This is the second in a series of blog posts about How to build with GenAI- From strategy to implementation. In this series, we explore the following questions:
- Is GenAI the right strategy for your product roadmap?
- How do you choose the right AI model?
- How do you navigate the complexity of data to deliver clear results?
- Should you take a Human-in-the-loop approach?
- How do you manage costs while developing with AI?
***
Once you’ve explored the Gen AI landscape and defined your goals, the next big decision is how to implement it. Should you build your own model, fine-tune an existing one, or rely on off-the-shelf APIs?
At Yotascale, we wrestled with the same question. “We started with an open mind,” recalls Jeff Harris, our Director of Strategy and Operations. “But pretty quickly, we realized building our own model didn’t align with our goals—or our users’ needs.”
Here’s how we approached this decision and what we learned along the way.
Understand Your Use Case and Capabilities
Every AI strategy starts with your specific use case. Are you solving a highly specialized problem that requires a custom solution, or are you enhancing user experience in a more general way?
For Yotascale, the goal was clear: improve how users navigate complex cloud cost data. We weren’t trying to push the boundaries of AI innovation—we were focused on delivering practical solutions.
As Jeff explains: “We didn’t want to reinvent the wheel. Instead, we leaned on existing models and APIs that allowed us to focus on delivering value quickly and efficiently.”
This approach helped us avoid unnecessary complexity while keeping our resources focused on what mattered most: the user experience.
Building Isn’t Always the Right Move
There’s a certain allure to building your own AI model. But as Jim Meyer, our VP of Engineering, puts it: “Even if you have the budget to train a model, the effort rarely pays off unless you’re getting considerably better results.”
Here’s why we decided against it:
- Cost: Building and training a model is resource-intensive, requiring high-quality training data and significant infrastructure.
- Maintenance: Models don’t maintain themselves. The ongoing costs of monitoring and retraining can quickly add up.
- Relevance: For our use case, existing APIs provided more than enough capability without the overhead of building from scratch.
This isn’t to say building is never the right choice—but it’s important to weigh the trade-offs carefully.
When Fine-Tuning Makes Sense
Fine-tuning can be a tempting middle ground. By training an existing model on your own data, you can tailor it to your specific needs.
However, even fine-tuning has its challenges. “It still requires high-quality training data,” Jeff notes, “and for the types of use cases we were exploring, we found that well-crafted prompts got us nearly the same results without the extra effort.”
That’s not to say we ruled it out entirely. For highly specific use cases, fine-tuning might make sense. But for now, focusing on dynamic prompts and API integrations has allowed us to move faster while keeping costs low.
Focus on Strategic Alignment
Ultimately, your choice should align with your broader strategy. For Yotascale, this meant enhancing the user experience rather than building AI infrastructure.
Jeff sums it up perfectly: “We didn’t want to plaster in AI just to say we had it. Instead, we focused on solving real pain points—like simplifying finOps and budgeting and forecasting workflows, and making cloud cost data more accessible.”
By keeping our strategy front and center, we were able to avoid shiny distractions and prioritize what truly mattered.
Takeaways for Making Your Decision
If you’re deciding whether to build, buy, or fine-tune an AI model, ask yourself:
- What’s the problem we’re solving, and how specialized is it?
- Do we have the resources and expertise to maintain a custom model?
- Can existing tools deliver the value our users need?
When in doubt, start small. As Jeff says: “You don’t need to solve every problem upfront. Start where you can add the most value and build from there.”
Looking Ahead
Choosing the right model is just one part of the puzzle. In our next post, we’ll explore how to use Gen AI to simplify complex workflows and enhance user experience—without overwhelming your team or your users.
