The following column is by Siqi Chen, the CEO of Runway Financial. Any opinions are those of the author.
In the first half of 2025, companies eliminated thousands of roles tech, retail, media and finance.
Most of the statements sounded the same: streamlining operations, reshaping cost structures and leaning on AI. It was hard to argue with the logic. The numbers checked out. And from a distance, many of the roles that disappeared may even have looked redundant.
Finance was one of them.
But the thing about finance is that it always looks automatable from the outside. You see the data, the models, the repeatable workflows and assume this part of the business can largely run itself.
That’s a mistake.
What Finance Is
Finance looks mechanical: reporting, budgeting, last close, forecast.
But the real work happens underneath.
It’s the only function trained to simulate the whole business. It’s the one place where product roadmaps, sales targets, hiring plans, customer behavior and market volatility all come together.
Sometimes the job is to check if the numbers add up. More often, it is to answer strategic questions: What happens if we pull this lever instead of that one? What breaks if we delay? What bet is worth making?
The spreadsheet isn’t the value. The reasoning is.
Which is why handing it over too quickly to machines comes with a cost you can’t always see right away.
The Automation Paradox
Some of the most obvious wins in finance are also the least important. You can automate invoice matching, forecasting and pipeline rollups. You can even have AI generate the board deck. And all of that saves time.
But AI doesn’t answer questions like:
- Can we afford to take this risk?
- What do we lose by waiting another quarter?
- How sensitive is this outcome to the inputs behind it?
AI can help finance get there. It can point out outliers, flag correlations and highlight variables that matter. It can even propose next steps.
But it doesn’t know where you’re trying to go, or what tradeoffs you’re willing to make to get there.
That’s what experience and judgment are. And those still belong to humans.
“One day, when someone asks why customer acquisition cost is spiking, no one knows anymore—because no one’s had to explain it in months.”
The Line Gets Crossed
It starts small.
First, you automate month-end. Then forecasts. Then scenario planning. Then decision support.
Each one saves time. Things go faster, and everyone seems relieved.
But then people stop touching the model. They stop revisiting the assumptions. Edge cases stop getting logged, and slowly the intuition fades.
One day, when someone asks why customer acquisition cost is spiking, no one knows anymore—because no one’s had to explain it in months.
A Familiar Pattern
When AI fails in finance, it rarely looks like an AI failure. But the pattern is recognizable:
- GPT‑4 Turbo failed 81% of finance questions in FinanceBench.
- Knight Capital lost $440M in 45 minutes due to a trading script that no one had reviewed.
- Interactive Brokers lost $48M from misinterpreting NYSE data.
That’s what happens when humans step away from systems too soon.
At most companies, something slips every quarter. A product launch gets delayed, a sales target gets missed, or a major campaign ends up underperforming. When that happens, most teams recalibrate and move on.
Finance doesn’t.
Finance is where all of those slips get caught, simulated, explained and absorbed. It’s the only function designed to see how one team’s decision affects another’s forecast—to tell you where your model just snapped and how to reroute. Without that role, the whole system starts to drift.
When Finance Steps Back
Every good finance team builds a kind of muscle memory over time.
You remember which quarters were messy. You know which metrics always run hot. You develop a sense of when 8% growth means actual progress (and when it doesn’t).
When you over-automate, you lose that intuition. You lose the context behind the numbers, as well as the reflex to ask if something even makes sense.
The dashboards continually update themselves, the metrics continually move and decisions start to be made with no one left to challenge them.
What’s Worth Automating
Automation isn’t the problem; not knowing where to apply it is.
AI should speed up the work without replacing the thinking behind it.
- Automate: Vendor matching, workflow routing, forecast generation and report formatting.
- Protect: Scenario modeling, capital allocation, tradeoff arbitration, judgment, intuition, and narrative.
The best finance teams use AI every day. They use it to:
- Surface what they might’ve missed.
- Save time on the tedious stuff.
- Ask better, deeper questions.
But they keep humans in the loop:
- Every material decision gets a second look.
- If the model can’t explain itself, it doesn’t ship.
- A person, not a system, always owns assumptions.
The goal is to make human judgment and intuition sharper.
Finance makes the consequences of each business decision visible. It’s where your company reasons about the future. It’s also where all the messy, inconsistent, contradictory inputs from across the org get reconciled into one coherent story.
Creating this kind of shared understanding and intuition goes far beyond predicting the future.
You can’t automate that away.
Because if you cut the part of your company that thinks and models how everything connects, you don’t get more efficient; you get lost.
