right, once the CS team got involved.
Strategy work was being submitted fully generated by AI. Not just polished, fully written. It sounded impressive (and it wasn’t baaad…) but once we dived into the details it made no sense within our team structure or our market.
As a result, people stopped reading content (maybe so they could get back to their AI conversation). Skimming hit an all-time high. Slack became a disaster. It was obvious we were creating a giant slop of AI content and spending a tremendous amount of time doing it, without seeing a single bottom-line result.
The company still wasn’t doing that hot.
AI, in all its forms, should help you be quicker, more accurate, and more persuasive. Structured well, it can, especially for FP&A. But to do so, you need to set the expectations and standardize its use so that it enables better outcomes rather than obfuscating the company’s knowledge.
Quick note: This is for those who want to move fast, quickly with AI. This doesn’t necessarily apply to big or public companies that have a host of data-type restrictions and giant tech teams.
The Bullets
|
Let’s get to work.
1. Encourage, Encourage, Encourage
Everyone is using it. And if they aren’t, they should. So, the first step is to embrace and encourage its use in producing better, well-informed results. We’ll get to how to set the expectations next, but first, encourage where it’s already being used well.
Bring it out into the open. Everyone is using it. Have an informal conversation about how people are using it, and start with how you are using it.
Example: I write a deep draft of the newsletter, and then use AI to copy-edit or provide resources. It saves me a tremendous amount of time on being grammatically correct and linking to other articles, but not much beyond that.
Track how it is being correctly used. Put up a quick spreadsheet on examples of its use within the finance team. This could be invoicing tagging, a Python analysis created by AI, data tagging and cleaning, etc.
Keep track of new wins. Whenever someone uses AI effectively to build something useful, put it in the tracker. Go ahead and assume that nearly everyone is using it for writing.
Be honest about the losses. Share the horror stories as well. That time you sank in tens of hours into building an analysis that only would’ve taken an hour to do. Or, when you spent five hours deep into an analysis only to realize that AI wouldn’t work in that case.
By taking these quick steps, you will have already encouraged a much healthier culture around AI than letting everyone do it quasi-secretly. The wins feed keeps morale high, sparks fresh ideas, and nudges even skeptics onto the bandwagon. Momentum compounds without formal mandates.
Tip: Publish the ‘AI win examples’ on a public Notion page so that everyone can see what it means to have success.
2. Set the expectations for results
Better outcomes, not more slop. You want to set a clear expectation for the outcomes of using AI. You don’t need to lock down every detail as the landscape is still the wild-wild-west, for better or worse. But the bar should be clear.
Time spent vs. Time saved. The ROI needs to be there. (See the earlier point about being honest about the losses.) The goal is to spend less time, not just shift time around.
Reproducibility. The great thing about spreadsheets (usually) is that you can update them. AI makes this harder. The expectation should be: if you use AI for an analysis, you share the exact directions or prompts that led to the result, so it can be redone later. Even better, have the AI turn it into code to ensure everyone can run.
Persuasiveness. Like any other work, what matters in the end is whether it’s persuasive. That means: Written work shouldn’t turn people off by sounding like AI. Analyses should show the ‘why,’ not just the result. Your team needs to be convincing people to take relevant action.
Accuracy. Every number should be double-checked and understood. The expectation is that every data point can be traced back to a clear calculation or source.
AI doesn’t change the expectation for great work. If anything, it raises the bar because people who use it effectively should be able to do things faster and spend more time actually improving the outcomes.
Tip: Ownership. Ownership. Ownership. At the end of the day, whoever puts the AI-generated output into the work is responsible for the result. If it’s wrong, they’re wrong. If it’s great, they’re great.
3. Standardize and document use
To make AI easier, and more useful, you need to standardize, at least on the tools you’re using and how they’re being used. This allows people to share learnings and build on what others are already doing.
Use the same tool. Everyone should be using the same AI tool for analysis. Pick one. Use it. Provide access to everyone.
Standardize the data access. This takes a bit more work, but it’s critical. Everyone should be pulling from the same data sources, not copy-pasting things in from different places. Enable a setup where anyone doing analysis uses the same structure and definitions.
Document the prompts used. Treat AI-driven projects like reports. Document how AI was used and how it can be used again. This doesn’t have to be lengthy. Just make sure it’s clear and reproducible.
Formalize the hypothesis. Every project, task, or analysis should start with a clear view of what you’re trying to solve and how AI could be applied. This avoids wild goose chases.
Example: “AI, build my five-year strategy!” Let’s not.
This will immediately get your team speaking the same language and producing work that others can understand and build on. At this point, nearly all major general-purpose AI tools are functionally similar. It’s less about features and more about consistency.
Tip: It helps to have a single data layer that ingests all your key sources, then you can run trained AI models on top of that.
4. Embed ownership & automation into role and project
Yes, most people have seen the trend: companies saying, “No more hires unless it’s proven AI can’t do it.” Crude? Yes. But also—maybe not the worst mindset. Here’s how you can apply the same principle more constructively within your team.
Quarterly mandate. Each teammate automates one high-pain task: cash waterfall, ARR bridge, QBR refresh. One quarter, one real improvement.
Bot-first rule. Every new analysis starts with /ai first_pass. Let the machine rough it out. Then humans frame the narrative and finish the story.
Constant improvements. The pace of work is speeding up. Expectations should too. By pushing toward constant improvements, you’re naturally pushing toward more automation and better outcomes.
Ask good questions that push AI. It’s not “Can AI do that?” It’s: “You’re spending a lot of time on this reconciliation, what would an easier version look like in an ideal world?” Then maybe: “Talk to X about setting that up.”
These aren’t necessarily less crude than “Don’t hire,” but they’re more constructive. They show how you, as a leader, can actively push AI adoption in a way that’s sustainable and empowering.
Tip: The easiest way to learn is by doing. No need for formal AI training—just start. Ask the AI. Take on a project. If people get stuck, they should ask someone else on the team who’s done it.
5. Looking into the future
In the near future, or now, with cutting-edge tools, you’ll have a data structure that enables expert AI to generate insights instantly, or even deploy agents to do the work for you. To build toward that future, focus on these fundamentals:
Good system hygiene. The cleaner your data, the more useful your AI. Just like with people, messy data leads to bad results. Structured, clean, well-labeled data is the foundation.
Keep up to date with tools. There’s a lot of hype but also real advancements. Stay plugged in. Take a demo or two. You don’t need to chase every shiny object, but you do need to stay aware.
An informed team. As people rack up their own AI wins, they’ll naturally become more excited about larger projects. You’re building literacy through action.
Consolidation. The real promise of AI in FP&A isn’t about plugging it into every tool. It’s when AI can ingest all your data and synthesize insights across functions. That’s the real game-changer. Prioritize building centralized visibility before chasing isolated use cases.
This approach is a level above everyone running around using AI differently and producing wildly different results. Very few people can run reinforcement training or structured outputs effectively at the individual level, and it’s a lot more powerful when built centrally.
In conclusion
AI can supercharge your results, or result in a meaningless jumble of information. If you want it to drive results, treat it like any other tool: set expectations, build structure, and track outcomes. Without that, you’ll end up with a graveyard of Notion pages and Slack threads no one will trust. But with it? You'll achieve better outcomes as a company.
