Stop Writing Blank Checks for “AI”: A Project Manager’s Guide to Real Budgeting

I see it every day one of my feeds: A CEO announces a “transformative AI initiative,” a top-tier consultancy hands over a seven-figure bill, and six months later, the project is stuck in “PoC Purgatory” with nothing to show but a soaring cloud bill. If you are a leader just getting started, you don’t need a $500/hour consultant to tell you that AI is expensive. You need to understand why it costs what it does, so you can challenge the estimates and build a sustainable roadmap. Here is the “No-Nonsense” framework for determining your AI project costs without getting fleeced.

1. The 80/20 Rule of AI Data

Consultants love to talk about “Neural Networks” “AI Agents” and “Large Language Models.” But in reality, 80% of your cost is usually data prep and engineering. Before you sign a contract, ask for a “Data Readiness Audit.” If your data is siloed, messy, or unlabelled, the model development cost is irrelevant because the model won’t work.

• The Trap: Buying a high-end model license before your data is clean.
• The Math: For every $1 spent on a “Data Scientist,” expect to spend $3 on “Data Engineers” and “Data Pipe-lining” in the first phase.

2. Infrastructure: The “Hidden Tax” of AI

Most companies budget for Training (building the AI) but forget about Inference (running the AI).

• Training Costs: These are one-time “burst” costs. Think of this like the cost of building a car.
• Inference Costs: These are recurring “per-use” costs. Think of this like the cost of gasoline. If your AI handles 10,000 customer queries a day, your cloud bill will scale with that volume.

Pro-Tip: Ask your vendor for an “Inference Scaling Estimate.” If they can’t tell you what the monthly bill will look like when you have 1,000 active users, they haven’t done their homework.

3. Build vs. Buy vs. Tune

You don’t always need to build a custom model from scratch. In 2026, the cost spectrum should include: API Integrations; Fine-Tuning (Existing models); Custom Model (Build from scratch) to name a few.

Recommendation: Start with an API. Prove the value for $10k before you spend $500k building a proprietary version of something that already exists.

4. The “Model Drift” Maintenance Fund

Unlike traditional software, AI decays. The world changes, and the model’s accuracy will drop (this is called “Drift”). A common mistake is treating AI like a “set it and forget it” purchase. You must budget for MLOps (Machine Learning Operations). This includes:

• Monitoring: Checking if the AI is still giving correct answers.
• Retraining: Updating the model with new data (typically 15-25% of the initial build cost per year).

The “Consultant Reality Check” Questions

Next time you’re presented with an AI proposal, ask these three questions to see if the numbers are padded:

• “What percentage of this budget is allocated to data cleaning vs. model architecture?” (If it’s less than 50% for data, they are being optimistic).
• “What is the estimated ‘Cost per Inference’ at scale?” (This forces them to look at your long-term cloud ROI).
• “What is the ‘Degradation Plan’?” (How much will it cost us in 12 months when the model starts losing accuracy?)

Final Thoughts

AI isn’t a product you buy; it’s a capability you build. Don’t let fancy terminology distract you from basic unit economics. If the “AI expert” can’t explain the costs in plain English, they likely don’t understand them either. Before you commit to an AI roadmap, you need a baseline. I invite you to spend 10 minutes with our Phase Zero diagnostic to surface your organization’s hidden blind spots.

Get your StAIR-Ready™ Assessment here →www.wheredoistartai.com

Next Article 3 In the Series: “Stay tuned for my next piece, where I’ll be diving into ‘Navigating AI Vendor Contracts: What Legal & Technical Teams Must Look For.'”

By George Fountain, Founder — Fountain Innovation Group (FIG) AI Readiness • Phase Zero Strategy • Responsible AI #AI #ProjectManagement #Budgeting #Innovation #ArtificialIntelligence #CostManagement #Strategy #DigitalTransformation

📖 Read on LinkedIn

Visit fig-io.com for more insights