AI in Asset Management: What Data Scientists Build at BlackRock

A data scientist at an asset manager builds the models, analytics and, more and more, AI agents that the rest of the firm invests with. Portfolio managers, risk and compliance teams, researchers and client businesses are the customers. The product is a tool they use, not a trade.

At BlackRock, the world's largest asset manager, the seat sits in a central AI lab, inside investment and research teams, and inside the technology platform the firm sells to other investors. The way in is still a quantitative degree, Python and statistics. What is changing is the work.

At BlackRock's investor day in June 2025, chief operating officer Rob Goldstein described a platform called Asimov: "While everyone else is sleeping at night, we have these virtual AI agents that are scanning research notes, company filings, emails, to generate portfolio insights."

Agents like those have to be proven before anyone relies on them, and building that proof now sits at the centre of the data scientist's job. An August 2026 posting for BlackRock's AI lab asks for "benchmark construction, metric design, and statistically sound measurement of quality, safety, and reliability." For a candidate, that changes the job description more than the entry ticket.

~$14trn

Assets BlackRock manages

as the firm states it in a September 2026 posting

5

Offices of its central AI lab

New York, Edinburgh, Atlanta, San Francisco, Seattle

3 to 6

AI scientists under one lab director

Edinburgh posting, August 2026

~60

Engineers and data scientists on one portfolio-tech team

December 2025 profile

What does a data scientist do at an asset manager like BlackRock?

Most of the job is turning a loose business question into a system someone can rely on. BlackRock's AI Labs makes a vice-president data scientist responsible for "the entire solution lifecycle."

The work, day to day

A working week touches each stage of that lifecycle:

• Scoping. Sitting with a portfolio manager, risk officer or compliance specialist to "translate business needs into well-scoped AI/ML solutions."
• Prototyping. Building a first version fast with "agent frameworks, orchestration platforms, and foundation model APIs" on the firm's own data.
• Testing. Deciding what counts as a right answer, measuring how often the system reaches it, and finding where it fails.
• Shipping. Working with engineers so the system runs in production, then "clearing internal review processes."
• Explaining. Presenting the work "to senior leaders and general audiences," sometimes as a white paper.

Juniors do this inside small teams. An August 2026 posting asks a senior applied AI researcher on BlackRock's data science team to mentor junior data scientists; a director's posting from the same month gives that role "end-to-end ownership of flagship AI Labs initiatives."

A worked example: a compliance assistant

Take an illustration. A compliance officer needs to know whether a fund's investment guidelines allow a trade. The data scientist builds an assistant that finds the relevant clause and answers yes, no, or refer to a person.

1. Build the test before the tool. Compliance specialists label 400 questions with the right clause and the right answer.
2. Measure retrieval. The right clause lands in the assistant's top three search results for 364 of the 400 questions: 91%.
3. Measure answers. The assistant answers 340 correctly, 85%. Every question that fetched the wrong clause is answered wrongly, so 36 of the 60 misses are retrieval failures (400 minus 364) and the other 24 are clauses it found and then misread (364 minus 340).
4. Fix the two failures separately. Retrieval misses call for better document splitting and search; misreadings call for tighter instructions or a rule check on the output. Rerun the same 400 questions after every change.
5. Ship with a watch on it. In production, a sample of live questions goes back to the compliance specialists each week, so the score keeps being measured rather than assumed.

The number the business cares about is not 85%. It is which 60 questions failed, and whether any of them would have waved a restricted trade through.

Which BlackRock teams hire data scientists?

An assistant like that could come from several teams at BlackRock, and where a team sits decides what it builds.

The central lab

BlackRock's AI Labs began as a lab in Palo Alto in February 2018, led by Stanford engineering professor Stephen Boyd, alongside a separate data science unit co-led by Rachel Schutt, who came from News Corp. Boyd and Schutt now co-lead AI Labs, which the firm calls "BlackRock's advanced AI science and engineering organization."

Its own description is the best guide to the seat: "We are not a research lab detached from the business, nor are we a traditional software engineering organization. We are a hybrid team of scientists and engineers who build solutions that run at scale in production environments and create capacity for our business partners."

Inside the investment teams

Closer to the money, the seat moves into the investment teams themselves. An August 2026 posting for an AI engineer in BlackRock's Global Fixed Income team asks for someone who can "translate investment problems into data science solutions," working alongside portfolio managers and researchers.

Kirsty Craig, head of research, data and AI strategy for portfolio management technology, runs a team that "sits at the horizontal" across investment groups to drive investment research. BlackRock named her a Tech Fellow in December 2025, a title held by only two dozen of its thousands of engineers.

BlackRock Systematic is the older version of the same idea. The firm says it has used AI and machine learning for nearly two decades, and in 2024 it described a model trained on more than 400,000 earnings-call transcripts from over 17,000 public companies, which it says forecasts the market's reaction to an earnings call more accurately than OpenAI's GPT models. There, quant researchers own the investment models; the lab's data scientists build AI systems for the whole firm.

How AI is changing the data scientist's job

Those AI systems are changing shape. At BlackRock the work is moving from models that inform a person toward agents that do part of the work first.

From models to agents

The lab's published papers read like applied mathematics; its 2026 hiring reads like agent engineering.

What the lab publishes

• Tax-aware portfolio construction via convex optimization
• Optimal claiming of Social Security benefits
• Strategic asset allocation with illiquid alternatives

What its 2026 postings ask for

• AI agents with tool use, memory and multi-step reasoning
• Evaluation frameworks for agent quality, safety and reliability
• Testing built for non-deterministic, agentic applications

An agent here means a language-model system that chooses its own tools and steps on the way to an answer. Ask it the same question twice and it can take two routes. A good demo proves little about a system like that; a test set run every day proves far more, which is the habit the right-hand column describes.

The firm's researchers are moving the same way. In August 2025, six BlackRock researchers published AlphaAgents, a study in which three language-model agents, one each for fundamentals, sentiment and valuation, analyse stocks together and are measured against benchmarks at different levels of risk tolerance.

What BlackRock's leaders have said

What is Aladdin Copilot?

The same letter calls AI "a powerful business accelerator for Aladdin," and Aladdin Copilot is where clients see it: the generative AI layer across the investment and operations platform BlackRock runs on and sells to other firms. BlackRock describes Aladdin itself as "a tech platform that unifies the investment management process through a common data language."

A September 2026 engineering posting puts Aladdin's reach at "hundreds of institutional clients and advisors globally, supporting over $21 trillion in assets." It is a business in its own right: technology services and subscription revenue reached $530 million in the first quarter of 2026, up from $436 million a year earlier.

What the assistant does

BlackRock launched eFront Copilot for its private markets software in 2023, and by September 2024 Aladdin Copilot was available to Aladdin clients. That month Syril Smith Garson, BlackRock's head of AI product, co-wrote a post naming faster onboarding, report generation, research summaries and proactive alerts as the next capabilities.

The product page says the assistant serves "to strengthen the connective tissue across the Aladdin® platform," and sets its limits in plain terms: no investment advice, no answers "outside Aladdin platform boundaries," and "content filtering and parameters to limit risk of hallucination, misinformation or inappropriate outputs."

The client side reaches wealth managers too. In October 2025 Aladdin Wealth launched an AI tool that turns portfolio analytics and each client's investment preferences into short written commentary for advisors, with Morgan Stanley's wealth advisors the first to use it.

Who builds Aladdin's agents

BlackRock's AI engineering team built the production agents that power the platform. At LangChain's Interrupt conference in 2025, two of its engineers, Brennan Rosales and Pedro Vicente Valdez, showed how they built them on LangGraph, and LangChain's summary of the talk describes a plugin registry "that enables 50+ engineering teams to contribute tools and agents," with development driven by tests that run daily in the build pipeline.

It is the worked example above at platform scale. A tool counts as finished when it is registered and passing its tests every day, not when it answers a demo question.

How do Vanguard, JPMorgan and Schroders compare?

BlackRock's set-up is one template. Other large managers arrange the same work differently, and it shows in what they say their data scientists are for.

At Vanguard, a senior data science manager describes growing from an individual contributor to lead "a team of data scientists developing AI/ML solutions for Vanguard’s Institutional and Intermediary Business."

Mary Callahan Erdoes, who runs JPMorgan's asset and wealth management arm, told investors in May 2025 that Smart Monitor is "taking things that take days and weeks and turning them into seconds." It is separate from the bank's firmwide assistant, LLM Suite.

Schroders draws the line most explicitly. Parimal Patel, head of its Investment Insights Unit, said in 2023 that his team does not "use AI to build models and algorithms to trade," and called the approach "augmented intelligence rather than artificial intelligence."

Hedge funds organise the seat another way, as the page on AI engineers at hedge funds shows, and the vendor-side version, the forward deployed engineer, is another job entirely.

What skills does a data scientist need at BlackRock?

Back at BlackRock, the postings put measurement first and the agent layer on top of it. The fixed income AI engineer, for one, is asked to judge models with "statistical methods and hypothesis-driven experimentation."

How much finance you need

Less for the engineering seats, more for the analytics ones. The fixed income AI engineer posting calls prior financial services experience "a plus," and the Mexico City AI engineer role asks for interest in the investment management industry rather than experience in it. BlackRock's careers pages say a large majority of its engineers "don't have formal backgrounds in financial services."

The analytics and risk side is different. Its models measure risk across fixed income, equities, derivatives and alternatives, and the campus function asks for knowledge of "finance, econometrics, statistical analysis or advanced math."

How do you get a data science job at BlackRock?

For a student, whichever side they aim for, the front door is a campus programme.

Campus programmes

BlackRock's Summer Internship in the Americas runs nine weeks from June to August and is designed to replicate, in the firm's words, "as closely as possible, the experience of being a full-time BlackRock Analyst." The Full-Time Analyst Program lasts two years. It opens with an orientation in early August; analysts then join their business teams for specialized training, and in some business areas they rotate across teams.

The technology functions open to campus applicants are Analytics & Modeling, Software Engineering and Technology Operations, and the 2027 internship posting lets a candidate apply to up to two functions within a programme.

BlackRock for Universities offers an earlier way in. It grew out of a 2020 internal hackathon pitch and runs a four-week course each fall and spring that trains students who manage university investment funds to use Aladdin, each class taught by a BlackRock employee.

The quantitative master's route

Master's students in quantitative fields get their own door. The Quantitative Master's Internship runs nine weeks in the Americas and eight in the UK. In London it sits in BlackRock Systematic, which wants people comfortable "translating statistical models and algorithms into code."

Inside BlackRock's Analytics & Modeling function, a quantitative research team supports almost 200 analytics models, from single securities and securitized products to portfolio risk, performance attribution, trading and liquidity.

Lateral hires

Experienced hires come in at a level set by years and degrees. BlackRock's 2026 postings asked for:

• A director in AI Labs: a PhD and eight years of experience, or a master's and eleven.
• A vice-president modelling data scientist in AI Labs: a PhD in a quantitative field and four years of experience, or a master's and seven.
• An AI Labs AI and machine learning engineer: eight years of software engineering, four of them shipping AI or machine learning into production.
• The Budapest senior applied AI researcher: six years of experience.
• The fixed income AI engineer: a bachelor's or master's degree and four years of relevant experience.

What the postings pay

BlackRock's US postings print a base salary range for each location. Its own postings in 2026 listed:

• January 2026, summer intern in Technology: $50.48 to $60.10 an hour in California, New York City, Washington state and DC
• July 2026, full-time analyst: $105,000 to $125,000 in Technology and $105,000 in Analytics & Risk, in the same locations
• September 2026, AI solutions research associate at the BlackRock Investment Institute: $145,000 to $190,000 in New York
• August 2026, vice-president modelling data scientist in AI Labs: $167,500 to $225,000 in New York or San Francisco
• August 2026, vice-president AI and machine learning engineer in AI Labs: $190,000 to $240,000 in New York

What do BlackRock interviews test?

BlackRock publishes its campus process step by step, and the technical functions add a coding test to it.

The rule on AI

BlackRock's guidance splits AI use into before and during. Before an interview, candidates may use AI for "brainstorming questions, preparing verbal or written communication, researching BlackRock, reviewing financial concepts and refining their resume/CV." During one, it is out: "BlackRock does not permit the use of AI tools during any part of the live interview process, including conversations, technical assessments and video-recorded assessments (e.g., via HireVue)."

The principles behind the questions

Interviews are "focused on the BlackRock Principles, your experience or your capabilities," and the five principles are short enough to learn by heart:

1. We are a fiduciary to our clients.
2. We are One BlackRock.
3. We are passionate about performance.
4. We take emotional ownership.
5. We are committed to a better future.

The third carries the line most useful to a technical candidate: BlackRock calls its people lifelong students of markets, of technology and of the world.

How to prepare for a data science job at BlackRock

Preparing for that seat means showing you can build a tool and prove it works, in that order.

1. Build one retrieval tool on public fund documents and grade it. Write 50 to 100 test questions with known answers before you tune anything, and log what failed and why, as in the worked example. That log is your interview story. Five more built the same way, on filings, sentiment labels, fraud data and economic series, are in AI projects for finance students.
2. Be able to rebuild everything you submit. Walk through your own code line by line, out loud; the guidance on work you cannot explain makes this non-negotiable.
3. Rehearse the recorded format. Practise answering in under three minutes on camera, and sit a timed coding set, since both come before any conversation.
4. Map two stories to each principle. Use the STAR structure BlackRock's own guidance recommends, and make at least one about a model or dataset that failed and what you learned from it.
5. Read what the firm publishes about its own AI. The lab's papers and the client assistant's product page give you a specific answer to "why BlackRock" that general enthusiasm for AI cannot.
6. Join a student fund that uses BlackRock for Universities. You learn the firm's platform before you interview.

A structured self-check lives on the interview readiness tool, and the guides cover the AI systems finance firms have built for themselves.

The bottom line

A data scientist at an asset manager like BlackRock is paid to build what the rest of the firm invests, complies and advises with, and to prove it works. At BlackRock the seat sits in a central lab, inside investment teams and behind the client platform, and the way in is still a quantitative degree, Python and real statistics.

What has changed is the product: less a single model a person reads, more agents that take the first pass for the people who decide. The candidate who can show exactly how they tested one is the candidate BlackRock's 2026 postings describe.