Eric Colson on machine learning
From Why 90% of Data Science Fails — And How to Fix It — With Eric Colson · · Delphina
“We reframed the recommendation problem at Stitch Fix as logistic regression and scaled scores between zero and one to represent purchase probability — that made the scores interpretable, learnable from data, and far easier to extend and scale than the ad‑hoc rules we had before.”
On , Eric Colson, Chief Executive Officer & Director at ARTISAN PARTNERS ASSET MGMT, spoke about machine learning during Why 90% of Data Science Fails — And How to Fix It — With Eric Colson on Delphina.
About Eric Colson
Eric Colson, CEO of Artisan Partners and a former data science executive at Netflix and Stitch Fix, has discussed the challenges companies face in leveraging data science effectively. In a February 2025 podcast, Colson argued that many firms treat data scientists as a support function, limiting their impact by only executing ideas from business teams. He advocated for giving data scientists autonomy and accountability for measurable outcomes, and for using trial-and-error experimentation with cheap failures to find winshol. Colson also emphasized the importance of decoupling algorithms from applications and enabling data scientists to frame problems rather than simply optimize within inherited constraints. In earlier appearances, Colson addressed the asset management industry, stating that "a lot of true active management got diluted" as firms prioritized growth over differentiation. He described Artisan's model as centered on investments, people, and trust, and noted the firm's introduction of "investment degrees of freedom" to allow teams to deviate from benchmarks. Colson also discussed value investing, saying Artisan seeks stocks that are out of favor and positions itself differently from the herd, and highlighted the firm's expansion into global and alternative strategies, including a China post-venture strategy.