This four-part series explores how FP&A can use AI without losing the challenge, ownership and judgement that make planning effective. Across the series, we examine the risks of over-relying on AI, the importance of keeping people and business context in the planning process, and the guardrails, checkpoints and accountability needed to use AI well in FP&A.

If your finance team is small and has already started budgeting and forecasting using AI, the numbers suggest you are onto something. According to the 2026 FP&A Trends Survey, The Readiness Gap, 88% of those teams that use AI stated that data now influences the majority of their decisions, and 64% assessed their forecast accuracy as good or great.

The survey is careful to describe this as an association rather than proof, and so am I. It does not demonstrate that AI by itself leads to a better forecast. But it does show that the teams who have reached for these tools are more likely to say their planning has improved.

So go ahead and use AI. It is more accessible than ever before for smaller finance teams. But be warned: cheap, easy access to the capability is not the same as a capability you can rely on.

Clever Is Not the Same as Adequate

Recently, I met with an accountant who had built an in-house for FP&A, using a public AI. It was clever, and it was genuinely impressive. But there was a pretty significant “but” attached. They did not fully trust the numbers, and they were worried the token costs would eat any benefit they got from it. Anyone who has run AI tools at volume knows that worry. The costs mount every time you revise and iterate, and they vary wildly depending on which model you are using.

This is a position that many smaller teams navigating AI for FP&A are likely to find themselves in. The tools are accessible enough that you can build something clever without a data science department behind you. But you need to ask yourself whether clever is adequate. Reaching for a public AI brings a set of challenges with it that you need to be aware of.

Some of these challenges you have already heard about. The tool can hallucinate a number. It can introduce biases without you realising. It can expose the confidential data you feed it. All of these are real challenges worth guarding against, and all are reasonably well understood by now.

But there is one more that you probably have not thought about, and in my view, it is the most dangerous of the lot.

The Risk Nobody Warns You About

This risk is sycophancy. AI is just too nice! This is somewhat intentional because it is designed to be a helpful assistant. But this helpfulness can overstep and become flattery and ingratiation. When this happens, the AI reflects your own thinking, based on what you have told it, back to you with the polish and confidence of an independent answer.

The result? A budget that confirms what leadership already expected, which is not the same as an accurate budget. The really alarming part? You’ll be none the wiser about what is missing.

AI sycophancy works in two ways. The more obvious way is that it agrees with your assumptions and inputs, endorsing the conclusion you were already leaning towards. The second is more subtle because it starts to mirror your disposition. If you are naturally cautious, this will be reflected in how you frame your questions and the answers you get back. If you are more bullish, the opposite will happen. In both cases, the model adapts to you rather than to what is actually accurate or realistic. It feels like independent validation, but it is not.

What Sycophancy Does to a Planning Process

This looks harmless and feels reassuring. But when the tool you are increasingly relying on only ever props up your view and never challenges you, your outlook and options narrow. And this happens exactly when the world is becoming increasingly complex and uncertain. Your process is less questioning and you feel more confident, which is precisely the wrong attitude for anyone whose job is to plan for what might go wrong.

The Question I Had to Ask It

I will admit this almost caught me, too. I was well into a piece of work with an AI service, and it was being very encouraging about the direction I had taken. After a while, though, the encouragement started to feel too smooth, so I asked it straight out whether it was agreeing with me because I was right, or because I wanted it to.

That question changed the answer. Once I had insisted it be less agreeable, it pointed out weaknesses that the first version had smoothed over, and the work ended up in a better place. The AI was just doing what it was designed to do, and me not thinking critically about the process was the thing that nearly let me down. Fortunately, a small moment of friction, through a single honest question, was what caught it.

History Was the Foundation. Is It Still?

Along with AI sycophancy, there is also the challenge of pattern recognition. We often hear that machines are better at this than people. AI leans on historical patterns because that is what it has to work with. But think about the last 24 months. How many of them were actually typical? If the honest answer is very few, then we have to ask whether history can still tell us much about what happens next. That is not a small concern, because predicting from past trends is exactly what machine learning does. So it seems that at the very moment the past has stopped being a reliable guide, we are putting more faith than ever in a technology that has nothing else to go on.

The 2026 FP&A Trends Survey shows that finance teams are noticing this problem. Only 29% say they can plan confidently more than six months ahead, a drop of 10 percentage points in just one year. The horizon over which finance can plan is shrinking, even as the business asks it to see further ahead and we supposedly have the modern tools to do this.

Who Does the Challenging?

Which brings us to the question this series of articles is really about. In a good planning process, value comes from a mix of views. It includes the person who looks at the tidy, comforting projection and says, "That is not what I am seeing on the ground." The AI mirror will never be that person. It relies on past patterns, and unless prompted, has no appetite to challenge your assumptions. But increasingly, it is the only voice in the room when the budget gets built.

If AI will not challenge your assumptions, who will? That is where this article series goes next. Because AI used well does not make the accountant less important. It makes them more so because the more the machine does, the more it depends on the person who knows when to trust it. And when not to.

Before Your Next Planning Cycle

1. Tell your AI to give you the steel man argument for the exact opposite conclusions to the ones you have arrived at. Deliberately challenge your results before you accept them.

2. Track your plan’s key assumptions back to their sources. Be suspicious of anything without a wide range of inputs, and actively correct that before arriving at a conclusion you are happy to stand behind.

3. Reflect critically on your engagement with AI. When last did it change your mind or present counterarguments? If the answer is never, you are not consulting a tool; you are consulting a mirror.