I remember sitting in a boardroom in San Francisco back in 2019, pitching a machine learning initiative to a CFO who looked like he wanted to be literally anywhere else. He crossed his arms, looked at my projected budget, and asked a simple question. "What is the exact dollar amount this will return to the bottom line in the next six months?"
I hesitated. I talked about innovation. I talked about future-proofing. I talked about the potential of the technology.
He shut the project down right there.
That was a hard lesson, but it shaped how I approach business. Since then, I have built and sold two companies, RemoteTeam to Gusto and MovieLaLa to Gfycat. I have made over 200 angel investments, backing companies like Anthropic, OpenAI, Scale AI, and Hugging Face before they were household names. I wrote "Becoming Top 1%" because I wanted to share what actually works in Silicon Valley. And let me tell you what works when you are dealing with a skeptical CFO: hard numbers, clear timelines, and undeniable ROI.
Right now, the market for AI talent is completely insane. Recruiters are charging a fortune. Candidates with six months of PyTorch experience are making wild salary demands. If you walk into your CFO's office and ask for a massive budget to hire a team of prompt engineers and data scientists without a bulletproof ROI calculation, you will get laughed out of the room.
You do not need a massive budget or a big-name brand to find and hire top AI talent. You do not need to burn millions on research and development. You just need to know how to structure your AI initiatives so they print money instead of burning it.
The AI Talent Trap
Let us talk about the elephant in the room. The cost of building an AI team.
I see founders and enterprise leaders making the same mistake every single week. They read a few articles about large language models, panic that they are falling behind, and immediately try to hire a Head of AI. They get hit with a $400,000 salary demand, plus equity, plus a massive compute budget.
The CFO sees this and immediately hits the brakes. And honestly? The CFO is right.
You cannot justify that kind of spend if you do not have a clear path to revenue or massive cost reduction. When I was building RemoteTeam, we did not hire expensive specialists for every new feature. We found smart, hungry engineers who could learn quickly. The same applies to AI today. You can upskill your existing engineering team to use APIs from OpenAI or Anthropic. You can build incredibly powerful internal tools using off-the-shelf models.
Do not let recruiters convince you that you need a PhD in machine learning to build a customer service chatbot or automate your invoice processing. You need practical builders. Builders cost less, move faster, and actually ship products.
When I invested in Hugging Face, I saw firsthand how the open-source community was democratizing access to powerful models. You do not need to build everything from scratch anymore. The building blocks are already there, waiting for someone with business sense to snap them together. Your CFO does not care about the elegance of your neural network architecture. They care about whether the tool works and how much it costs to run.
A Pilot Is Not A Test
Here is where most AI initiatives die. The team pitches a "pilot project" to the CFO. They frame it as an experiment. A way to test the waters.
A successful pilot is not a test. It is the first step of a full implementation.
If you frame your AI project as an experiment, the CFO will treat it as a disposable expense. When budgets get tight, experiments get cut. You have to frame the pilot as phase one of a system that will fundamentally improve the unit economics of the business.
I have seen this play out across dozens of my portfolio companies. The ones that succeed do not run science experiments. They run targeted strikes on inefficiency.
I will give you the five criteria we use to ensure our AI pilots are set up for success from day one. If your project does not meet these five criteria, do not even bother pitching it to finance.
1. Tied to a Specific, Measurable Business Metric
"Improving efficiency" is not a metric. "Making the team faster" is not a metric.
Your pilot must target a specific number on the P&L. For example, reducing customer support ticket resolution time from 48 hours to 2 hours. Or cutting the cost of processing a vendor invoice from $15 to $2.
When you walk into the CFO's office, you need to say: "We currently spend $50,000 a month processing these documents manually. This pilot will cost $10,000 to build and $2,000 a month to run, saving us $38,000 a month."
That is a conversation a CFO wants to have. I remember a specific instance at RemoteTeam where we needed to streamline onboarding documents for international contractors. We did not pitch it as an "AI document parser." We pitched it as a way to cut onboarding time by 70%, which directly translated to faster time-to-value for our customers. The budget was approved in ten minutes.
2. Uses Existing Data
Do not pitch an AI pilot that requires a six-month data cleaning project before it can start.
Data cleaning is a black hole for time and money. If your AI initiative requires you to migrate three legacy databases and standardize ten years of messy records, the ROI timeline stretches out to infinity.
Find a problem where the data is already clean and accessible. Maybe it is your Zendesk support logs. Maybe it is your Stripe transaction data. Start where the friction is lowest. Prove the value there, and then use those savings to fund the harder data projects later.
When I look at companies like Scale AI, their entire business model is built around the fact that data preparation is incredibly hard and expensive. Do not take that burden on yourself for your very first pilot. Pick the low-hanging fruit.
3. Clear Path to Production in 30 Days
Speed is your best friend when proving ROI. If a pilot takes six months to build, the business context will change before you even launch.
I tell my founders to scope their AI pilots to 30 days. If you cannot get a working version into the hands of users in a month, the scope is too big. Cut features. Use a simpler model. Hardcode the edge cases. Just get it live.
Once it is live, it starts generating data. It starts saving time. It starts proving its own ROI. At MovieLaLa, we had to move incredibly fast to keep up with the entertainment industry. If a feature took more than a few weeks to ship, it was already obsolete. Apply that same urgency to your AI pilots.
4. Requires Minimal Specialized Maintenance
Your CFO is terrified of technical debt. They know that building the tool is only 20% of the cost. Maintaining it is the other 80%.
If your AI pilot requires a dedicated machine learning engineer to babysit the model, retrain it every week, and manage complex infrastructure, the ROI calculation falls apart.
Build your pilots using managed services. Use OpenAI's API. Use Anthropic's Claude. Let them handle the infrastructure, the scaling, and the model updates. Your team should focus entirely on the business logic and the user experience. This keeps your maintenance costs near zero, which makes the ROI calculation look fantastic.
5. Solves a Problem the CFO Actually Cares About
Engineers love to solve interesting technical problems. CFOs love to solve cash flow problems.
If you want budget, solve the CFO's problem.
Look at the areas where the company is bleeding cash. High customer churn. Expensive manual compliance checks. Inefficient supply chain routing. Point your AI pilot directly at the biggest pain point on the balance sheet. When you solve a problem that keeps the CFO awake at night, you will never have to beg for budget again.
The Actual ROI Formula
Let us get into the math. How do you actually calculate the ROI of an AI project?
It is simpler than most people think. You do not need complex probabilistic models. You just need basic arithmetic.
ROI = (Value Generated - Total Cost) / Total Cost
But you have to be brutally honest about the inputs.
Calculating Value Generated
Value comes in two forms: new revenue and cost savings.
New revenue is hard to attribute directly to an AI tool unless it is a customer-facing feature that you charge for. If you build an AI feature and charge an extra $10 per user per month, the math is easy.
Cost savings are where most internal AI projects shine. But you have to calculate it correctly.
Do not just say "it saves 10 hours a week." You have to convert that time into dollars. If the tool saves a $100,000 per year employee 10 hours a week, that is roughly $25,000 in saved time per year.
But here is the catch. The CFO will ask: "Are we actually reducing headcount, or are they just spending those 10 hours browsing Reddit?"
You have to show how that saved time translates to business value. "By saving the sales team 10 hours a week on data entry, they can make 50 more outbound calls per week, which historically converts to $50,000 in new pipeline."
Calculating Total Cost
This is where teams lie to themselves. They only calculate the API costs.
Your total cost must include:
- Development Time: The fully loaded cost of the engineers who built it.
- Compute and API Costs: The monthly cost of running the models.
- Maintenance: The estimated time spent fixing bugs and updating prompts.
- Training: The time spent teaching the team how to use the new tool.
When you present the true, fully loaded cost alongside a realistic, conservative estimate of the value generated, you build trust with the finance team. I have seen founders pitch AI tools with wildly optimistic cost projections, only to get crushed when the AWS bill arrives. Be conservative. If the ROI still looks good with conservative numbers, you have a winner.
Finding the Right Talent
I mentioned earlier that the market for AI talent is insane. But you still need people to build these tools. How do you do it without breaking the bank?
Stop looking for unicorns.
You do not need someone who can write a transformer architecture from scratch in C++. You need software engineers who understand how to chain prompts together, how to manage context windows, and how to evaluate model outputs.
I have seen incredible AI products built by self-taught developers who just spent a few weekends reading the OpenAI documentation and building side projects.
When I invest in early-stage AI startups, I do not just look at their academic credentials. I look at their GitHub. I look at what they have actually shipped. A developer who has built three practical, working AI tools using APIs is infinitely more valuable to your business than a researcher who has spent three years writing theoretical papers but has never deployed code to production.
Look inside your own company. Find the engineers who are already playing with these tools on the weekends. Give them the space and the mandate to build internal prototypes. You will be amazed at what they can create when you remove the red tape.
At RemoteTeam, some of our best features were built by engineers who simply had a deep curiosity and a willingness to read documentation. You can build that same environment in your company without spending millions on recruiting fees.
The Cost of Doing Nothing
When you are pitching your AI pilot, you have to address the alternative. What happens if we do nothing?
The CFO might think that waiting is the safe option. Let the technology mature. Let the prices come down. Let the competitors take the arrows in their backs.
That is a fatal mistake.
AI is not just another software update. It is a fundamental shift in how work gets done. The companies that figure out how to integrate AI into their workflows today are building a compounding advantage. They are learning how to manage the data. They are learning how to train their teams. They are learning where the models fail and where they succeed.
If you wait two years to start, you will not just be two years behind on the technology. You will be two years behind on the organizational muscle memory required to use it effectively.
Your competitors will have lower customer acquisition costs. They will have higher margins. They will be able to ship products faster.
When you present your ROI calculation, make sure you include the cost of inaction. Show the CFO what happens to your market share if a competitor achieves a 20% reduction in operating costs and passes those savings on to the customer.
Final Thoughts
Dealing with a skeptical CFO is not a barrier. It is a filter.
It forces you to think clearly about your business. It forces you to abandon vanity projects and focus on what actually matters.
I have built my career on finding the signal in the noise. Whether I am evaluating a seed-stage startup for an angel investment or deciding which features to build at RemoteTeam, I always come back to the same basic principles.
Solve real problems. Move fast. Prove the value.
If you can do that, you will never have a problem getting budget for your AI initiatives. You will not just get the budget. You will get a mandate to transform the company.
Stop treating AI like a magic trick. Start treating it like a business tool. Do the math, build the pilot, and show them the money.
Frequently Asked Questions
How long does it take to calculate ai roi for skeptical cfos.?
The timeline varies depending on your starting point and resources. For most founders, expect 2-4 weeks for initial setup and 2-3 months to see meaningful results. I've seen teams move faster when they focus on one thing at a time rather than trying to do everything at once.
What are the most common mistakes when calculating ai roi for skeptical cfos.?
The biggest mistake I see is overcomplicating things early on. Start with the simplest version that works, get real feedback, and iterate from there. Another common trap is copying what worked for someone else without understanding the context behind their decisions.
How do I measure success with this approach?
Pick one or two metrics that directly tie to your goal and track them weekly. Vanity metrics like page views or follower counts rarely matter. Focus on metrics that reflect real engagement or revenue impact.
What tools do I need to get started?
Start with the basics. You don't need expensive software or fancy tools. A spreadsheet, a note-taking app, and direct access to your customers will get you further than any enterprise platform. Add tools only when you hit a specific bottleneck.