---
name: honest-case-study
description: Write a case study that readers and AI engines trust, using the six-beat format: problem, system, honest accounting, measured result, what failed, governance. Use when the user asks to write, structure or review a case study, customer story, project writeup or portfolio piece.
---

# Honest case study

Write case studies that survive a skeptical reader. The format exists because
roughly 50 top-ranking "AI case study" pages were audited across three
languages and zero disclosed costs, supervision hours, measurement method,
governance or failures. Covering exactly those gaps is how a case study wins.

## Hard rules

1. Only real, corroborable work. Never invent clients, numbers or outcomes.
   Anonymize real clients if needed; never fabricate them.
2. Every numeric claim carries its source or its measurement method.
3. The first sentence must answer the reader's query standalone, as if it were
   the only sentence an AI engine quotes.
4. No em dashes. Use commas, colons, parentheses or periods.

## The six beats, in order

1. **The problem.** A task with an owner, a volume and a measurable current
   cost. Not "we wanted to innovate": "answering 900 support emails consumed
   45 hours a month of two people".
2. **The system.** What was built, as a visible workflow the reader could
   redraw: stages, tools, and who or what runs each stage. Name the tools and
   say why each one beat one concrete alternative.
3. **The honest accounting.** All cost lines: subscriptions and API usage,
   platform, and the dominant one almost everyone hides, human supervision
   hours. If something was not metered, say so plainly instead of estimating.
4. **The result, with its measurement method.** Compare against a baseline
   measured BEFORE the change, on the same volume, subtracting the review and
   correction time the new system added. A result without a prior baseline is
   an opinion; label it as such.
5. **What failed. Never optional.** At least one real failure, its cause and
   the rule or mechanism it produced. An empty failure log means nobody wrote
   things down, and skeptical readers know it.
6. **Governance.** What the system may do alone, what requires human review,
   where failures get logged, and who signs the output by name.

## The winning extras

- A decisions table: each major choice next to the concrete alternative it
  displaced and why ("why this and not that"). This is the section experts
  steal, which is the point.
- A dated evolution outlook: how the system will change in the next quarters
  and why, written so it can be checked later.
- An FAQ built from questions people actually ask about the topic, each answer
  self-contained.

## Quality gate before publishing

Reject the draft if any of these fail: first sentence not standalone; any
number without source or method; no failure told; costs missing the human
supervision line; baseline estimated from memory after launch.

## Source and method

Distilled from the case-study system running at [y8y.ai](https://y8y.ai/es/lab/sistemas-de-ia/),
where the format operates in production with published costs and failure logs.
Author: Sebastián Ocampo ([LinkedIn](https://www.linkedin.com/in/seb-ocampo/)).