The Mars Mission Principle: Why Pricing Agentic AI Systems Needs a Playbook
A Practical Guide to Budgeting for Autonomous Systems That Evolve
What’s the Quick Take?
Today I’m going to give you the low down on the mysteries of pricing Agentic AI systems, show you how to navigate the uncharted waters of AI economics, and get you some insights into calculating both capital and operational expenditures for your next-gen AI project.
This guide will equip IT leaders and manager with the tools to make informed decisions in the rapidly evolving landscape of artificial intelligence.
Boldly Going Where Nobody’s Gone Before
Imagine you're an IT leader tasked with pricing up a holiday to Mars. Sounds barmy, right? Well, that's precisely the conundrum many face when attempting to budget for a shiny new Agentic AI system.
It's uncharted territory, with no precedent and a myriad of unknowns.
In this post I’ll be your guide through the economic wilderness of AI, helping you navigate the treacherous terrain of CapEx, OpEx, and everything in between.
In the next few minutes, I'll demystify the process of pricing an Agentic system, providing you with a framework that's as adaptable as the AI itself.
The Agentic AI Pricing Paradox
Pricing an Agentic AI system is akin to estimating the cost of colonizing Mars. It's a venture into the unknown, fraught with variables and uncertainties.
Traditional pricing models fall short when confronted with the complexities of true artificial agency. Why? Because we're not just dealing with lines of code or processing power – we're talking about systems that can make decisions autonomously.
Breaking Down the Costs: The CapEx Conundrum
Capital expenditure (CapEx) for an Agentic system involves more than just servers and software licenses. Here's what you need to consider:
Infrastructure
This includes not just hardware, but the entire ecosystem needed to support your AI. Think high-performance computing clusters, storage systems, and networking equipment.
The main issue to consider with infrastructure is going to be your choice of Large Language Model (LLM) - Are you happy with an API based one like OpenAI, or does the governance and security policy of your organisation mean you need to deploy, train and refine your own model and have that running internally. These are big technology decisions with financial impact.
The CapEx for an Agentic AI system is like building the launchpad for your Mars mission – it's the foundation that everything else relies on.
Software Licensing
Agentic systems often require specialised software. for the tools that they’ll use to carry out their work. Underestimating this can be costly. You’ll need to know exactly what your Agentic system is expected to do in order to figure this out.
Initial Data Acquisition and Preparation
If you plan on training or fine tuning an LLM, this will involve purchasing datasets or dedicating resources to clean and structure existing data.
Integration Costs
Connecting your new AI system with existing enterprise software isn't a trivial task. While it’s easy to do with the right interfaces and Agentic tools being in place, this could change the use patterns of the underlying system which in turn will have vendors stood their with their hands out wanting more money.
The OpEx Odyssey: Navigating Ongoing Costs
Operational expenditure (OpEx) is where things get really interesting.
Unlike traditional software, Agentic AI systems have ongoing costs that can be harder to predict. Here’s my top 3.
Compute Resources
As your Agent AI grows in use, its computational needs may change. Cloud computing can help manage this, but costs can fluctuate based on usage. This is where making sure the right guardrails are in place for your Agents, making sure they strictly adhere to their job descriptions. Otherwise, they could run wild!
Human Oversight
While Agentic systems are autonomous, they still require human supervision. Factor in costs for domain experts to keep your AI on track and support the new system.
Software Licensing
The cost of underlying software isn’t just a capital expense, it’s an operational expense too.
If your capitalize your licenses, chances are you;’ll have a maintenance uplift on top that is due every so often. Don’t forget about this.
Where your licenses are OpEx only (which is becoming more and more popular, particularly for software Agentic tools will need access to) make sure you know what the metrics are that drive that price. once again, be aware of the need to bound the agency of an Agent - otherwise, if you a “per call” operational cost to an interface or end point, you could be faced with a very big bill.
The ROI Riddle: Measuring Success in Uncharted Territory
Determining the return on investment for an Agentic AI system is perhaps the trickiest part of the equation. Traditional metrics often fall short when dealing with systems that can fundamentally transform business processes.
Here's how I approach it.
Time Savings
Measure the reduction in time for tasks that your Agents handle. This is the first step towards understanding what level of cost reduction and cost avoidance your new system can deliver for your business.
Quality Improvements
It is common for Agentic solutions to be viewed as the solution to reduce error rates in data entry and processing. Make sure you have a baseline to work against if this is your goal.
New Capabilities
Quantifying the value of entirely new functionalities that weren't possible before is key to measurement success. While many new IT systems released into the enterprise are focused on efficiency and cost reduction, it’s not just e-commerce and mobile apps that driven revenue growth in 2024.
Agentic systems also have this potential, so if you’re planning to introduce Agents as part of a wider effort to increase margins, it’ll be important to build a Lifecycle Cost Model to show the money spent versus the money coming in.
Remember, the true value of an Agentic system often reveals itself over time as it learns and adapts to your specific business needs.
Factoring in the Unknown: The AI Uncertainty Principle
When budgeting for an Agentic AI system, it's crucial to account for the unknown. This isn't just about setting aside a contingency fund; it's about creating a flexible financial model that can adapt as your AI evolves.
Here’s 3 things to remember to keep in mind.
1. Scalability Costs
As your Agentic capability starts to show value, you might want to increase the scope of work or deploy new ones to support other processes. This could require investment in more powerful infrastructure or more skilled human resource.
2. Regulatory Compliance
The legal landscape around AI is rapidly changing. Budget for potential future compliance requirements for your Agentic system.
3. Ethical Considerations
As your AI becomes more advanced, you may need to invest in ethical audits and bias detection tools. Although as already mentioned, bounding the agency of your Agents with guardrails will significantly help you with this.
Key Takeaways
As I wrap up our journey through the economic landscape of Agentic AI, let's distill the crucial insights.
🚀 Treat AI Pricing Like Space Exploration
Just as we can't precisely budget a trip to Mars, pricing an Agentic AI system requires flexibility and a willingness to navigate the unknown.
💰 Balance CapEx and OpEx Carefully
Your initial investment sets the stage, but ongoing costs will likely be more significant and variable over time.
📊 Be Structured, But Prepared to Look Beyond Traditional ROI Metrics
The value of Agentic AI often manifests in unexpected ways. Be prepared to measure success using new, AI-specific metrics. Take a Lifecycle Costing approach so that you capture all the relevant dimensions.
🎯 Budget for the Unknown
Set aside resources for the unexpected. The true power of your AI system may reveal itself in ways you haven't anticipated.
If you're venturing into the world of Agentic AI or grappling with the economics of advanced AI systems, I'd love to hear about your experiences and challenges.
Drop a comment below or reach out directly - I read every response.
Until next time,
Chris