Glossary

60 terms you'll meet in interviews

Short definitions written for career changers, each linked to the careers where the term comes up.

A/B test: A randomised experiment splitting users between variants to measure a causal difference. Most tests are underpowered, so calling a winner after three days on a 2% lift is usually reading noise. Seen in: Data Scientist, Product Manager, Data Analyst
Alpha: Return above what the market or a chosen benchmark explains. Genuine alpha is rare, decays as others find the same signal, and is the entire product a quantitative fund sells. Seen in: Quantitative Analyst
AUM: Assets under management, the total client money a firm or advisor oversees. Advisor pay is usually a percentage of it, so a planner with $80m at a 1% fee is working with roughly $800k of annual revenue. Seen in: Financial Planner, Quantitative Analyst
Backtest: Running a strategy or model against historical data to estimate how it would have performed. Backtests flatter almost everything, because look-ahead bias, survivorship bias and repeated tweaking all push results up. Seen in: Quantitative Analyst, Data Scientist
Big-O: Notation describing how an algorithm's time or memory grows with input size, such as O(n log n). It matters in interviews far more than in most jobs, but it is the shared language for arguing about whether something will still work at ten times the data. Seen in: Software Engineer
Bulge bracket: The largest global investment banks, including Goldman Sachs, Morgan Stanley, JPMorgan and Bank of America. They pay roughly $110k-$120k base to first-year analysts, and elite boutiques often pay more for the same hours. Seen in: Investment Banking Analyst
Cap rate: Net operating income divided by property price, the yield a property throws off before financing. A 6% cap rate means $60k of annual net income on a $1m building, and cap rates move inversely to prices. Seen in: Real Estate Agent or Investor
Carry: Carried interest is the share of investment profits, typically 20% above a hurdle rate, that a fund's investment team keeps. It is why senior private equity pay can dwarf the cash salary, and why it arrives in lumps years after the work. Seen in: Private Equity Associate
Cash-on-cash return: Annual pre-tax cash flow divided by the cash you actually put in, rather than the full property value. Leverage raises it while also raising the chance that one bad year wipes you out. Seen in: Real Estate Agent or Investor
Churn: The rate at which customers or revenue leave over a period. In subscription businesses it caps growth arithmetically, so a product with 3% monthly churn is replacing its entire customer base roughly every three years. Seen in: Tech Sales Account Executive, Product Manager, Data Analyst
CI/CD: Continuous integration and continuous delivery: every commit is built and tested automatically, and passing builds deploy without manual steps. It converts deployment from an event into a routine and is the single biggest predictor of how fast a team can ship. Seen in: Cloud or DevOps Engineer, Software Engineer
Cohort analysis: Grouping users by when they joined and tracking each group over time, rather than pooling everyone together. It reveals whether retention is genuinely improving or whether new signups are masking a leaky product. Seen in: Data Analyst, Product Manager, Data Scientist
Commission split: How a sale's commission divides between the brokerage and the agent, often starting near 50/50 and improving with volume. New agents also pay desk fees, dues and marketing, so gross commission and take-home are very different numbers. Seen in: Real Estate Agent or Investor, Tech Sales Account Executive
Comps: A comparable company analysis values a business against the trading multiples of similar public companies. Choosing the comparison set is the judgement call, and a defensible set is worth more than a precise calculation on a bad one. Seen in: Investment Banking Analyst, Financial Analyst, Private Equity Associate
Continuing education: Ongoing credit hours required to keep a licence active, such as CPA CPE or the CFP's 30 hours every two years. It is a real recurring cost in both money and time that credential comparisons routinely leave out. Seen in: Certified Public Accountant, Financial Planner, Project Manager, Cybersecurity Analyst
Critical path: The longest chain of dependent tasks in a project, which sets the earliest possible finish date. Work that is not on the critical path can slip without moving the deadline, and knowing the difference is most of scheduling. Seen in: Project Manager
CVE: A Common Vulnerabilities and Exposures identifier, the public reference number for a specific known flaw. Severity scores are only a starting point, because whether a CVE matters depends on whether the vulnerable path is reachable in your systems. Seen in: Cybersecurity Analyst, Cloud or DevOps Engineer
Dashboard: A maintained set of charts answering a recurring question for a specific audience. The ones that survive answer a question somebody has to act on weekly; the rest quietly stop being opened. Seen in: Data Analyst, Product Manager, Financial Analyst
DCF: A discounted cash flow model values a business by projecting its future free cash flows and discounting them back to today at the cost of capital. Small changes to the discount rate or the terminal growth assumption swing the answer enormously, which is why bankers run sensitivity tables rather than quoting one number. Seen in: Investment Banking Analyst, Financial Analyst, Private Equity Associate
Design system: A shared library of components, tokens and rules that keeps a product visually and behaviourally consistent. It is infrastructure, so the payoff shows up as speed on the twentieth screen, not the first. Seen in: UX or Product Designer, Software Engineer, Product Manager
Due diligence: The investigation a buyer runs on a target before closing, covering financials, legal exposure, customers, technology and management. Associates live inside data rooms during diligence, and finding the one broken assumption is the whole job. Seen in: Private Equity Associate, Investment Banking Analyst, Management Consultant
EBITDA: Earnings before interest, taxes, depreciation and amortisation, used as a rough proxy for operating cash generation. It is the standard unit in which private companies are bought and sold, usually quoted as a multiple such as 8x EBITDA. Seen in: Investment Banking Analyst, Private Equity Associate, Financial Analyst, Certified Public Accountant
ETL / ELT: Moving data from source systems into a warehouse, either transforming before loading (ETL) or after (ELT). Cheap cloud compute pushed most teams to ELT, where raw data lands first and dbt models it in place. Seen in: Data Analyst, Cloud or DevOps Engineer, Data Scientist
Feature engineering: Turning raw data into inputs a model can use, through aggregation, encoding, ratios and time windows. On tabular problems it still moves accuracy more than swapping model architectures does. Seen in: Machine Learning Engineer, Data Scientist
Fiduciary duty: A legal obligation to act in the client's best interest ahead of your own, including on fees and product selection. CFP professionals are held to it, and it is the clearest dividing line between advice and sales in financial services. Seen in: Financial Planner
Fine-tuning: Continuing training on your own examples to adapt a pretrained model to a specific task, format or tone. It changes behaviour rather than adding knowledge, so retrieval usually beats it when the goal is getting facts right. Seen in: AI Engineer, Machine Learning Engineer
GAAP: Generally Accepted Accounting Principles, the US rulebook for financial reporting. Knowing where GAAP allows discretion, such as revenue recognition timing or impairment testing, is what separates a reviewer from a preparer. Seen in: Certified Public Accountant, Financial Analyst, Investment Banking Analyst
Inference latency: The time from sending a request to a model until you get a usable response. It drives both user experience and serving cost, and cutting it through batching, caching or quantisation is a large share of production ML work. Seen in: Machine Learning Engineer, AI Engineer, Cloud or DevOps Engineer
Infrastructure as code: Defining servers, networks and permissions in version-controlled files, usually Terraform, instead of clicking through a console. You get review, rollback and a repeatable rebuild, which is what turns a lucky environment into a reproducible one. Seen in: Cloud or DevOps Engineer
IRR: Internal rate of return is the annualised return that makes an investment's net present value zero. It rewards getting money back early, which is why a quick flip can post a higher IRR than a deal that returns more total cash over a longer hold. Seen in: Private Equity Associate, Real Estate Agent or Investor, Financial Analyst
Kubernetes: An orchestration system that schedules containers across machines and restarts them when they fail. It solves real problems at scale and creates a full-time job's worth of complexity for teams that do not have those problems yet. Seen in: Cloud or DevOps Engineer, Machine Learning Engineer
LBO: A leveraged buyout acquires a company mostly with borrowed money, using the target's own cash flows to service the debt. Returns come from paying down that debt, growing earnings, and selling at a higher multiple than you paid. Seen in: Private Equity Associate, Investment Banking Analyst
Loss ratio: Claims paid divided by premiums earned, the core measure of whether an insurance book is priced correctly. Actuaries spend their careers moving this number, and a few points of improvement is worth hundreds of millions at scale. Seen in: Actuary
MECE: Mutually exclusive and collectively exhaustive, the standard for breaking a problem into parts that do not overlap and leave nothing out. It is the structure consultants are graded on in case interviews and the reason their slides feel so tidy. Seen in: Management Consultant
MEDDIC: A qualification checklist covering metrics, economic buyer, decision criteria, decision process, identify pain and champion. It exists because deals mostly die from talking to someone who cannot sign, not from a weak pitch. Seen in: Tech Sales Account Executive
North star metric: The single measure a team agrees best tracks the value it delivers to users, such as weekly active editors or nights booked. Picking a bad one is expensive, because teams optimise exactly what you point them at. Seen in: Product Manager, Data Analyst
OTE: On-target earnings, the total pay a salesperson makes at 100% of quota, usually split about half base and half commission. RepVue's 2026 data puts median account executive OTE at $200,000 against a $100,000 base, but only a minority of reps hit target in any given year. Seen in: Tech Sales Account Executive, Real Estate Agent or Investor
Overfitting: A model that has learned the noise in its training data rather than the underlying pattern, so it performs well in development and badly in production. Holdout sets, cross-validation and simpler models are the usual defences. Seen in: Data Scientist, Machine Learning Engineer
p-value: The probability of seeing a result at least this extreme if the null hypothesis were true. It does not tell you the probability that your hypothesis is right, and treating 0.05 as a decision threshold has produced a large pile of unreproducible research. Seen in: Data Scientist, Data Analyst, Actuary
Pass rate: The share of candidates who pass a professional exam on a given sitting. CFA levels run around 40-45%, the CFP exam is near 67%, and planning for one repeat is realistic budgeting rather than pessimism. Seen in: Certified Public Accountant, Actuary, Financial Planner, Quantitative Analyst, Project Manager
Phishing: Tricking a person into giving up credentials or running something, usually through email that impersonates someone trusted. It remains the most common initial access method in real breaches, which is why training and hardware keys beat most technical controls. Seen in: Cybersecurity Analyst
PRD: A product requirements document setting out the problem, the users, the proposed solution and how success will be measured. Its function is to force disagreement into the open before engineering starts building. Seen in: Product Manager
Prompt injection: An attack where text inside the data a model reads is crafted to override its instructions. There is no complete fix, so the practical defence is treating model output as untrusted and limiting what any agent is allowed to do. Seen in: AI Engineer, Cybersecurity Analyst
Quota: The revenue a salesperson must book in a period, typically set at four to six times their OTE. Quota attainment rates across SaaS commonly sit near 40-60%, so a company's median rep earns meaningfully less than the advertised OTE. Seen in: Tech Sales Account Executive
RACI: A chart marking who is responsible, accountable, consulted and informed for each task. Its real value is forcing a room to admit that two people thought they owned the same thing. Seen in: Project Manager, Management Consultant
RAG: Retrieval-augmented generation retrieves relevant documents and puts them in the model's context before it answers. It is the standard way to ground a language model in private data, and retrieval quality, not model choice, is what usually limits the result. Seen in: AI Engineer
Scope creep: Requirements growing after a project is agreed, without a matching change to time, budget or staffing. It is the most common cause of missed dates, and the defence is a written change process rather than better effort. Seen in: Project Manager, Product Manager
Sharpe ratio: Excess return divided by the volatility of that return, used to judge whether a strategy earns its risk. A Sharpe above 2 in live trading is excellent, and a Sharpe of 6 in a backtest usually means you have a bug. Seen in: Quantitative Analyst
SIEM: A security information and event management platform that centralises logs and alerts on suspicious patterns. Working in one is the day job for most SOC analysts, and tuning out false positives is most of the skill. Seen in: Cybersecurity Analyst
SLO: A service level objective is an internal reliability target, such as 99.9% of requests succeeding over 30 days. The gap between the target and 100% is the error budget, which is what lets a team argue about shipping speed with numbers. Seen in: Cloud or DevOps Engineer, Software Engineer, Project Manager
SQL join: Combining rows from two tables on a matching key, with inner, left, right and full variants. Most analyst interview failures are join failures, usually producing duplicate rows from a many-to-many relationship nobody checked. Seen in: Data Analyst, Data Scientist, AI Engineer
TAM: Total addressable market, the full revenue available if you won every possible customer. Almost every TAM slide is too big; the useful version is the serviceable segment you can actually reach with the team you have. Seen in: Product Manager, Management Consultant, Tech Sales Account Executive
Technical debt: Shortcuts in design or implementation that make future changes slower. Some is a deliberate trade for speed and worth taking; the dangerous kind is accidental and unmeasured, showing up as estimates that keep growing. Seen in: Software Engineer, Product Manager, Cloud or DevOps Engineer
Three-statement model: A linked model in which the income statement, balance sheet and cash flow statement all update from one set of drivers. Building one from a filing without circular reference errors is the standard test of whether someone can actually do finance. Seen in: Investment Banking Analyst, Financial Analyst, Private Equity Associate
Token: The chunk of text a language model actually processes, roughly three quarters of a word in English. Pricing, context limits and latency are all counted in tokens, so cost work in AI engineering is token accounting. Seen in: AI Engineer
Usability test: Watching five to eight representative people attempt real tasks with your interface while you keep quiet. Five participants typically surface most of the serious problems, which is why the method survives despite tiny sample sizes. Seen in: UX or Product Designer, Product Manager
Variance analysis: Comparing actual results against budget or forecast and explaining the gap by driver: price, volume, mix, timing. It is the recurring core of an FP&A month-end and the skill that gets analysts invited to operating reviews. Seen in: Financial Analyst, Certified Public Accountant
Vesting: The schedule over which granted equity actually becomes yours, commonly four years with a one-year cliff. Leaving before the cliff means you get nothing, and Levels.fyi total compensation figures assume you stay long enough to collect. Seen in: Software Engineer, Product Manager, Machine Learning Engineer, AI Engineer, Cloud or DevOps Engineer
Wireframe: A low-fidelity layout showing structure and hierarchy with no visual styling. Working in wireframes first stops the conversation about colour from starting before the conversation about what goes on the screen has finished. Seen in: UX or Product Designer
Zero trust: A security model that authenticates and authorises every request instead of trusting anything inside the network perimeter. It assumes the attacker is already inside, which after twenty years of breaches is the realistic assumption. Seen in: Cybersecurity Analyst, Cloud or DevOps Engineer