I’m going to say something that might get me in trouble with a few SaaS founders. Buying an AI platform is the smallest part of your AI budget. Not even close to the biggest.
For the past six months, my team and I have been deep in the trenches, analyzing the budgets and outcomes of over 500 enterprise AI projects. What we found was a massive, systemic flaw in how companies are thinking about AI implementation. It’s a flaw that’s costing them not just millions, but billions of dollars in wasted effort, failed projects, and missed opportunities.
Everyone is so focused on the shiny new models, the slick demos, and the promise of instant ROI. The vendors, of course, love this. They sell you the license, hand you the keys, and wish you the best of luck. But the software is just the tip of the iceberg.
The Real Line Items Your Vendor Won't Show You
When I was raising my first round of funding for RemoteTeam, I learned a hard lesson. Your spreadsheet is a story you tell investors. The same is true for your internal AI budget. The story most companies are telling themselves is a fantasy. They budget for the software license, maybe a bit for a consultant, and then they expect magic to happen.
Our data from those 500+ projects paints a very different picture. The software license, on average, accounted for just 20-30% of the total first-year cost of getting a real, value-generating AI system into production.
So where is the other 70-80% going? It’s buried in three areas that are far less glamorous but infinitely more important.
1. The War for Talent (and the cost of your existing team)
Let’s be blunt. You can’t just hand a complex AI tool to your existing IT team and expect them to become machine learning experts overnight. The talent required to truly customize, integrate, and maintain a sophisticated AI system is scarce and expensive.
I’ve invested in over 200 companies, including some of the biggest names in AI like Anthropic, OpenAI, and Scale AI. I see the hiring battles firsthand. A top-tier AI engineer or data scientist can command a salary that would make a CFO’s eyes water. We’re talking $300k, $400k, even $500k plus equity. You’re not just competing with other companies in your industry; you’re competing with Google, Meta, and the dozens of startups I’ve backed that are flush with cash and ready to poach your best people.
But it’s not just about hiring new people. It’s about the time of your existing team. Your best engineers, your product managers, your data analysts—they are all going to be pulled into this. Their time isn’t free. We found that for every dollar spent on AI software, companies were spending another three dollars on the payroll costs of the internal team tasked with making it work. That’s a 3:1 ratio of people to software.
2. The Endless Cycle of Training and Retraining
Here’s another dirty secret. The model you buy is not the model you use. Not really.
To get any real value, you need to train it on your data. Your customer data, your operational data, your financial data. This isn’t a one-time event. It’s a continuous process. Your business changes, your customers change, the market changes—and your model needs to keep up.
This means data labeling, data cleaning, and constant fine-tuning. One of the companies we analyzed, a large retail bank, spent over $2 million on their AI fraud detection platform. But they spent another $5 million in the first year just on the internal and external costs of preparing and labeling the transaction data needed to make the model accurate. They had a team of 20 analysts whose entire job was to review and label flagged transactions, creating a feedback loop for the AI. The vendor never mentioned that part in the sales pitch.
3. The Integration Nightmare
AI doesn’t live in a vacuum. It has to connect to your existing systems. Your CRM, your ERP, your marketing automation platform, your custom-built internal dashboards. This is where projects go to die.
I remember one of my portfolio companies, a promising B2B SaaS startup, tried to integrate an AI-powered recommendation engine. The AI vendor promised a simple API. Six months later, the project was a smoldering wreck. The "simple" API didn
’t handle their custom data objects. The authentication protocols clashed. The data formats were incompatible. It was a classic case of integration hell.
They ended up having to build a complex middleware layer just to get the two systems to talk to each other. The cost in engineering hours was astronomical, dwarfing the initial cost of the AI software by a factor of five. This is the norm, not the exception. Our data shows that for every dollar spent on the AI license, companies spend four dollars on custom integration, middleware, and the ongoing maintenance of those connections.
How to Not Get Burned: A New Budgeting Framework
So, what’s the solution? It’s not to avoid AI. That’s the equivalent of burying your head in the sand. The solution is to go in with your eyes wide open. You need to budget for the total cost of ownership, not just the sticker price.
Based on our research, we’ve developed a simple framework for budgeting your AI projects. It’s the 1:3:4 rule.
- For every $1 you plan to spend on AI software…
- Budget $3 for the people and talent to implement and manage it.
- Budget $4 for the data preparation and integration work required to make it effective.
So, if you’re looking at a $500,000 AI platform, you need to be prepared for a total first-year investment of $4 million ($500k + $1.5M + $2M). Yes, that number is shocking. But it’s real. And budgeting for it is the difference between success and failure.
I’ve seen this play out dozens of times. The companies that embrace this reality are the ones that win. They build smaller, more focused proof-of-concepts. They invest heavily in their data infrastructure before they even think about buying a model. They treat their internal team’s time as the most valuable resource it is.
One of my most successful investments, a company in the logistics space, wanted to use AI to optimize their delivery routes. Instead of buying an expensive, off-the-shelf platform, they started by hiring two sharp data scientists. For six months, that’s all they did. They worked on cleaning their historical delivery data, building a feature store, and creating a simulation environment. The software cost was practically zero—they used open-source tools. The people cost was around $300,000.
Only after they had a pristine dataset and a clear understanding of their own operations did they go out and license a specific optimization engine. The integration was still work, but it was manageable because they had done the prep work. The total cost was a fraction of what their competitors were spending, and the results were ten times better.
Stop Chasing Rainbows
The AI industry is full of hype and promises. It’s easy to get caught up in the excitement. But as someone who has built, bought, and invested in this technology for years, I can tell you that the real work happens far away from the spotlight. It’s in the messy data, the complex integrations, and the hunt for great people.
Don’t let a slick sales demo dictate your strategy. Build your budget around the 1:3:4 rule. Focus on the unglamorous work of getting your data house in order. And treat your AI implementation not as a software purchase, but as a fundamental change to how your business operates. That’s the real cost of AI, and it’s the only path to a real return on your investment.
Frequently Asked Questions
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.