--- title: What Breaks a Machine Learning Pipeline Before It Deploys description: Five common failure points in ML pipelines are outlined: treating preprocessing as a one-time script (leading to training-serving skew), defaulting to accuracy... canonical: https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: What Breaks a Machine Learning Pipeline Before It Deploys | daily.dev og:description: Five common failure points in ML pipelines are outlined: treating preprocessing as a one-time script (leading to training-serving skew), defaulting to accuracy... og:url: https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk og:image: https://api.daily.dev/og/posts/vrQUOlGjK.png og:image:alt: What Breaks a Machine Learning Pipeline Before It Deploys og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # What Breaks a Machine Learning Pipeline Before It Deploys **[Medium](https://daily.dev/sources/medium_js)** · 6 min read · 0 upvotes · 1 comments ## Summary Five common failure points in ML pipelines are outlined: treating preprocessing as a one-time script (leading to training-serving skew), defaulting to accuracy as a metric without considering prediction costs, testing only one model instead of benchmarking multiple candidates, skipping actual deployment in favor of notebook demos, and failing to document decisions. Each mistake is paired with a concrete fix, such as using scikit-learn's ColumnTransformer for reusable preprocessing, selecting task-appropriate metrics like precision or R-squared, and wrapping models in a deployable web app. The post ends with interview questions to identify whether a practitioner avoids these pitfalls, followed by a Fiverr service pitch. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is training-serving skew and how do I prevent it in a machine learning pipeline? Training-serving skew is when a model performs well during testing but degrades in production because incoming data doesn't match the training data exactly, often because preprocessing was redone by memory instead of reused consistently. Preventing it means building preprocessing as a saved, reusable transformation, such as scikit-learn's ColumnTransformer, fit once and applied identically to every new batch of data. _daily.dev surfaces practical mlops guidance for teams trying to keep preprocessing consistent between training and production._ ### Should I use accuracy or precision to evaluate a classification model that predicts whether content will perform well? Precision is usually the better choice when a false positive is more costly than a false negative, such as predicting whether a recipe will drive high homepage traffic, where telling someone content will perform well when it won't is an expensive mistake. In that kind of pipeline, a target of at least 80% precision on the positive class is a reasonable benchmark, while accuracy treats every error as equally costly, which rarely reflects real business trade-offs. _developers weighing metric choices for a classifier can compare approaches through daily.dev before committing to one._ ### What R-squared value indicates a well-built regression model for predicting a numeric outcome like exam scores? An R-squared around 0.88 is achievable for a properly built regression pipeline predicting a numeric outcome such as exam scores, tested across as many as eight regression models before selecting the best performer. R-squared measures how much of the actual variation in outcomes the model explains, making it the appropriate metric for regression problems rather than accuracy or precision, which apply to classification. _anyone benchmarking regression models can track solid R-squared references on daily.dev before picking a final metric target._ ## Community discussion Top comments from developers on daily.dev. **@trevorsuna** · 0 upvotes > Training-serving skew is the quiet failure here. Reusing the exact preprocessing pipeline in both places, then testing the deployed path rather than only the notebook, catches a surprising number of problems before users do. ## Similar posts on daily.dev - [Why Most Machine Learning Projects Fail to Reach Production](https://daily.dev/posts/why-most-machine-learning-projects-fail-to-reach-production-opcx6gdce) · InfoQ · 0 upvotes · 0 comments - [Critical Mistakes Companies Make When Integrating AI/ML into Their Processes](https://daily.dev/posts/critical-mistakes-companies-make-when-integrating-ai-ml-into-their-processes-1oopopkx8) · Towards Data Science · 0 upvotes · 0 comments - [Prevent Data Leakage in ML Pipelines](https://daily.dev/posts/prevent-data-leakage-in-ml-pipelines-sx8xnnel4) · Daily Dose of Data Science \| Avi Chawla \| Substack · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#backend](https://daily.dev/tags/backend), [#mlops](https://daily.dev/tags/mlops), [#scikit](https://daily.dev/tags/scikit) [View this post on daily.dev](https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"What Breaks a Machine Learning Pipeline Before It Deploys","url":"https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk"},"datePublished":"2026-08-02T03:51:36.017Z","dateModified":"2026-09-13T21:32:30.164Z","description":"Five common failure points in ML pipelines are outlined: treating preprocessing as a one-time script (leading to training-serving skew), defaulting to accuracy...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4181a4a638fa1f1b5b8c1ac153f39e1e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4181a4a638fa1f1b5b8c1ac153f39e1e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"machine-learning,backend,mlops,scikit","timeRequired":"PT6M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"What Breaks a Machine Learning Pipeline Before It Deploys"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk","comment":[{"@type":"Comment","text":"Training-serving skew is the quiet failure here. Reusing the exact preprocessing pipeline in both places, then testing the deployed path rather than only the notebook, catches a surprising number of problems before users do.","datePublished":"2026-08-02T04:00:38.461Z","url":"https://daily.dev/posts/vrQUOlGjK#c-73uMCkKBU","author":{"@type":"Person","name":"Trevor Suna","url":"https://daily.dev/trevorsuna","image":"https://media.daily.dev/image/upload/s--dZ7gXxpp--/f_auto/v1784081551/avatars/avatar_EMoP47rpuw8DNjhp6R1b6?_a=BAMAMicg0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/what-breaks-a-machine-learning-pipeline-before-it-deploys-vrquolgjk#faq","mainEntity":[{"@type":"Question","name":"What is training-serving skew and how do I prevent it in a machine learning pipeline?","acceptedAnswer":{"@type":"Answer","text":"Training-serving skew is when a model performs well during testing but degrades in production because incoming data doesn't match the training data exactly, often because preprocessing was redone by memory instead of reused consistently. Preventing it means building preprocessing as a saved, reusable transformation, such as scikit-learn's ColumnTransformer, fit once and applied identically to every new batch of data. daily.dev surfaces practical mlops guidance for teams trying to keep preprocessing consistent between training and production."}},{"@type":"Question","name":"Should I use accuracy or precision to evaluate a classification model that predicts whether content will perform well?","acceptedAnswer":{"@type":"Answer","text":"Precision is usually the better choice when a false positive is more costly than a false negative, such as predicting whether a recipe will drive high homepage traffic, where telling someone content will perform well when it won't is an expensive mistake. In that kind of pipeline, a target of at least 80% precision on the positive class is a reasonable benchmark, while accuracy treats every error as equally costly, which rarely reflects real business trade-offs. developers weighing metric choices for a classifier can compare approaches through daily.dev before committing to one."}},{"@type":"Question","name":"What R-squared value indicates a well-built regression model for predicting a numeric outcome like exam scores?","acceptedAnswer":{"@type":"Answer","text":"An R-squared around 0.88 is achievable for a properly built regression pipeline predicting a numeric outcome such as exam scores, tested across as many as eight regression models before selecting the best performer. R-squared measures how much of the actual variation in outcomes the model explains, making it the appropriate metric for regression problems rather than accuracy or precision, which apply to classification. anyone benchmarking regression models can track solid R-squared references on daily.dev before picking a final metric target."}}]} ```