Smart Fix Assignment Service
Situation
Stitch Fix’s core product — personalized styling — depended on matching the right items to each customer. The assignment of which items went into a customer’s “Fix” was a critical business lever, but this particular area lacked integration between product engineering and data science.
The two platforms lived in different organizations and used different technology stacks. Data science had models. Engineering had the fulfillment pipeline. There was no established real-time integration point between them for Fix assignment.
These were early days at Stitch Fix. There were no product managers — engineers were trusted for their product taste, and the company specifically hired for that.
Decision
I researched the problem, wrote the spec, and implemented the Smart Fix Assignment Service end-to-end. There was no PM because PMs didn’t exist yet for this work. The problem was well understood from the engineering and data science side.
This was a cross-platform effort: bridging engineering and data science organizations that used different technology. The service provided real-time integration, allowing algorithmic recommendations to flow directly into the Fix assembly process.
The real power of the implementation was establishing a data science plugin point. Once the service existed, data science could run experiments and iterate on their algorithms with no or minimal need for engineering intervention. This reduced the friction between “data science has a better model” and “that model is running in production” to nearly zero.
Risk
Building a cross-organizational integration meant I was accountable for scope, prioritization, and stakeholder communication across both engineering and data science — in addition to implementation. If the service underperformed or the integration created operational problems, there was no shared ownership to distribute the risk.
I also took the risk of building a pattern — not just a service. If the plugin point architecture was wrong, it would have propagated bad practices across the organization.
Change
Keep rates and average order values reached historic highs. At Stitch Fix’s scale and customer base, these improvements translated to millions in incremental revenue. The metrics are fuzzy after a decade, but the directional impact was clear: the 2017 IPO was successful and the stock eventually rose dramatically.
More durably, the data science plugin point pattern became the template. Other teams used the same architecture to embed algorithmic intelligence into their product workflows — and because the plugin point minimized engineering friction, data science could iterate independently. The pattern outlasted the original service.
What This Demonstrates
- End-to-end ownership: Research, spec, implementation, and cross-organizational integration — without waiting for organizational support that didn’t exist yet.
- Plugin point architecture: Built an integration pattern that let data science iterate independently, not just a one-time feature.
- Business impact at scale: Chose a problem that directly affected core business metrics (keep rate, AOV) and contributed to IPO-era revenue growth.
- Cross-platform fluency: Comfortable operating at the boundary between product engineering and data science, across different organizations and technology stacks.