Talk about your next project
Humanloop case study

The lesson

Name the buyer's fear in a customer quote early, then show the evaluation setup that answers it before you give the result.

The move

The hero shows the FMG logo, the headline "How FMG solves LLM evaluation with Humanloop", a three-sentence summary that names the regulated industry, and a wide portrait photo. "The Problem" lists FMG's AI uses, such as support email replies and compliance checks, quotes Lucas Jans asking how he will know if a use of AI is working, explains why building an internal tool was rejected, and lists three requirements. "The Solution" describes evaluators and datasets, the fine-tuned smaller models that matched GPT-4 for a 15x cost saving, and a captioned radar chart. "The Result" claims hundreds of engineering hours saved and closes with a second Lucas Jans quote, followed by a carousel of other customer stories and a "Book a demo" block.

Why it works

  • Our readThe opening quote puts the reader's own question, whether the model is working, in a peer's words before any product claim.
  • Our readThe build-versus-buy paragraph and the three listed priorities give an evaluator the criteria FMG used, so they can check their own fit.
  • Our readOne concrete comparison, smaller fine-tuned models against GPT-4, gives the solution section a specific proof point instead of general praise.

What not to copy

Don't copy FMG's regulated financial-advisor framing, the GPT-4 comparison or the radar chart of evaluation metrics unless your product measures the same things. Keep the summary hero, the Problem, Solution and Result sections with a section-level headline each, the customer quote that states the fear, the build-versus-buy reasoning, and the related stories carousel before the demo call to action.

More from Humanloop

More case studies

Steal it for your brand.

Tell us what you want to make. We’ll talk through how to put these ideas to work for your brand.