Score AI Against Your Real Problems

Last week I was talking to a CFO I respect a lot. He made an observation that I haven’t been able to stop thinking about. He said if you went around to all of his top people, independently, and asked them to list the seven biggest problems in the finance org, they’d all say the same things. Same list, every time. And his take was that probably isn’t unique to his firm. Walk into almost any finance department, corporate, consulting, public company, doesn’t matter, and you’d get roughly the same list back.

That stuck with me. Because if it’s true, it means the problems aren’t mysterious. Finance professionals know exactly what’s broken. The question is whether anything is actually capable of fixing it.

Around the same time, I came across a Hank Green video where he proposed a different way to measure AI. Forget benchmarks. Forget demos. The real question is: can AI solve the biggest problems people actually face? He used housing as the example. AI can help with permits, admin, some data processing. But the housing crisis is about zoning, supply constraints, and capital access. AI has a role to play. It’s not the fix.

So I started thinking…what happens when you apply that same lens to finance and accounting?

The Known List

Think about it like deferred maintenance on a building that everyone walks past every day. Everyone can see exactly what needs fixing. Nobody is confused about the problem list. Here’s what that list almost always looks like in a finance org:

• The close process is too slow and too manual
• Data lives across too many systems and won’t reconcile cleanly
• Reporting is backward-looking by the time leadership sees it
• Finance spends more time gathering data than analyzing it
• The ERP is either behind, mid-migration, or about to be replaced
• Headcount is always under pressure relative to scope

None of that is a surprise to anyone who’s worked in finance. These problems predate AI by decades. They’ll still be there after the current AI wave if nothing structural changes.

Score AI Honestly — Problem by Problem

Here’s the exercise I’d encourage you to run. Take that list, or your organization’s version of it, and score AI against each item honestly. Not what AI can do in theory. What it actually moves today.

Close process is too slow. AI can automate reconciliations on clean data, flag anomalies, and draft variance explanations. That’s real. But the close is slow because of disconnected systems, manual handoffs, and review chains built around distrust of data. AI speeds up a few steps. It doesn’t fix the process architecture. Score: partial.

Data won’t reconcile across systems. AI can surface where the breaks are happening faster. But if you have two ERPs with different chart of account structures and no integration layer, AI is reading the mess…not cleaning it up. That’s a data infrastructure problem. Score: partial at best.

Reporting is backward-looking. AI can speed up production of the report and generate narrative commentary. Some tools are starting to draft early MDA language. But real-time reporting requires real-time data infrastructure — feeds, pipelines, warehouses, governance. Most orgs don’t have that yet. Score: helps at the output layer, not the input.

Finance gathers more than it analyzes. This is where AI shows up most clearly. Summarizing documents, formatting outputs, first-pass drafts, routine data prep — the assistant-layer use cases. This is real. Probably the strongest area on the list. Score: genuine partial win.

Headcount pressure. AI can extend individual capacity. But if the scope of finance keeps expanding and the efficiency gains just get absorbed by more work, the mismatch doesn’t close. Score: complicated.

The Distinction That Actually Matters

At PwC and in my own work since, I’ve found that probably 70 to 80 percent of the use cases people bring up when they start thinking about AI aren’t actually AI problems. They’re workflow problems. ETL problems. Training problems. Process design problems. Something that needs a deterministic output, not a probabilistic one.

AI has been great at accelerating the conversation around transformation. But learning to distinguish the tool problem from the structural problem is itself a form of professional judgment worth developing. The structural blockers — bad data, legacy ERPs, incentive misalignment, broken processes — are still structural. AI can work alongside them. It doesn’t shortcut them.

Key Takeaways

• Finance’s biggest problems are largely universal — and they existed long before AI arrived
• The right benchmark for AI isn’t a demo or a vendor scorecard — it’s whether it moves your actual pain points
• In most cases today, AI plays a partial role against the biggest problems, not a complete fix — and that’s an honest assessment, not a knock on AI
• Structural blockers — broken data, legacy ERPs, misaligned incentives — are still structural; AI doesn’t shortcut them
• Running a “score AI against your real problems” exercise is a practical way to prioritize AI investments for yourself, your team, or your clients

Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]