Intelligence Systems

Strategy

My AI adoption strategy

I don't force AI down the team's throat — that's unproductive and intimidating in a rapidly changing environment. I give people space to play with it first. That builds a real understanding of how it works and what it can do, and that understanding is what makes it possible to design for the people actually using it.

Execution

Three times we bottled lightning

Below are three examples of how we used AI as a strategic partner — and helped our users do the same.

01 · CloudSpeaker

Turning scattered feedback into signal

Leadership & Execution: Katie Sebkhi

Problem

Customer feedback was scattered across too many channels to reliably shape product decisions. When I dug into it, I found existing research tools that almost no one used — not because they weren't valuable, but because no one had time to go dig for them.

Solution

I joined forces with machine learning engineers to build CloudSpeaker: a platform that ingests feedback daily from support tickets, community, GitHub, Discord, Reddit, and more, classifies it with AI, and turns it into a searchable, on-demand feedback warehouse anyone at the company could query.

Impact

• Automated feedback classification, sentiment analysis, and summarization across every major channel.
• Turned months of manual research into minutes — for every team at Cloudflare, including marketing, product, design, engineering, and support.
• Built a content warehouse specific to Cloudflare that could power far more than one team's questions.

02 · CLUE

A shared bar for good content

Leadership: Katie Sebkhi  |  Design Execution: Alexa

Problem

Content quality across products was inconsistent, and fixing it meant pulling in the content team every time. There was no shared, scorable bar for what "good content" actually meant.

Solution

My team built CLUE: a tool that scores customer-facing content — UI copy, error messages, API docs — against our glossary, voice and tone, and UX writing best practices. It started as a weighted scoring model in a spreadsheet, then Workers AI got layered on top to catch what formulas couldn't: tone and context. It shipped as a full production tool any team could run themselves. This quickly evolved into how we approach design assessment and critique — acting as the trusted partner in the room before bringing frames to the wider team or engineering.

Impact

• Gave every team a shared bar for "good content" instead of a subjective one.
• Freed reviewers to focus on strategic feedback instead of manual line edits.
• Raised the average content score from 7.42 to a perfect 10 — a 34.8% lift.

03 · Cloudy Monitoring

Teaching the AI to trust better sources

Leadership: Katie Sebkhi  |  Design Execution: Content Team

Problem

Once Cloudy — our in-product AI guidance — shipped, the real work started: watching how it performed. Accuracy was slipping, and it turned out Cloudy was referencing outdated or incorrect source material. The more information it had access to, the less accurate it got.

Solution

My content team built a custom tracker to monitor Cloudy's prompts, sentiment, and accuracy over time, then went back into the source material itself — reformatting, tagging, and chunking content so Cloudy had less to sift through and more reason to trust what it found.

Impact

• Built a system to continuously monitor AI prompts, sentiment, and accuracy in production.
• Traced accuracy problems back to source material, not just model behavior.
• Improved response accuracy from 20% to 80%.

Want to talk through how this kind of thinking could apply to your product?

• Get in touch ↗