If you are running a $2M–$20M company, every week brings another AI tool promising to transform your finance function. Most were built for companies three times your size, with IT teams you don't have and workflows that don't match yours. There is a better approach — and it starts with understanding that the real competitive advantage isn't which AI tool you buy. It's knowing your finance processes well enough to build exactly what your company needs, and keeping human judgment where it belongs.
The Boundary Between Preparation and Judgment
AI is exceptionally good at repetitive, structured tasks that consume time without requiring judgment: processing invoices, extracting data from documents, flagging anomalies against a known pattern, routing information to the right person with the right context already attached.
What it does not do well — at least not reliably — is exercise judgment. It cannot tell you whether to approve a $40,000 vendor payment when cash is tight and the relationship is new. It cannot weigh a strategic tradeoff. Those remain human responsibilities, and they should. AI belongs in the preparation layer of your finance function, not the decision layer.
The finance expertise is not in building the tool. It is in knowing what the tool should and should not do — where to place the control points, how to design exception handling, what human review looks like at each step, and which processes are ready for automation and which ones aren't. Get those decisions right, and AI makes your finance function faster and sharper. Get them wrong, and you have a workflow that runs smoothly right up until something goes wrong — and then nobody knows who is accountable.
One objection that comes up quickly — and fairly — is data security. Consumer AI tools are not the only option. Enterprise AI platforms, Microsoft Copilot, private APIs, and self-hosted deployments that run entirely within your own environment all allow companies to apply AI while maintaining appropriate security and governance. The right choice depends on your technology environment and your data sensitivity — but the options are there, and they are more accessible than most founders expect.
The Case For — and Against — Off-the-Shelf AI Tools
Packaged AI finance tools have real advantages. They are maintained by vendors, built with security and compliance in mind, and often integrate natively with major ERPs. For companies that want something working quickly with minimal internal effort, they can be the right call.
The limitation is that they are built for a generalized version of your company. Your chart of accounts, your approval workflows, your board reporting language — these reflect years of decisions specific to your situation. Fitting your operations into someone else's tool means accepting their assumptions about how finance should work. Sometimes that is fine. Often it produces friction, workarounds, and adoption problems that quietly erode the value you were promised.
Building your own lightweight AI-assisted workflows using tools like Claude or ChatGPT is more accessible than most founders expect. For simpler, well-defined processes, you may not need technical expertise — a clear description of the process and some well-designed prompts can get you surprisingly far. That said, as soon as workflows become multi-step, integrated with ERP data, or require permission controls, you will likely need someone with both finance process knowledge and technical comfort to design them properly.
Keep Your ERP Vanilla — Build Smart on Top
This is a position I hold firmly, and one I did not arrive at on my own. Early in my career, when I stepped into my second controller role at Optimum Talent, a mentor gave me a piece of advice I have never forgotten: run the software the way it was built to run. Do not bend it to fit your current habits. Your habits will change. The software's architecture will not.
When you implement NetSuite, Sage, Microsoft Dynamics, or any other ERP, resist the temptation to customize it. Customizations are expensive to build, fragile on upgrades, and create consultant dependency. They lock your finance function into decisions made when the company was a different size, with different priorities.
Run your ERP as designed. Then build your custom intelligence in the layer above it — the workflows, the approval processes, the reporting narratives, the exception handling. That is where customization is cheap, reversible, and does not require a consultant to undo. When your business evolves, you update a prompt, not a core module. You get a clean, upgradeable ERP and a tailored AI layer that handles the idiosyncratic needs of your business.
A Real Example: Accounts Payable
The traditional AP process at a $5M–$15M company: an invoice arrives, someone prints it, matches it to a PO or approval email, writes the GL code on it, and puts it in a pile. The approver reviews the pile — often without full context — signs, and passes it back. The invoice gets entered manually into the ERP. Paper gets filed.
It works. It is also slow, paper-heavy, and highly dependent on institutional knowledge that lives in people's heads rather than in a system.
What an AI-assisted AP workflow looks like instead
Invoices are captured digitally — no printing. An AI workflow reads the invoice, extracts the key fields, and builds an approval package with everything the approver needs: vendor name and payment history, the matching PO or prior approval, a GL code suggestion, and flags for anything unusual — an amount higher than typical, a new vendor, a duplicate invoice number. Critically, it also pulls context that your ERP stores in separate places but never assembles for you — the current cash position, the available budget on the relevant line — and presents it all in one place at the moment of approval. The ERP has the data. The AI connects it.
The approver receives a clean, complete package and approves online with a single action. Nothing is printed. Nothing is re-entered manually.
In a recent implementation I built for a client — an AI-assisted AP workflow that I later adapted into a reusable Excel tool when full AI access wasn't available on-site — the AP coordinator told me she was saving 50% of her processing time. That is not a benchmark from a vendor's marketing material. It is what one person measured in her own day, doing her actual work.
The employee's role does not disappear — it improves
Your AP clerk is no longer moving paper and re-entering data. They are verifying that the numbers are correct, confirming the vendor and amount match the original PO, and catching the exceptions the AI flagged for review. The approver still makes the approval decision with full context in hand.
AI prepares. Humans decide. That is not a limitation of the technology — it is the right way to run internal controls.
Your internal control framework depends on human review at key checkpoints. The goal is to make that review faster and better-informed, not to remove it. The accountability chain stays intact. The time spent on judgment goes up. The time spent on administration goes down.
A Second Example: Month-End Results Narrative
Every month, the numbers close. What does not close automatically is the story behind them — and that story is what your CEO and leadership team actually need to make decisions.
Why did gross margin drop two points? Why did one department overspend while another came in under budget? What does the variance against forecast actually mean for the quarter ahead? Traditionally, the finance director spends hours pulling this together manually: opening the ERP, pulling actuals, comparing against budget, drafting commentary line by line.
What changes with AI
An AI workflow trained on your chart of accounts, your budget structure, and prior months' commentary produces a first draft of the month-end narrative automatically once the books close. Variances are identified and flagged. Patterns across prior periods are surfaced. The finance director reviews the draft, corrects what the AI could not know — the sales rep who left mid-month, the one-time expense that will not repeat, the strategic context behind a number — and produces a final version in a fraction of the time.
The finance director or CFO always makes the final call. They own the narrative, the tone, and the conclusions. The AI handles the assembly. The human handles the judgment and the sign-off. The CEO gets a better-explained story, faster. The finance director spends their time on analysis and interpretation rather than building the document from scratch every month.
Where to Start
Pick one process — the most painful, repetitive task in your finance function. Map it in detail: every step, every handoff, every piece of information that moves, every point where a human needs to verify or decide. That mapping exercise often reveals bottlenecks and control gaps that aren't obvious day to day.
Then ask, step by step, where AI could handle the preparation work so your people can focus on verification and judgment. Build a simple version. Test it with real data. Refine it based on what your team actually encounters. The discipline is in the process design, not the technology.
The Architecture Decision That Matters
The companies that will build lasting advantage from AI in finance are not the ones with the largest budgets or the newest software. They will be the ones that understand their finance processes well enough to automate the right work — and keep human judgment exactly where it belongs.
Technology is becoming easier to acquire. Good financial process design remains a competitive advantage. Keep your core systems clean and standard. Build your intelligence in the layer above them. And make sure the people designing that intelligence understand finance first, and technology second.
That is the architecture decision most growing companies get wrong. And it is the one that compounds — quietly, in both directions — for years.