The CFO’s credibility gap

Most finance leaders say the right things about artificial intelligence (AI). Far fewer actually use it. Their teams notice — and that gap is quietly derailing transformation efforts across finance organizations — and nowhere more visibly than in technology companies, where the product itself is often built on the same AI capabilities the finance team is being asked to adopt. 

Finance and accounting professionals are, by training, expert observers of inconsistency. When a leader pushes AI adoption while visibly not adopting it themselves, the team hears: this is important enough to change how you work, but not how I work. That is not transformation — it is compliance theater, and it produces the half-hearted adoption rates that frustrate finance leaders six months into every initiative. In technology companies, the irony is particularly sharp: the CFO may be signing off on AI infrastructure spend while their own team still builds the board package in manually linked spreadsheets. 

The solution is not better communication or a more compelling change management deck. It is something harder and more personal: the CFO has to go first. 

The specific challenge in finance 

Finance carries structural tensions that make the leadership modeling problem especially acute. The professional identity of a controller, senior accountant, or FP&A analyst is tied directly to their domain expertise. For these professionals, AI does not simply automate a task — it introduces a perceived threat to the thing that makes them valuable. In SaaS and cloud businesses, where FP&A teams are often the institutional owners of ARR models, cohort analysis, and unit economics, this tension runs especially deep: if AI can draft an ARR bridge or model churn scenarios, teams begin to question where their edge lies. 

At the same time, finance has a legitimate reason for caution that other functions do not: outputs carry legal, regulatory, and fiduciary weight. A hallucination in a marketing email is embarrassing. A hallucination in an earnings release is a material event. In a pre-IPO technology company, a fabricated ARR or NRR figure shared with prospective investors can damage relationships that take quarters to repair — with no equivalent to an 8-K to correct the record. Finance professionals are right to apply scrutiny, and any leader who dismisses that concern as resistance is misreading their team. 

This combination — identity threat plus legitimate risk concern — creates a change environment that generic enterprise AI adoption frameworks are not built for. Technology companies face an additional layer: the engineering and product teams sitting down the hall are often early and confident AI adopters, making finance’s caution look like cultural lag rather than professional judgment. Effective transformation here requires both the modeling of confident personal AI use and governance structures that give teams permission to trust what AI produces. Neither works without the other. 

What modeling AI use actually looks like 

Leading by example on AI is frequently invoked and almost never operationalized. In practice, it means: 

• Use AI in your own analytical work, visibly. When reviewing a board package — ARR waterfall, net revenue retention bridge, or pipeline coverage analysis —, use an AI tool to stress-test the narrative against the numbers and share what it surfaced. When wrestling with a complex revenue recognition question – involving multi-element arrangements, contract modifications, or variable consideration, bring AI-assisted research into the conversation with your team — including your assessment of where it was useful and where it was wrong. 
• Narrate your reasoning, not just your conclusions. When AI saves two hours on an ARR bridge or cohort retention analysis, say so. When you catch a mistake in an AI output, explain what you caught and how. Making your process legible gives teams both permission and a model to follow. 
• Hold yourself to the same adoption standards you set for others. What specific AI capabilities are you personally developing this quarter? Leaders who can answer from direct experience — not from a vendor briefing — carry significantly more credibility when asking their teams to do the same. 

Change management for trained skeptics 

 The finance professional’s instinct for skepticism is a feature, not a bug — and in technology companies, where speed is cultural and “move fast” is the default operating posture, that instinct can feel like an unwelcome brake. The challenge is to redirect it toward evaluating AI outputs effectively rather than resisting adoption entirely. The message is to not stop being so careful. It is your expertise in identifying what is wrong that makes you the right person to govern how this tool is used. 

Practically, this means starting with low-stakes, high-time-cost work: first-pass ARR bridge commentary, summarizing customer contract terms or subscription schedules, churn and expansion variance analysis, formatting SaaS metrics packages for board or investor consumption. These use cases build fluency without exposing sensitive outputs during the learning curve. 

It also means investing in technical depth, not just tool familiarity. Finance professionals who understand where large language models fail — hallucination patterns, context limitations, sensitivity to question framing — create meaningfully more value than those who have only learned to write prompts. This is not computer science education; it is operational literacy. In a technology company, where colleagues in engineering already think fluently about model limitations and data integrity, a finance team that cannot engage at that level will find its credibility — and its seat at the table on AI decisions — quietly eroding. 

Governance as an operating standard, not a speed bump 

The dominant framing of AI governance in finance is risk mitigation — controls designed to slow things down enough to prevent bad outcomes. This framing is strategically costly. For technology companies operating in compressed planning cycles — quarterly board updates, annual recurring revenue reviews, and investor reporting that moves at the speed of fundraising — a governance posture built around brakes rather than guardrails is a competitive liability. Finance organizations that build governance as an operating standard — a definition of how AI work is done well — move faster and with greater confidence than those building it as a gate to prevent AI work from being done badly. 

Three components most governance frameworks underweight: 

• Auditability standards. Before deploying AI on any finance workflow, establish a clear answer to: if someone asks how this number was derived, can we show them? AI-assisted analysis is defensible when inputs, process and human review are documented — an extension of existing workpaper standards, not a new compliance burden. For technology companies, this is especially consequential for metrics like ARR, NRR, and CAC payback, which carry both internal governance weight and external investor scrutiny — and which are recalculated, not sourced directly from the general ledger. 
• Scope boundaries that are explicit and reviewed regularly. What is AI authorized to do, and what requires human origination? The absence of written scope boundaries does not prevent AI use — it ensures it happens inconsistently and without shared standards. In a SaaS business, a reasonable starting boundary: AI may draft commentary and flag anomalies, but all board-reported metric definitions, revenue recognition conclusions, and investor-facing figures require human origination and sign-off. 
• Tiered review protocols by output type. Teams need clarity on what a reviewed AI output looks like for an internal communication versus a board SaaS metrics package versus an investor data room update versus an SEC disclosure or 409A submission. Ambiguity here causes over-review that eliminates efficiency gains, or under-review that creates risk.