---
title: "AI Productivity Numbers in Case Studies Don't Transfer to Your Team"
description: "Stampli's widely-shared case study says Codex cut launch production from 243 hours to 77 — a real 68% reduction. That number describes Stampli's pre-AI…"
canonical: https://supergood.solutions/blog/ai-productivity-numbers-in-case-studies-don-t-transfer-to-your-team/
published: 2026-09-17
last-updated: 2026-10-05
updated: 2026-10-05
---

![AI Productivity Numbers in Case Studies Don't Transfer to Your Team — a fitted bespoke suit on a mannequin that clearly does not fit a much smaller toy robot standing next to it, green tape measure.](/blog/ai-productivity-numbers-in-case-studies-don-t-transfer-to-your-team/hero.webp)

BUSINESS

# AI Productivity Numbers in Case Studies Don't Transfer to Your Team

Published September 17, 2026 · Updated Oct 5, 2026 — 3 min read

**TL;DR:** Stampli's widely-shared case study says Codex cut launch production from 243 hours to 77 — a real 68% reduction. That number describes Stampli's pre-AI process, not yours, and buying a tool to chase someone else's delta is how teams overpay and underdeliver.

## Key Insight

Vendor case studies report a *ratio*, and ratios hide the denominator. In OpenAI's own write-up, Stampli says its go-to-market team modeled the Deep Finance launch at 243 active role-hours without AI tools, then delivered it in about 77 hours with Codex and ChatGPT Work, a 3.16x speedup, achieved while "design resources and outside contractors" were already committed elsewhere. That 243-hour baseline is the whole story. It's an estimate of how long Stampli's specific team, with its specific tooling gaps and its specific resourcing crunch, would have taken to hand-produce a seven-part blog series, launch emails, a webinar deck, paid creative, a PR release, and a hero animation. If your content team isn't running that lean, doesn't have that gap, or has already automated half of that workflow, the 68% doesn't apply to you. You're starting from a different 243.

## Why Teams Miss This

The mistake isn't believing the case study. The number is real and OpenAI's methodology is disclosed. The mistake is treating a single company's before/after delta as a transferable multiplier, then building a budget or a headcount decision on top of it. This is the same trap that shows up across enterprise AI adoption broadly. Research on generative AI pilots has repeatedly found that a large majority of enterprise GenAI initiatives fail to show measurable P&L impact, not because the models don't work, but because teams skip the step of measuring their own starting point before rolling out a tool. A pilot ROI number is also frequently measured in a clean, hand-picked environment (a forgiving use case, curated data) that doesn't resemble the messy production reality it gets compared against later. Without your own baseline, you can't tell whether a case study's multiplier is a floor, a ceiling, or irrelevant to your workflow.

## How to Actually Do It

1. **Baseline before you buy.** Pick the specific workflow you're evaluating AI for (content production, support triage, code review, whatever the vendor's case study matches) and time your team's current process for 2-3 real instances. Get actual hours-per-task, not a guess from a planning meeting.
2. **Match the shape of the case study to your shape.** Ask what made the vendor's baseline slow: understaffing, tool gaps, approval bottlenecks, manual handoffs? If your team's slowness comes from a different source (e.g., compliance review, not content drafting), a tool that speeds up drafting won't produce their delta.
3. **Run a scoped pilot on your own baseline.** Deploy the tool on the same workflow you just timed, with the same team, for a fixed window. Compare your before/after hours directly. Don't borrow the vendor's percentage.
4. **Price the decision off your delta, not theirs.** If your pilot shows a 20% reduction instead of 68%, that's still real information. Decide whether 20% clears your cost-benefit bar on its own, rather than anchoring to the number in the sales deck.
5. **Re-baseline periodically.** As your team adopts more tooling, your "manual" baseline keeps shrinking, so last year's pilot multiplier won't repeat next year. Measure again before the next purchase decision.

## What We've Learned

Before your next AI tool evaluation, timebox two hours to measure your team's current hours-per-task on the exact workflow in question. That single number is worth more than any vendor's published percentage, because it's the only variable in the ROI equation you actually control.

## Sources

- Case study with full methodology and hour figures: [Stampli cuts launch hours by 68% using ChatGPT Work — OpenAI](https://openai.com/index/stampli/)
- On pilot ROI not reflecting production reality: [How to Measure AI ROI (And Why Most Vendors Don't) — Accubits](https://accubits.com/how-to-measure-ai-roi-and-why-most-vendors-dont/)
- On AI ROI requiring operating-model change, not just tool adoption: [The Davos reality check on AI ROI — CIO](https://www.cio.com/article/4145044/the-davos-reality-check-on-ai-roi-why-tools-dont-pay-off-until-work-changes.html)

Have a specific workflow in mind?

The fit check comes first. It takes about 2 minutes. If it's a fit, Ryan follows up to schedule the paid Quick Scan ($1,000): a live working session where we'll tell you honestly whether it should be an agent, a workflow, or left alone, before you spend a dollar building it. You get 3 prioritized recommendations on the call, a one-page summary after, and the fee is credited toward a build if you sign within 14 days.

[Start the fit check →](/start/)

[See AI Agent Consulting](/ai-agent-consulting/)·[Start the fit check](/start/)

Get new posts + practical agent-ops notes

One email when something new goes up. No nurture sequence, no spam — unsubscribe whenever you want.

Thanks — you're on the list.
