AI Projects
Workflows I've built
There's a lot of AI noise right now. Everyone's adding it to their deck. I'm less interested in talking about it and more interested in shipping things with it.
These are five workflows I've actually built and used on real client work. Real problems, specific builds, real outcomes. I'm putting them here because I think the specifics matter more than the summary.
01
On-Brand Asset Generation
New product shots without a reshoot. The trick: AI never touches the product itself.
Most AI image tools take one photo and ask the model to imagine the rest. Fine for low-stakes products. On a $300 luxury item where logo, texture, and color all matter, the model guesses, and sometimes it guesses wrong.
Start from a 3D model
The brand has one or we commission it once. It pays for itself the first shoot it replaces.
Render the product exact
Logo, label, material, color: all come straight from the model, accurate down to the detail.
AI builds only the world around it
Background and lighting are generated to match the existing campaign. The product is never AI-generated.
New colorways, no reshoot
Aera had SKUs that didn't exist at the last shoot. Same lighting, same world, nothing that reads as out of place.
Accurate product photography at a fraction of the cost. Nothing that looks like it was generated.
02
The Content Multiplier
One shoot day becomes six weeks of assets. The extraction and packaging system finds what the footage contains, then formats it for every channel.
A full day of interviews with a coach, trainer, or educator generates four to six hours of raw conversation. Transcribed and without editing, most of it sits unused. The AI system finds what's packagable inside those hours, categories it by type, and auto-formats it for email, social, blogs, and video. No curation. No manual rewriting. Just systematic extraction.
Let the interview run
Longer takes. Tangents. The moments before and after rolling. Previously this was editor's nightmare. Now it's raw material. The system extracts usable content from all of it.
Transcribe and categorize
Feed the footage into Whisper for transcription, clean audio with enhancement tools, then search the transcript for distinct types: actionable advice, strong quotes, problem/solution pairs, substantive takes.
Package each moment into formats
A one-line quote becomes an email subject and social caption. A longer exchange becomes a blog paragraph or video script. The system applies templates; no rewriting needed.
Export the library
Hand off the full structured output. The team pulls, schedules, and publishes. No creators. No briefs. Just a stack of ready-to-use assets.
Real moments from one shoot day · Cindy Kramer, athletic trainer
To: Your list
Subject: What a 35-year coach takes every morning
Hi,
Cindy Kramer, coach for seven continents of triathletes and endurance athletes, shared something in a recent interview that stuck:
"I raised my level three points in two months, just by taking the Omega-3s."
She's measured it. That's not marketing. That's blood work.
Omega-3s are in our stack because the evidence is real and the effect is measurable. If you've been thinking about it, this is the signal.
Shop Omega-3
Extracted from one interview. Auto-formatted and ready to send.
One shoot day. Four hours of raw interviews. 26 distinct moments. 78+ total assets across email, social, blog, video. The system extracted what was already there.
03
Ad Copy Tuning Engine
The problem: copy decisions come down to whoever sounds most confident in the room. This scores every line instead.
Drop in a line of copy
A headline, a hook, a CTA. Anything you'd run as an ad.
Five checks score it 0–100
Hook, emotion, audience fit, brand voice, and CTA. Each gets a number and a reason.
Weak lines rewrite themselves
Anything under 90 gets rewritten automatically and re-scored until it clears.
Brand is a hard gate
The brand book is a non-negotiable. Off-voice copy is rejected no matter how well it scores elsewhere.
AERA · home fragrance
MOMENTOUS · performance
Illustrative demo. Copy and scores are sample outputs of the system, not live campaign lines.
04
Interview Mining for Paid Ads
Hours of old interviews, mined for the 30 seconds that work as an ad.
Brands sit on hours of long-form content that never reaches a paid channel: podcasts, founder interviews, testimonials. It exists. Nobody has time to watch six hours of tape looking for a 30-second ad.
Feed in a transcript
Any long-form footage: a podcast, an interview, a product deep-dive.
Find the ad-shaped moments
It searches for segments that match proven structures: problem/solution, before/after, testimony, transformation.
Mark the hook, value, and close
For each match it pulls the strongest opening, the core value beat, and the natural close, with timestamps.
Rank and hand off
Cuts are scored and ranked, then exported straight into the editor. The best concepts are already at the top of the stack.
Sample structured output · Aera perfumer interview
Ranked extract #1 [arc: problem/solution · hook score: 91] HOOK "Most people can't actually smell their own home." Timestamp: 00:14:32 VALUE Perfumer explains olfactory adaptation: why familiarity kills scent perception over time. Clip: 00:14:32 -> 00:16:48 (2m 16s) CTA "Aera resets your baseline. Your home should smell the way it did on day one." Timestamp: 00:16:51 -> Premiere XML ready. Rough cut: 2m 34s. Human pass: captions, B-roll.
Hours of unused long-form turned into ranked, ready-to-edit performance creative. Interviews sitting for a year became some of the strongest ads we ran.
05
Performance Data to Creative Decision
A Monday-morning brief that tells the creative team what to build, what to kill, and what's due, before anyone asks.
The loop between performance data and creative decisions usually breaks down. Reports go out, there's a standup, and by the time the brief updates the data is already two weeks old.
Pull the live numbers
Every Monday at 6am it grabs the last 14 days of Meta performance automatically.
Cross-reference plans and brand
It reads upcoming briefs from Notion and the brand book for messaging context.
Write the brief
What's working and why, what's fatiguing, what's launching this week, and the one creative priority.
Hand it straight to production
It can export Figma placeholders with the copy pre-filled, so there's no separate handoff.
Illustrative sample digest · numbers synthesized for demonstration
TOP PERFORMERS 1. "Without a word" (9:16) ROAS 4.2x CTR 3.8% Day 14 2. "Before / After" (4:5) ROAS 3.9x CTR 4.1% Day 9 3. "Science of scent" (1:1) ROAS 2.7x CTR 2.2% Day 21 (fatigue signal) HOOK PATTERN Question-as-challenge outperforming aspirational statements 3:1 this week. LAUNCHING THIS WEEK (from Notion) Colorway drop - June 19. Needs 3 static variants by Thursday. -> Priority: challenge hook on colorway drop. Pull "Science of scent."
The creative team knows what to build, what to kill, and what's due before the week starts.