Decision Scientist Job Description

A decision scientist is the analyst who gets handed a question with money attached: should we raise the subscription price, which of three onboarding flows should ship, how much of next year's marketing budget should move from paid search to retail media. The job is to build the evidence, quantify the uncertainty and hand a leader a recommendation they can defend. The output is a decision memo, not a model in production.

The title is not standardized. A 2026 hiring guide from the INFORMS Career Center describes two versions of it. In some companies the title names a product-and-experimentation role built on causal inference, A/B testing and quasi-experimental design. In others it means classical decision analysis under uncertainty: expected-value reasoning, decision trees, Bayesian updating and stochastic modeling applied to capital allocation, pricing or operations.

Experimentation and causal inference

On the experimentation side, a decision scientist designs tests before anything ships. That means picking the randomization unit (user, session, store, market), calculating the sample size needed to detect an effect worth acting on, choosing a primary metric and guardrail metrics, and agreeing on stopping rules so nobody calls a winner early. When a randomized test is impossible, because a price change hits a whole region or a policy applies to everyone at once, the work moves to quasi-experimental methods: difference-in-differences, synthetic control, regression discontinuity, instrumental variables or matching.

Experiment platforms such as Optimizely, Statsig, Eppo or in-house systems run the assignment and logging. The decision scientist owns the interpretation, including checks for sample ratio mismatch, novelty effects, interference between treatment groups and metrics that move for the wrong reasons, whether or not the platform flags them.

Decision analysis

On the decision-analysis side, the toolkit is older and more explicit about uncertainty. A decision scientist structures the choice as a tree or influence diagram, attaches probabilities and dollar outcomes to each branch, and runs Monte Carlo simulation and tornado charts to show which assumptions actually drive the answer. Spreadsheet add-ins such as @RISK, dedicated tools such as TreeAge Pro or Analytica, and Python libraries such as PyMC all show up in this work.

Who they work with

The daily counterparts are product managers, finance partners, marketing leads and engineers who instrument the data. The audience for the final readout can be a vice president or the executive team. That is the real distinction from a model-building role: a decision scientist is measured on whether leaders made a better call, and sometimes the right readout is "the data does not support this yet."

There is no license for this work and no single required degree. As a rough guide, people move into decision science from statistics, economics, operations research, industrial engineering, applied mathematics, psychology research or quantitative marketing. What the background has to supply is the ability to reason about cause and effect and about uncertainty, not just to fit models.

Education

For the nearest BLS occupation, operations research analysts, the Occupational Outlook Handbook says entrants typically need at least a bachelor's degree, that some employers require or prefer a master's degree, that students need extensive coursework in mathematics, and that computer science coursework matters because analysts rely on statistical and database software. For decision science specifically, a master's in statistics, economics, analytics or operations research is a sensible route, and economics PhDs with applied microeconometrics training are a natural fit for causal-inference-heavy roles.

Experience

As a rough guide, entry-level decision scientist roles fit candidates with 1 to 3 years of analytics work, and senior or staff roles fit people with 5 or more years who have owned experiment programs or high-stakes decision models. Useful prior jobs include product analyst, marketing analyst, pricing analyst, economist, operations research analyst and survey or research scientist.

Skills hiring managers test

• Experiment design: power analysis, randomization units, variance reduction methods such as CUPED, sequential testing and guardrail metrics.
• Causal inference: difference-in-differences, synthetic control, regression discontinuity, instrumental variables, propensity matching and the assumptions behind each.
• Decision analysis: decision trees, expected value, value of information, sensitivity analysis and Monte Carlo simulation.
• Data work: SQL against a cloud warehouse such as Snowflake, BigQuery or Databricks, plus Python (pandas, statsmodels, PyMC) or R.
• Communication: a one-page memo that leads with the recommendation, states the uncertainty plainly and survives questions from a CFO.

Certification

No certification is required. The closest professional credential is the Certified Analytics Professional program from INFORMS, the operations research and analytics society. It now has three levels. CAP-Essentials and CAP-Pro have no eligibility requirements beyond agreeing to the INFORMS Code of Ethics and passing the exam, while CAP-Expert requires meeting education, experience, soft skills and ethics criteria and filing an application before the exam. All three levels are based on the INFORMS Analytics Framework. INFORMS stopped enrolling new candidates for its older entry credential, aCAP, in June 2025. Treat the credential as a supplement to a portfolio of written experiment readouts, not a substitute for one.

Because BLS does not count decision scientists separately, the best public outlook data comes from the closest occupation, operations research analysts. In its current Occupational Outlook Handbook, BLS projects employment of operations research analysts to grow 12 percent from 2025 to 2035, much faster than the average for all occupations, with about 7,500 openings a year on average over the decade. BLS expects many of those openings to come from replacing analysts who change occupations or retire. Operations research analysts held about 113,100 jobs in 2025.

BLS ties that growth to companies and government agencies seeking efficiency and cost savings, and to demand for analytics that improves business planning and decision making. It also notes that improvements in analytical software have made operations research more affordable and applicable to a wider range of areas.

Where the jobs are

Finance and insurance employed 26 percent of operations research analysts in 2025, and professional, scientific and technical services employed 22 percent, according to BLS. Pay differs across those settings. In May 2025 OEWS data, the median for these analysts was $100,590 in management, scientific and technical consulting services and $80,870 in finance and insurance.

Experimentation is not limited to tech. A January-February 2025 Harvard Business Review article on scaling experimentation noted that, owing to the widespread availability and lower cost of experimentation tools, most organizations, even those outside the technology sector, conduct online experiments. Each of those programs needs someone who can tell a real effect from noise, and that is decision science work.

How AI is changing the job

Generative AI assistants can draft SQL, write first-pass analysis code and summarize experiment results. Where teams use them, the decision scientist's value sits in the parts that require judgment: framing the question, choosing an identification strategy and deciding whether an AI-produced number can be trusted. An INFORMS-announced study in the journal Decision Analysis, released Nov. 11, 2025, found that generative AI could help define viable objectives for organizational and policy decisions, but the overall quality of those objectives fell short unless humans intervened. Its authors proposed a four-step hybrid model that combines AI brainstorming with expert refinement.

AI product launches are also decisions to test. A model release, ranking change or assistant feature can be evaluated with an online experiment, because an offline benchmark score does not by itself show how customers respond.

Advancement

Senior decision scientists move into staff or principal individual contributor roles that set methodology for a whole data organization, or into management of a decision science team. Others shift sideways into product management, strategy or pricing leadership, where the habit of making uncertainty explicit is a direct advantage.

Dear Hiring Manager,

I am applying for the Decision Scientist role on your marketplace team. For the past four years I have been a product analyst at a regional grocery delivery company, where my job has drifted from building dashboards to deciding which ideas deserve a test and what we should do with the result.

The project I am proudest of started as a request to "prove" that a new delivery fee structure worked. A before-and-after comparison made it look like a big win, but the launch had coincided with a holiday period. I proposed a staggered rollout across metro areas and analyzed it with difference-in-differences. The real effect on order volume was close to zero, while the effect on basket size was positive and held up through two robustness checks. Leadership kept the fee change but dropped a planned marketing push built on the inflated number, which saved the team a quarter of wasted effort.

I also rebuilt our A/B testing checklist. We now run power calculations before launch, check for sample ratio mismatch on day one, and require a written prediction from the product owner before results are unblinded. Test readouts went from a slide of charts to a one-page memo that leads with the recommendation and the uncertainty around it.

My tools are SQL on Snowflake, Python with pandas, statsmodels and PyMC, and a lot of plain writing. I use AI assistants to draft queries and code, and I check every number they produce against the warehouse before it goes into a memo.

I would welcome the chance to bring that mix of rigor and plain communication to your team.

Sincerely, Jordan Ellison