Job description

Overview The Frontier Transformation Framework helps customers become Frontier Firms: organizations where AI and agents operate as part of the business. TokenOps makes the economics of those systems measurable and governable. Within the Frontier Transformation team, this role provides the quantitative planning capability that turns telemetry and uncertainty into defensible investment and operating decisions. We are seeking an Applied Scientist, AI Economics (TokenOps, FinOps) to own the proactive planner, forecasting, optimization, and calibration that make AI economics defensible. You will develop predictive models for cost and delivery distributions, design optimization methods that account for risk and constraints, and calibrate those models against production evidence. This role is for a scientist who can connect rigorous methods with practical customer decisions. You will move between Python implementation, probabilistic forecasting, stochastic optimization, causal inference, and executive explanation; partner with the Telemetry & TokenOps Engineer on the evidence foundation; and turn the work into a reusable forecast and calibration assessment. Responsibilities Core Responsibilities: • Own proactive-planner algorithms and predictive-model calibration for TokenOps and FinOps decisions. • Develop probabilistic models that forecast cost, token demand, latency, quality, and delivery outcomes as distributions rather than point estimates. • Design stochastic optimization methods that recommend model, agent, routing, and capacity choices under uncertainty. • Apply CVaR, chance constraints, and related risk measures to keep recommendations within customer cost, reliability, and delivery tolerances. • Build conformal and empirical calibration methods that quantify uncertainty and show when predictive confidence is no longer reliable. • Use causal inference and experimental evidence to distinguish the impact of TokenOps interventions from correlation or external effects. • Develop Monte Carlo simulations and scenario analyses that make tradeoffs, tail risks, and sensitivity visible to decision makers. • Implement production-quality scientific software in Python and establish monitoring, validation, versioning, and MLOps practices for deployed models. • Partner with the Telemetry & TokenOps Engineer to define the fact-store data, attribution, quality, and lineage needed for forecasting and calibration. • Translate model outputs into clear customer recommendations, assumptions, constraints, and decision thresholds. • Create a reusable forecast and calibration assessment that can be applied consistently across customer engagements. Success in this role looks like: • Customer decisions are supported by calibrated cost and delivery distributions with explicit assumptions and uncertainty. • Planner recommendations improve expected outcomes while respecting customer risk tolerances, budgets, and operational constraints. • Predictive performance and calibration are monitored in production, with clear triggers for investigation, retraining, or model retirement. • Causal and experimental evidence makes the economic impact of TokenOps interventions defensible to technical and business stakeholders. • Forecasting and calibration methods become reusable assets that the delivery team can apply and explain consistently. Qualifications Required Qualifications: • Master's Degree in Computer Science, Engineering, Data Science or related field AND 4+ years experience applying machine learning, forecasting, or optimization in production • OR Bachelor's Degree in Computer Science, Engineering, Data Science or related field AND 6+ years experience applying machine learning, forecasting, or optimization in production • OR equivalent experience. Preferred Qualifications: • Demonstrated ownership of calibrated decision models operating under uncertainty and tied to real production decisions. • Strong Python and scientific machine-learning skills, including disciplined implementation, testing, and reproducibility. • Deep experience with probabilistic forecasting and evaluation of predictive distributions. • Experience formulating and solving stochastic optimization problems with operational or economic constraints. • Working knowledge of CVaR, chance constraints, and other methods for modeling tail risk and decision confidence. • Experience with conformal prediction, empirical calibration, or comparable uncertainty-quantification techniques. • Strong grounding in causal inference and the design or analysis of experiments and observational studies. • Experience deploying and monitoring predictive models through production MLOps practices. • Published or production work in AI economics, FinOps, TokenOps, or decision science. • Experience building Monte Carlo simulations, scenario models, and sensitivity analyses for executive or operational decisions. • Understanding of model, agent, token, infrastructure, and delivery cost drivers in enterprise AI systems. • Experience working with telemetry, attribution, and fact-store data in partnership with data or platform engineers. • Familiarity with model evaluation, drift detection, data-quality controls, and lineage for regulated or high-stakes decisions. • Customer-facing experience translating quantitative results into clear recommendations, tradeoffs, and limitations. • Ability to collaborate with solution architects, engineers, finance stakeholders, and delivery leaders. • A record of turning bespoke analytical work into reusable methods, software, assessments, or intellectual property. Solution Architecture IC4 - The typical base pay range for this role across the U.S. is USD $101,800 - $193,800 per year. There is a different range applicable to specific work locations, within the San Francisco Bay area and New York City metropolitan area, and the base pay range for this role in those locations is USD $126,100 - $207,600 per year. Certain roles may be eligible for benefits and other compensation. Find additional benefits and pay information here: careers.microsoft.com This position will be open for a minimum of 5 days, with applications accepted on an ongoing basis until the position is filled. Microsoft is an equal opportunity employer. All qualified applicants will receive consideration for employment without regard to age, ancestry, citizenship, color, family or medical care leave, gender identity or expression, genetic information, immigration status, marital status, medical condition, national origin, physical or mental disability, political affiliation, protected veteran or military status, race, ethnicity, religion, sex (including pregnancy), sexual orientation, or any other characteristic protected by applicable local laws, regulations and ordinances. If you need assistance with religious accommodations and/or a reasonable accommodation due to a disability during the application process, read more about requesting accommodations.

Requirements

Must-haves

Own proactive-planner algorithms and predictive-model calibration for TokenOps and FinOps decisions.
Develop probabilistic models that forecast cost, token demand, latency, quality, and delivery outcomes as distributions rather than point estimates.
Design stochastic optimization methods that recommend model, agent, routing, and capacity choices under uncertainty.
Apply CVaR, chance constraints, and related risk measures to keep recommendations within customer cost, reliability, and delivery tolerances.
Build conformal and empirical calibration methods that quantify uncertainty and show when predictive confidence is no longer reliable.
Use causal inference and experimental evidence to distinguish the impact of TokenOps interventions from correlation or external effects.
Develop Monte Carlo simulations and scenario analyses that make tradeoffs, tail risks, and sensitivity visible to decision makers.
Implement production-quality scientific software in Python and establish monitoring, validation, versioning, and MLOps practices for deployed models.
Partner with the Telemetry & TokenOps Engineer to define the fact-store data, attribution, quality, and lineage needed for forecasting and calibration.
Translate model outputs into clear customer recommendations, assumptions, constraints, and decision thresholds.
Create a reusable forecast and calibration assessment that can be applied consistently across customer engagements.
Customer decisions are supported by calibrated cost and delivery distributions with explicit assumptions and uncertainty.

Nice-to-haves

Demonstrated ownership of calibrated decision models operating under uncertainty and tied to real production decisions.
Strong Python and scientific machine-learning skills, including disciplined implementation, testing, and reproducibility.
Deep experience with probabilistic forecasting and evaluation of predictive distributions.
Experience formulating and solving stochastic optimization problems with operational or economic constraints.
Working knowledge of CVaR, chance constraints, and other methods for modeling tail risk and decision confidence.
Experience with conformal prediction, empirical calibration, or comparable uncertainty-quantification techniques.
Strong grounding in causal inference and the design or analysis of experiments and observational studies.
Experience deploying and monitoring predictive models through production MLOps practices.
Published or production work in AI economics, FinOps, TokenOps, or decision science.
Experience building Monte Carlo simulations, scenario models, and sensitivity analyses for executive or operational decisions.
Understanding of model, agent, token, infrastructure, and delivery cost drivers in enterprise AI systems.
Experience working with telemetry, attribution, and fact-store data in partnership with data or platform engineers.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

aipythonmachine learning

Platform: eightfold
Effort: medium