Updated November 8, 2026: labeled the cost-saving figures as anecdotes, added sources and linked the FinOps Foundation and AWS tools.

Short answer: FinOps in a startup means making every engineer responsible for the cost of what they build. What worked at DockTech, a 12-person startup: cost estimates in design reviews, cost visibility for the whole team, cost checks in code review, and anomaly alerts. For AI workloads, add token-level monitoring, caching and context management.

I recently had the privilege of participating in the "FinOps más que costes" podcast to share my experience as CTO of DockTech implementing FinOps in a 12-person startup. The conversation made me reflect on how we've managed to make cost optimization not just a necessity, but a competitive advantage.

Why Does Every Line of Code Affect Your Cloud Bill?

In my 20+ years of technology experience, I've witnessed how the relationship between engineering and costs has evolved. Before, as programmers, we cared little about costs. We had a finance department that handled that. But in a startup, especially in a niche like maritime where you can't easily access large funding rounds, every engineering decision is literally a purchasing decision.

Working primarily with serverless and Lambda, I've learned that a poorly optimized line of code can turn into thousands of dollars per year. A synchronous loop that waits unnecessarily, excessive logs filling CloudWatch, or poor connection management can spike costs in unpredictable ways.

How Do You Build a FinOps Culture in a Startup?

The key isn't just implementing cost monitoring tools, but fundamentally changing how we think about development. In any early-stage startup, you must establish that:

Every programmer is responsible for the costs of what they design. When a developer presents the design of a new feature, they must include not only functional and non-functional requirements, but also a cost estimate using tools like the AWS Pricing Calculator.

This isn't easy. Indirect costs are especially complex to predict. In one early-stage company I worked with, they implemented stream processing with Flink, calculated direct costs well, but didn't anticipate that GuardDuty would analyze all those intensive operations, triggering unexpected costs.

Why Share Cloud Costs With the Whole Team?

One of our most important decisions was giving complete visibility to the entire team about our cloud spending. Every person in the company can see how much we pay for AWS, MongoDB Atlas, and other services. This transparency creates awareness and collective responsibility.

We also try to understand our "cloud efficiency rate" even before having significant revenue. We need to know what percentage of our future income will go to infrastructure costs to make informed business decisions.

How Do You Add Cost to Code Reviews?

We've integrated cost considerations into our code review process. When reviewing code, we don't just look for bugs or style issues, but also potential cost impacts. Is this synchronous call necessary? Are we logging too much information? Could this function be optimized to reduce execution time?

This level of scrutiny might seem excessive, but when you're working with limited runway, every optimization counts.

Could Static Analysis Catch Cost Problems?

I'm passionate about static code analysis for security, and I believe the same principles should apply to FinOps. Imagine having tools that could analyze your code before deployment and flag potential cost issues, just like we do for security vulnerabilities.

This could be part of the CI/CD pipeline – if your code doesn't pass the cost efficiency check, it doesn't get deployed. We're not there yet, but I see this as the future of FinOps tooling.

Does the Economic Cycle Matter for FinOps?

The economic environment when this was written (early 2025), with higher interest rates and tighter capital, has actually been beneficial for startups that embrace FinOps early. While it's more challenging to raise money, companies that build cost efficiency into their DNA are better positioned for the long term.

As I mentioned in the podcast, everything is cyclical. We know the easy money will return eventually. But companies that survive and thrive during the tough times will be the ones that seize opportunities when the market improves.

Is Good Design Also Cost-Efficient Design?

I believe there's a strong correlation between good design and cost efficiency. A system with harmonious architecture, with a certain beauty in its design, usually has a relationship with both product quality and cost optimization. When you optimize for costs thoughtfully, you often end up with better overall system design.

Key Takeaways for Startup CTOs

  1. Make cost responsibility part of every engineer's job description
  2. Implement complete cost transparency across the organization
  3. Include cost estimation in your design review process
  4. Set up anomaly alerts – they'll save you from unexpected spikes
  5. Think about unit economics from day one, even before significant revenue

The reality is that I'd prefer to have more capital and not worry about these details. But we don't have that luxury. This constraint has forced us to build better, more efficient systems, and ultimately, I believe it's made us a stronger company.

FinOps isn't just about cutting costs – it's about building sustainable, efficient technology that supports long-term business success. In a startup, that discipline can be the difference between survival and failure.

How Does FinOps Change for AI Workloads?

As AI becomes central to many startup offerings (see also the architecture of token pricing), we're facing an entirely new category of cost management challenges. Unlike traditional cloud resources that charge for compute time or storage, AI services operate on token economics – and this changes everything.

The Token Economics Revolution

Every interaction with an LLM costs money based on both input and output tokens. This means prompt engineering becomes cost engineering. A verbose prompt consuming 200 tokens versus an optimized 50-token version doesn't just improve response time – it cuts the input cost of that prompt by 75%.

In my experience, some startups have cut AI costs by around 60% through simple prompt optimization (an anecdote, not a benchmark): being specific about output length, removing unnecessary context, and crafting precise instructions. When you're paying per token, every word has a price tag.

Smart Caching Strategy

The most underutilized AI cost optimization is caching. Implement semantic caching for similar user queries, cache expensive embedding generations, and leverage provider-specific prompt caching for repeated system messages. How much a well-implemented caching layer saves depends on how repetitive your traffic is, but it can cut AI costs substantially while improving response times.

Context Window Management

Maintaining conversation history or large context windows can increase costs sharply, because every turn resends the whole history. Implement intelligent context pruning – keep only essential information for the task. Consider breaking complex operations into smaller, focused AI calls rather than one expensive, context-heavy request.

The key is treating AI costs with the same rigor we apply to traditional infrastructure. Monitor token usage by feature, track cost per user interaction, and always consider the model performance versus cost trade-off. Sometimes a smaller, faster model at a fraction of the cost delivers most of the value.

In the AI era, FinOps expertise isn't just about infrastructure anymore – it's about understanding the economics of intelligence itself.

Frequently Asked Questions

What is FinOps?

FinOps is a practice that brings engineering, finance and business teams together to manage cloud spend and make cost a shared responsibility. See the FinOps Foundation's introduction for the formal definition.

Who is responsible for cloud costs in a startup?

Every engineer. Each person who designs a feature should include a cost estimate in the design, and the whole team should be able to see what the company pays for cloud services.

How do you estimate the cost of a new feature?

Use a tool such as the AWS Pricing Calculator during design review, and look for indirect costs. For example, a stream-processing job can trigger extra charges from services that analyze its activity, such as GuardDuty.

How can you reduce LLM costs?

Tighten prompts, cache repeated work, prune the context you send, split large tasks into focused calls, and use a smaller model when it is good enough. Monitor token usage by feature and track cost per user interaction.

Further Reading