Trust by Design: How AI Can Earn Its Place in Finance

• Financial professionals are eager for AI to automate tedious, error-prone tasks like data entry and reconciliation, but only on their terms.

• True trust in finance AI must be designed deliberately through robust controls, transparency, and a clear audit trail, rather than assumed.

• Organizations must approach AI adoption as a change-management priority, offering formal training to address employees' anxiety and build fluency.

• Finance teams display a spectrum of AI readiness across distinct postures—the Cautious Skeptic, Pragmatic Optimist, and Proactive Innovator.

Our latest research shows a profession that's already leaning in, eager to hand off the tedious, error-prone work that has consumed accountants' time for years. But "eager" doesn't mean "easy." The same people who want AI to move faster are also the ones drawing the hardest lines around how it gets there: on their terms, with control firmly in their hands, and a clear line of sight into how it arrived at an answer. The opportunity is real, but so is the bar for delivering the control, transparency, and trust that finance requires.

We surveyed over 100 financial professionals — a group weighted heavily toward large, complex organizations, with over 80% coming from Enterprise and Mega-Enterprise companies spanning manufacturing, financial services, healthcare, and beyond.

What came back wasn't a story of resistance to AI. It was something more interesting: a profession standing at a genuine inflection point, pulled in two directions at once. People want efficiency; however, they're wary of the black box that delivers it. Closing that gap means earning trust deliberately — through what gets built, how it's sequenced, and what controls people are given along the way. That's the lens for everything that follows, and the implications reach well beyond any one company: trust in AI is a shared, industry-wide project, not something any single vendor hands down.

Before we get to what finance professionals want from AI, it's worth pausing on where they're starting from. "AI adoption" isn't a single starting line; it's a set of habits, and most of finance hasn't formed them yet.

A companion study we ran back in February with 40 customers looked at how AI actually shows up in someone's day, not how they feel about it in the abstract. For a lot of people, that's not at all: 65% of respondents weren't using AI in their workflow, not by choice so much as by policy, lack of a standard company approach, or simply sitting in an early evaluation phase with no broad rollout yet. Where usage does exist, it tends to live in two very different places. There's the occasional, reach-for-it-when-stuck habit — pulling up ChatGPT, Gemini, or Copilot to draft an email, summarize a document, or debug a formula. And there's a smaller, steadier habit of AI folded into the actual workday, often through tools like Microsoft Copilot in Excel or AI already built into core finance software.

Two things stand out in that pattern. First, even where the habit exists, it remains almost entirely assistive rather than autonomous — "help me write this," "check my formula," and not "handle this for me." Among users who've adopted AI at all, 35% are using it for assistive tasks like drafting emails — the ceiling on current habits is help, not delegation. Second, there's a real gap between the general-purpose tools people reach for on their own and the workflows they actually spend their day in. Roughly 89% named only general-purpose LLMs for performing non-finance activities. The same people fluent in ChatGPT are not yet fluent in AI that lives inside reconciliation, matching, or close. That gap between borrowed familiarity and embedded trust is exactly the "last mile" we get to close — and it's worth remembering, in everything that follows, that a meaningful share of the market is starting this conversation from close to zero.

The most honest tension in this research isn't about technology at all — it's about identity. For a profession that has spent decades finding security in manually tying out every number, AI doesn't just remove a task ... it removes a comfort zone. Left unaddressed, that shift shows up as quiet resistance: shadow IT, reversion to spreadsheets, and even the occasional deliberate hunt for an AI error just to justify going back to the old way.

The organizations navigating this well aren't the ones with the most sophisticated models — they're the ones treating the transition as a change-management problem first. Nearly two-thirds of respondents (from our survey of over 100 professionals) pointed to formal, instructor-led training as their preferred way to build fluency, well ahead of self-guided learning.

AI won't replace the professionals who learn to work alongside it, but those who never get the support to make that shift may find themselves competing against colleagues who did.

Ask people where AI will change their work most, and the answer isn't glamorous. It's not forecasting, fraud detection, or strategic advisory. It's data entry and reconciliation — cited by over 75% of respondents as the area facing the most disruption over the next five years.

There's something almost reassuring in that. The market isn't asking AI to replace judgment. It's asking AI to finally take the tedious, repetitive, error-prone work off professionals' plates so humans can focus on the exceptions that actually require it. Respondents cited financial reporting and analysis as a secondary priority, placing fraud detection and compliance monitoring further behind — areas viewed as a second horizon rather than a starting point.

If there's a lesson in the ordering, it's this: The fastest path to trust isn't a flashy predictive model. It's nailing the boring stuff first, visibly and reliably, so people have a reason to extend that trust further.

If there's one thread that runs under everything in this research, it's this: We must design trust in AI, rather than assume it. People aren't asking to be convinced that AI is safe in the abstract — they're asking to retain the choice of when and how it touches their work. Humans need to do the "final approval," "actual tying out," or "validation." Not because the AI is necessarily wrong, but because in an environment with zero tolerance for financial error, the cost of being wrong once outweighs the convenience of being right most of the time.

This is also, notably, not a call for AI to disappear behind the scenes. It's a call for visibility — explainability, audit trails, and a legible logic path from input to output. The underlying commitment we must uphold: People, not the model, hold the final word.

You likely have a team member who has expressed this concern: I don't want my job "taken away by things they do today happening automatically." That's not reflexive fear of new technology; it's a rational response from someone whose entire professional identity is built on getting numbers exactly right. For this person, the conversation starts and ends with risk. Data privacy and security top the list of concerns, and just as often, the issue is simpler: this colleague does not yet trust what the model gives back. Call them the Cautious Skeptic.

Then there is the colleague who is not opposed to automation at all and is just waiting for the math to work. Their objections are practical, not philosophical: cost and the difficulty of proving ROI. Show this person real time saved on a real task, and their guard comes down fast. Some of these colleagues are already comfortable with AI features being on by default — proof that trust here isn't a fixed trait; it's earned incrementally, through demonstrated value. Call them the Pragmatic Optimist.

Finally, the proactive innovator is already a few steps ahead, quietly building solutions in-house or serving on a formal AI Ethics or Governance Board, which a growing share of finance organizations have already established. This person has moved past "should we use AI" entirely and is asking how to govern it responsibly at scale, and how to shift from doing the work to overseeing it. Call them the Proactive Innovator.

Here's why that matters beyond any one survey: "AI readiness" isn't a single dial; it's a spectrum. Most finance organizations already contain all three of these postures, often on the same team, sometimes on the same task. Giving people real control over their own exposure to AI is what addresses market segmentation — the same feature, different defaults, different pacing, and different options for different people.

Pull back far enough, and the picture that emerges isn't one of an industry resisting change. It's one that's ready to move, but only on its own terms — terms that center control, transparency, and evidence over promises.

Control isn't a constraint on the vision; it's the foundation of it. We must design trust into our solutions rather than bolting it on, earning it through mundane tasks before spending it on ambitious initiatives. A brilliant model nobody switches on doesn't move the adoption needle.

The organizations that internalize this won't just adopt AI faster; they'll help define what responsible, durable AI adoption in finance actually looks like.