4 questions every finance leader should ask of their AI stack

Generative AI tools are extraordinarily powerful and helpful, but they’re not perfect. That means two things can be true at once:

• AI has incredible potential to help finance departments get more done faster
• Misused, AI can have the opposite effect

As a finance leader, it is essential that you be your company’s most informed AI buyer. Because while other departments can roll back campaigns or product releases, it is very difficult to roll back inaccurate financials and accounting. Especially when the books were certified and already used to make crucial company decisions.

To help you navigate this fast-changing world, I’ll share four questions you should ask every AI vendor before allowing them to access your sensitive financial data and workflows.

1. How do you ensure AI is accurate and consistent enough for financials?

Some AI finance partners are carefully guarding your data and verifying outputs. Others are simply putting your data into generic versions of LLMs without the necessary data protections and hoping for the best. If you use the latter type of vendor, you open yourself up to a lot more risk and potential problems.

Using AI models that aren’t fine-tuned to the needs of finance teams can result in what Harvard Business Review called workslop: “AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” For example, when AI invents numbers or rationale. This happened to Deloitte’s analysts in a report for the Australian government, leading to fines.

But these challenges shouldn’t stop you from finding the right applications of AI. Your fintech partners simply must restrict the AI to working from your actual, human-checked financial books. Run the results through multiple layers of self-correcting checks, and continuously train the model.

With a proper, thoughtful approach, AI can work autonomously in your back-office alongside your team to increase productivity.

“Your fintech partners simply must restrict the AI to working from your actual, human-checked financial books.”
—Sandeep Chopra, Co-CEO, Everest

2. Is AI essential to helping handle the task faster or better?

AI is a technology, not a feature. You must ensure your fintech vendors are applying it to places that are actually useful to your team. Not simply tacking it on to say they offer “AI-powered tools” so you can check a box with leadership that you’re adopting AI.
Any AI-powered feature you invite into your fintech stack should be smart enough to acknowledge when it is wrong or ask for more context when it can’t complete the task. And, failing that, escalate it to a human. Both Everest and Ramp are great examples of applying this approach correctly:

Using AI to code, then capturing that code

When an Everest user asks its AI bot for a new type of report, the Everest AI writes Python code to produce the document and stores it for future reference. That way, it will always perform as expected.

Ramp’s AI escalates select transactions to humans

After companies put their expense policies in Ramp, they can use AI to automatically suggest whether each expense should be approved, reviewed or rejected. Controllers have the option to let AI auto-approve certain low-risk expenses. This policy agent saves hours of review time, escalating expenses to a human only in the select cases where it’s necessary.

3. How do your tools de-risk AI use?

As agentic finance approaches, and AI agents autonomously make decisions, finance platforms should offer you guardrails. Tools should keep humans in the loop for critical or impactful decisions to prevent costly errors. Ramp’s AI tools, for example, are trained on your policies and controls. And Everest’s ERP offers a Live Sandbox™ where you can test changes on your actual data in total confidence before activating them. This allows you to see how agents handle your transactions and reports before releasing them to your production environment.

We believe that lack of trust is the top reason finance teams aren’t adopting more AI technology. But if you have tools to de-risk those AI-powered tools, it becomes easier to trust. It opens the door to autonomous finance, where finance managers and accountants evolve into financial system orchestrators and borrow more and more methodologies from developers.

4. How does your AI tool change my team’s roles?

This is a bit of a tricky question, but it reveals how deeply a fintech vendor actually understands your work and your most crucial responsibilities.

Any salesperson who struggles to come up with a reasonably specific answer is effectively saying “We don’t know.” Which is maybe not the product team you want to bet your finance org’s future on. This is certainly the case with ERP vendors who say, “We integrate with every model, do whatever you like,” instead of, “Here is a workflow that uses AI to automate revenue recognition in a way that’s ASC 606 compliant.”

A fintech vendor that is actually helping its customers innovate will be full of useful insights. They should be able to teach you something about your business, too.

Judge correctly, scale your finance org

The role of the CFO is changing rapidly. The duties listed in CFO job description increased 19% over the past five years, says Deloitte. We believe that’s a sign that in an environment of so much uncertainty, everyone else is looking to your team for certainty.

Are you able to provide it? Without adding headcount?

Every finance team should be figuring out their AI strategy and guardrails right now. It’s the only way to scale. But they can ease concerns and reduce risk by asking the right questions.