Marketing ops · Production

Stitch Fix uses expert-in-the-loop generative AI for ad headlines and product descriptions

The problem

Writing ad headlines and product descriptions at scale was time-consuming, costly, and produced inconsistent quality; Stitch Fix had hundreds of thousands of styles requiring descriptions that copywriters alone could not handle.

First attempt

Using few-shot learning alone for product descriptions produced generic, limited quality output that was insufficient for Stitch Fix's scale and brand standards.

Workflow diagram · grounded in source
1
Map outfit to style space
ai_action
“we map the outfit and a set of style keywords (such as effortless, classic, romantic, professional, boho etc.) to the latent style space, and then find the style keywords closest to the outfit in that space”
2
GPT-3 generates ad headlines
ai_action
“we use GPT-3 to generate headlines based on the selected style keywords”
3
Copywriter reviews and edits headlines
human_review
“Our human experts (copywriters) then review and edit the headlines generated by AI to ensure they capture the style of the outfit and align with the brand's tone and messaging”
4
Fine-tune model on expert examples
ai_action
“we gathered a task-specific dataset by having our human experts write several hundred high-quality product descriptions (the "completion", or training output) based on product attributes (the "prompt", or training input). We then fine-tu…”
5
Algo generates product descriptions
output
“This resulted in accurate and engaging descriptions tailored to meet the needs of our clients, all written in the Stitch Fix brand voice”
6
Expert feedback improves algorithm
feedback_loop
“This intelligence can help us further fine-tune the algorithm and improve the quality of the output through regular quality assurance checks”
Reported outcome

The expert-in-the-loop approach is now used for all Facebook and Instagram ad headlines, with copywriters reporting significant time and effort savings, and a blind evaluation showing algo-generated product descriptions achieved higher quality scores than human-written ones.

Reported metrics
Copywriter time and effortsaving significant time and effort
Product description quality scorehigher quality scores
Reported stack
GPT-3
Source
https://multithreaded.stitchfix.com/blog/2023/03/06/expert-in-the-loop-generative-ai-at-stitch-fix/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The expert-in-the-loop approach is now used for all Facebook and Instagram ad headlines, with copywriters reporting significant time and effort savings, and a blind evaluation showing algo-generated product descriptio…

What tools did this team use?

GPT-3.

What results were reported?

Copywriter time and effort: saving significant time and effort; Product description quality score: higher quality scores (source-reported, not independently verified).

What failed first in this deployment?

Using few-shot learning alone for product descriptions produced generic, limited quality output that was insufficient for Stitch Fix's scale and brand standards.

How is this marketing ops AI workflow structured?

Map outfit to style space → GPT-3 generates ad headlines → Copywriter reviews and edits headlines → Fine-tune model on expert examples → Algo generates product descriptions → Expert feedback improves algorithm.

Related marketing ops cases