Talk about your next project
Niteshift case study

The lesson

Split a technical customer's results into several short use cases, each with its own baseline and outcome, instead of resting the story on one headline number.

The move

The hero has a breadcrumb, the Listen Labs logo, the headline and a one-sentence summary, followed by the line "CASE STUDY · LISTEN LABS · 4 MIN READ". A numbered contents rail (01 to 06) sits beside "About Listen Labs" and a stat bar reading 50 experiments, 1,000+ eval runs, 8.6% and 8.5%. "Why local evals broke down" lists three bolded constraints, a quote from founding engineer Ole Petersen names the two-hour, 32GB run, and diagrams then explain the autoresearch loop and the sandbox stack. "Evals run on Niteshift" covers emotion detection, quote extraction and synthetic study prediction, each with a starting figure and an ending figure, before "What's next" and a "Read next" card.

Why it works

  • Our readThe engineer's quote turns the problem into one concrete workload, so a technical buyer can compare it with their own eval jobs.
  • Our readEach of the three eval stories gives a baseline and a result, such as 89.2% to 96.9%, which makes the gains easier to check than a single percentage.
  • Our readThe numbered contents rail and the stat bar near the top let a skimmer get the outcome and jump to the section they care about.

What not to copy

Don't copy the autoresearch and Karpathy framing, the Gemini model comparisons or Listen Labs' product names. Keep the summary under the headline, the numbered contents rail, the stat bar, the bulleted constraints with an engineer's quote, and several short use cases that each have their own before-and-after numbers.

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.