Lots of people like to make fun of large companies…

And while these might be true in certain cases, there’s 1 hidden benefit nobody talks about.

Big companies have budget to do cool things.

And luckily for me, that means I’ve been working with and testing AI in FP&A for years.

Sure, the landscape is new/different today with the kinds of applications that are possible, but I’ve at least seen firsthand what works and what doesn’t in a company context over time.

Let’s spend the next 5 minutes dissecting what that means for your finance team…

Before we talk about best practices for AI in FP&A, we need to talk about where I’ve seen finance leaders burn their time and energy before learning a hard lesson.

When I see overwhelmed finance leaders go into a reactive sprint to implement AI, it typically comes across as a “let’s put AI on everything - report, analysis, forecasting, consulting - all of it, NOW!”.

And while this is a fine vision, it’s a terrible implementation plan.

The most recent I saw (that I’ve shared before) is that a SVP of finance engaged an AI vendor to ‘run the forecast through AI to see if it’s more accurate’.

Here’s why that’s a bad idea:

A decent amount of what we find in our financial results is manual adjustments - accounting adjustments to correct an erroneous journal entry or expense allocation.

Or it’s driver-based - for example, call volume from an outage caused a spike in call center expenses and ultimately customer cancellations.

And while these are decent use cases individually for AI to assist with, simply running your aggregated historical results through AI without giving it that additional context will produce a mediocre result.

Another pitfall is assuming AI will get you an outcome that is only attainable by a better process.

Years ago we had an initiative to systematically catalog all the analysis that happened by all finance teams across the company (across a 300+ person finance org). The goal was to share insights and give future analysts a head-start if they saw a similar result show up and didn’t know what might be the cause.

A couple years later the effort was abandoned - the finance org wasn’t consistently uploading their analysis (less than 30% adoption rate) so it was shut off.

But could you imagine how powerful that data would have been with an AI chat layered on top today?? Lot’s of potential was lost here due to bad process management.

That’s a clear lesson to 1) fix the process then 2) layer AI on top.

With that out of the way, let’s get to the good stuff…

Like any thoughtful strategy, this one comes in 3 layers. There is nothing unique about this compared to what your software strategy should be for FP&A - meaning you can simply remove the word AI and replace it with the word software.

Let’s jump in…

Which part of a house gets the most attention?

Definitely not the foundation, framing, or drywall. And the AI in FP&A equivalent to that is your infrastructure layer. While it’s not the most popular with the masses, it’s the most critical to get right if you want to enable AI.

This layer varies widely between startups (a simple FP&A tool will work) and enterprise (custom data pipelines and warehousing feeding a BI layer) but is the foundation of an FP&A team’s ability to produce insights with AI.

As they say in the data world: garbage in, garbage out.

Which is why it’s critical to get all your data in 1 place, sanitized, and organized for analytics. It’s also why enterprise finance teams have data governance, data engineering, and data scientist sub-teams at scale. KPI definitions matter a lot especially at scale.

If you haven’t already, now is you time to pull the trigger on improving your infrastructure layer within FP&A. There are very few opportunities you’ll get to embed advanced analytics and AI into the DNA of your FP&A team practically overnight.

This won’t solve all of your problems or cover all your needs.
That’s fine.
But it’ll probably cover 80% of your outputs:

But that last 20% of your outputs is what causes your team a headache each month. Let’s talk about how to address that…

Every industry is different.
Every company is different.
Every month board is different.

Which gives rise to the need for more customization outside of your infrastructure layer.

I’m talking about those tasks you generally repeat every month or quarter (or on-demand) but specifically look different every time.

Here’s a few examples to make it real:

Both of these are examples of use-case tools for finance. And the coolest part about these in the age of AI of 2026 is that both of these can be built internally by finance teams.

Historically only enterprise companies who employ the right kind of talent could build internal software (it’s cool, I’ve lived it), but with tools like Claude Code it’s entirely possible for an above-average technical finance person to build internal tools.

If you haven’t already, I’d highly encourage you to look at your processes that require some level of: data retrieval, data processing, analysis, forecasting, or communication (so pretty much everything in finance) and ask Claude if it’s possible to build with Claude Code.

Here’s a prompt you can use:

Share your output, what you’ve built, or where you get stuck. I’d love to hear.

The final layer is the shiny object of 2025 and 2026 (so far).

It’s tools like:

Just to be clear, I’ve used almost all of these specific examples. And they are super cool.

But for most FP&A teams they aren’t the highest ROI. In fact, many of them are shiny objects and distractions from adding real value to the business.

Yet, they still have a place.

I recently dove deep into Claude in Excel to give my honest first impression. It’s awesome to watch a robot build a model, update a model, and analyze the data faster (and better) than a person.

But again, that’s not the constraint of any FP&A team I’ve worked with.

My assessment of these tools is that you might get a 10-20% productivity boost out of them - and your team should all be using one of them (especially the Excel add-ins).

But a non-technical person setting up Moltbot is a waste (check out twitter if you’re wondering what they hype is about)

The homework here is to find one tool (I’ve heard Shortcut is great), get IT access for it, and start becoming fluent in it.

That’s what I’m doing.

Say hi 👋 on LinkedIn