You think like a product manager and work like a data scientist. You have shipped ML-powered products to real customers and can point to the decisions you made, not just the ones made around you. Messy data, ambiguous model outputs, and stakeholders who don't yet see why the data matters are familiar territory.

You think in terms of compounding: how every data point collected today improves predictions tomorrow, and how that improvement builds a position competitors cannot close. Data insights is nShift's path from a tool company to an intelligence platform, and that scope motivates you rather than overwhelming you.

You are as comfortable in a Jupyter notebook as in a leadership meeting, and you know when each is needed.

The mission is to define what nShift builds with that data, validate it before engineering commits, and ship intelligence products that help shippers make better decisions. A shipper asking "how can I improve my delivery experience" should get an answer powered by the full network, not just their own history. Done well, that becomes a compounding intelligence layer: every data point collected today improves suggestions tomorrow, and the improvement creates a position no competitor can close.

The intelligence layer draws from two sources. Structured operational data, such as shipment telemetry and carrier performance, feeds statistical models and predictions. Qualitative data, such as customer support interactions, delivery exception messages, and community feedback, carries signal that numbers alone cannot surface. LLMs make that signal extractable and usable at scale. This role owns both.

The role sits at the intersection of product management and data science. You have the technical depth to build ML proof-of-concepts independently and the commercial judgment to translate those capabilities into products customers pay for. You write the strategy and the first notebook.

Define and own the data driven product strategy

• Own the roadmap for nShift's data intelligence layer: carrier performance prediction, cross-customer benchmarking, ML-driven carrier recommendations, and related capabilities.

• Translate nShift's data asset into a sequenced product strategy that delivers near-term value while building toward the compounding intelligence layer vision.

• Define the information architecture for the data insights layer: what data is collected, how it is normalized (in partnership with adjacent teams), and how it is exposed to upstream products.

• Make hard prioritization calls: what to build first to validate the ML hypothesis, what to defer, and what to kill when the data doesn't support the bet.

• Own the qualitative intelligence pipeline alongside the quantitative one: define which unstructured sources carry decision-relevant signal (support tickets, delivery exception messages, customer forum, carrier communications), how LLM-extracted insights are structured for downstream use, and how qualitative and quantitative signals combine into a coherent intelligence product.

• Own the positioning and differentiation narrative: why this dataset is uniquely nShift's, why no competitor can replicate it, and how to articulate that defensibility to enterprise buyers.

Build and validate before engineering commits

• Run ML proof-of-concepts independently: data exploration, feature engineering, model selection, and initial validation, using Python, notebooks, and the data available in nShift's data layer.

• Define the evaluation framework for both ML models and LLM pipelines. For ML features: what accuracy thresholds matter, how to measure model performance in production, how to distinguish signal from noise. For LLM-based features: how to assess output consistency, factual grounding, and quality at scale, where traditional accuracy metrics don't apply.

• Design lightweight experiments that answer the key questions before a full engineering team is committed. Does the prediction accuracy justify the product? Does cross-customer data improve the model significantly? Is the data we have sufficient, or do we need to instrument more?

Must-have

• Hands-on data proficiency across both paradigms: you can build and validate ML proof-of-concepts (data exploration, feature engineering, model training, evaluation) using Python and standard ML tooling (scikit-learn, pandas, XGBoost, or equivalent), and you can design and evaluate LLM-based analysis pipelines (prompt design, retrieval-augmented approaches, output quality evaluation). You are not a data scientist or an ML engineer, but you can work like one when the problem requires it.

• Shipped ML-powered products: you have personally driven an ML or data intelligence product from initial hypothesis to production, including defining evaluation criteria, partnering with data engineers and scientists, and making commercial decisions based on model performance data.

• PM fundamentals with commercial depth: 3+ years shipping B2B SaaS products. You can build a business case, define pricing, run customer discovery, and hold direct conversations with enterprise buyers about ROI.

• Data literacy that earns engineering respect: you understand data models, pipeline design, normalization challenges, and the practical constraints of real production data. You can review a data schema and immediately ask the right questions.