Machine Learning and Data Processing
regularization
noun
Meaning
Any technique that stops a model from becoming too complex during training, in order to reduce overfitting and help it work better on new data.
Where you use it
Common forms are L1 and L2 regularization (weight decay), dropout and early stopping. Engineers add, increase or reduce regularization depending on whether the model overfits or underfits. The verb is regularize; the UK spelling is regularisation.
Examples
Stronger regularization closed most of the gap between training and validation accuracy.
We use dropout and weight decay for regularization.
Common phrases
L2 regularization · add regularization · regularization strength · weight decay
More about this term →pretraining
noun
Meaning
The first and largest training stage, in which a model learns general patterns from a huge amount of mostly unlabeled data before it is adapted to specific tasks.
Where you use it
For LLMs, pretraining usually means predicting the next token on trillions of tokens of text, and it is by far the most expensive stage. It is followed by post-training steps such as fine-tuning and RLHF. Also written pre-training; the adjective is pretrained.
Examples
Pretraining took eight weeks on a few thousand GPUs.
We don't do pretraining ourselves; we fine-tune open models.
Common phrases
pretraining data · pretraining run · large-scale pretraining · pretrained model
More about this term →temperature
noun
Meaning
A setting that controls how random a generative model's output is. Low values give more predictable, focused answers; high values give more varied and creative ones.
Where you use it
Set for each request, usually between 0 and 1 or 2. Teams use a low temperature for extraction, classification and code, and a higher one for brainstorming. Even at 0, outputs are not always exactly identical. Often adjusted together with top-p.
Examples
Set the temperature to 0 for the invoice extractor; we want output that is as consistent as possible.
At a higher temperature the slogans were more creative, but some made no sense.
Common phrases
set the temperature · low temperature · temperature of 0.7 · sampling temperature
More about this term →prompt engineering
noun
Meaning
The practice of designing, testing and improving prompts so that a model gives accurate, useful and consistent outputs for a task.
Where you use it
It includes writing clear instructions, adding examples, setting an output format and testing prompts against an evaluation set. Some people find the name too grand, so 'prompt design' and 'context engineering' are also used.
Examples
A week of prompt engineering gave us most of the gains we had expected from fine-tuning.
Good prompt engineering starts with a test set, not with clever wording.
Common phrases
prompt engineering techniques · do prompt engineering · prompt engineer · prompt design
More about this term →few-shot prompting
noun
Meaning
Giving a model a few worked examples of a task inside the prompt, so that it can follow the same pattern for a new input without any extra training.
Where you use it
Contrasted with zero-shot prompting, where you give only instructions and no examples. People say 'add a few shots' or 'a three-shot prompt'. The examples should be varied and match the real input format closely.
Examples
Few-shot prompting with five labeled tickets worked better than a long list of rules.
Zero-shot was fine for English, but Turkish needed few-shot prompting.
Common phrases
few-shot examples · zero-shot prompting · few-shot setting · three-shot prompt
More about this term →transfer learning
noun
Meaning
Reusing what a model learned on one task or dataset as the starting point for a different but related task, instead of training a new model from scratch.
Where you use it
The standard approach when data is limited: take a pretrained model, replace or add a few layers and fine-tune it on your own data. Fine-tuning is the most common form of transfer learning.
Examples
Thanks to transfer learning, 2,000 labeled images were enough to reach 95% accuracy.
We used transfer learning from a general image model to detect defects in steel.
Common phrases
use transfer learning · transfer learning approach · pretrained backbone · domain adaptation
More about this term →base model
noun
Meaning
A model that has only been pretrained, before any fine-tuning or instruction training. It continues text well but does not reliably follow instructions or chat.
Where you use it
Contrasted with instruct or chat models, which have been tuned to follow instructions. Teams pick a base model as the starting point for their own fine-tuning. In general talk, 'base model' can also just mean the model you start from.
Examples
We fine-tuned the base model, not the chat version, because we needed a very specific output format.
The base model just continued my question instead of answering it.
Usage note
'Foundation model' is a broader term for any large pretrained model that many applications are built on; it is not the same as 'base model'.
Common phrases
choose a base model · base model checkpoint · base versus instruct model · fine-tune the base model
More about this term →CNN
abbreviation
Meaning
Short for convolutional neural network: a network that slides small learned filters over an image to detect patterns such as edges, textures and shapes.
Where you use it
For years the standard model in computer vision, and still widely used on phones, cameras and embedded devices because it is efficient. It is often compared with vision transformers. Also called a ConvNet.
Examples
A small CNN runs on the doorbell camera and detects people in real time.
We compared a CNN with a vision transformer, and the CNN was faster at similar accuracy.
Common phrases
train a CNN · CNN backbone · CNN architecture · convolutional layer
More about this term →object detection
noun
Meaning
A computer vision task that finds objects in an image or video and gives both their class, such as 'car' or 'person', and their position, usually as a box.
Where you use it
Used in traffic cameras, retail shelves, factory inspection and self-driving cars. Engineers talk about detectors, bounding boxes, confidence scores and mAP (mean average precision). It goes further than image classification, which gives one label per image.
Examples
Our object detection model misses people who are partly hidden behind cars.
For object detection we need boxes, not just one label per photo.
Common phrases
object detection model · real-time object detection · object detector · detection threshold
More about this term →tokenizer
noun
Meaning
The component that splits text into tokens and turns them into numbers a model can read, and that turns the model's output numbers back into text.
Where you use it
Every model comes with its own tokenizer, and using the wrong one breaks the results. Engineers look at a tokenizer's vocabulary size and how efficiently it handles a language; some split Turkish words into many small pieces. The process is called tokenization.
Examples
This tokenizer splits long Turkish words into many pieces, so our prompts cost more.
Make sure you load the tokenizer from the same checkpoint as the model.
Usage note
The UK spelling is tokeniser, but code and documentation almost always use the US form tokenizer.
Common phrases
load the tokenizer · tokenizer vocabulary · subword tokenizer · tokenization
More about this term →sentiment analysis
noun
Meaning
An NLP task that detects the opinion or emotion in a text, for example whether a product review is positive, negative or neutral.
Where you use it
Used on reviews, social media posts, surveys and support chats. It can be simple (positive or negative) or more detailed, by aspect: 'the battery is great but the screen is dim'. Sarcasm and mixed opinions make it hard.
Examples
Sentiment analysis shows that complaints about delivery doubled after the price change.
Our sentiment analysis model struggles with sarcasm in Turkish tweets.
Common phrases
sentiment analysis model · run sentiment analysis · aspect-based sentiment analysis · sentiment score
More about this term →fairness
noun
Meaning
The goal that a model treats different groups of people equitably, for example by giving similar error rates or approval rates across genders, ages or regions.
Where you use it
Teams evaluate fairness with metrics that compare groups and report it in model cards and audits. There are several definitions of fairness and they can't always be met at the same time, so the choice depends on the use case and the law.
Examples
Before launch, we checked fairness across age groups and regions.
The fairness review found that error rates were twice as high for one dialect.
Common phrases
fairness metrics · evaluate fairness · fairness audit · group fairness
More about this term →converge
verb
Meaning
To reach a stable state during training, where the loss stops going down in a meaningful way and more training brings little benefit.
Where you use it
Said of models, training runs and losses: 'the model converged after 20 epochs', 'the loss isn't converging'. The noun is convergence. When training never converges, people check the learning rate, the data and the code.
Examples
The model converged after about 12 epochs, so we stopped there.
With the new learning rate, training converges twice as fast.
Common phrases
converge quickly · fail to converge · training converges · convergence speed
More about this term →class imbalance
noun
Meaning
A situation in which some classes in a dataset have far more examples than others, for example 990 normal transactions for every 10 fraudulent ones.
Where you use it
Very common in fraud, medical and defect detection. It makes accuracy misleading, so teams look at precision, recall and F1 instead. Typical fixes are resampling, class weights in the loss function and collecting more examples of the rare class.
Examples
Because of the class imbalance, the model learned to predict 'no defect' for almost everything.
We handled the class imbalance by oversampling the rare class.
Common phrases
severe class imbalance · handle class imbalance · imbalanced dataset · class weights
More about this term →data leakage
noun
Meaning
A mistake in which information that won't be available at prediction time, such as test examples or future data, gets into training and makes results look better than they are.
Where you use it
A classic reason why a model looks excellent in testing and then fails in production. Common causes are duplicates across splits, features that contain the answer, and random splits on time-based data. Also called target leakage or test set contamination.
Examples
99% accuracy on day one usually means data leakage, not a great model.
We found data leakage: the 'refund date' feature only exists after a customer complains.
Usage note
In security and privacy, a data leak means that private data was exposed to outsiders. In ML evaluation, data leakage is a mistake in method.
Common phrases
check for data leakage · cause data leakage · target leakage · test set contamination
More about this term →state-of-the-art
adjective
Meaning
Describes the best result or method known at a given time, usually measured on a well-known benchmark.
Where you use it
Very common in papers ('achieves state-of-the-art results') and in marketing. It is shortened to SOTA, pronounced 'SO-tuh'. Because the field moves fast, a state-of-the-art model may be overtaken within months.
Examples
The paper claims state-of-the-art results on three Turkish benchmarks.
We don't need a state-of-the-art model for this; a small one is fast and good enough.
Usage note
Use hyphens before a noun ('a state-of-the-art model') but usually not after a verb ('this model is state of the art').
Common phrases
state-of-the-art results · state-of-the-art model · achieve state-of-the-art performance · SOTA
More about this term →multimodal
adjective
Meaning
Able to work with more than one type of data, such as text, images, audio and video, like a model that can answer questions about a photo.
Where you use it
Used before nouns: multimodal model, multimodal input, multimodal search. Each type of data is called a modality. Many current LLMs are multimodal and accept images and audio as well as text.
Examples
With a multimodal model, technicians can photograph a broken part and ask what it is.
Our search is now multimodal: users can search with text or with an image.
Common phrases
multimodal model · multimodal input · multimodal search · multimodal LLM
More about this term →AI agent
noun
Meaning
A system built around an LLM that can plan steps and use tools, such as search, code or company software, to complete a task with little human help.
Where you use it
Engineers build agents, give them tools, limit what they may do and log every action. Related phrases are agentic workflows, tool use and multi-agent systems. Because agents act on real systems, permissions and human approval steps are a big topic.
Examples
The AI agent reads the bug report, finds the right file and opens a draft fix for review.
We don't let the agent send emails unless a person approves them first.
Usage note
In Turkish, 'ajan' can also mean a spy, so some teams prefer 'yapay zeka asistanı' or simply use the English word 'agent'.
Common phrases
build an agent · agentic workflow · tool use · multi-agent system
More about this term →chain of thought
noun
Meaning
Step-by-step reasoning that a model writes out before giving its final answer, which often improves results on math, logic and multi-step problems.
Where you use it
Written with hyphens before a noun: chain-of-thought prompting, chain-of-thought reasoning. Shortened to CoT. It was first triggered with prompts such as 'think step by step'; many newer models are trained to reason this way on their own.
Examples
Chain-of-thought prompting fixed most of the arithmetic mistakes.
We hide the chain of thought from users and show only the final answer.
Common phrases
chain-of-thought prompting · chain-of-thought reasoning · CoT · think step by step
More about this term →vector database
noun
Meaning
A database designed to store embeddings and quickly find the ones most similar to a query, which makes it a common part of semantic search and RAG systems.
Where you use it
You index documents, upsert vectors and run similarity or nearest-neighbor searches. Some teams use a dedicated vector database; others add vector search to a database they already have. Also called a vector store.
Examples
All 200,000 help articles are chunked, embedded and stored in the vector database.
Do we really need a separate vector database, or can our existing database handle it?
Common phrases
similarity search · vector store · nearest-neighbor search · index documents
More about this term →LoRA
abbreviation
Meaning
Short for low-rank adaptation: a fine-tuning method that keeps the original weights frozen and trains only small added matrices, which saves a lot of memory and storage.
Where you use it
The most popular parameter-efficient fine-tuning (PEFT) method. Teams train a LoRA adapter for each task or customer and swap adapters on top of one base model. QLoRA combines it with quantization so that it fits on smaller GPUs.
Examples
With LoRA we fine-tuned the 8B model on a single GPU in one afternoon.
Each customer has its own LoRA adapter, but they all share the same base model.
Usage note
Don't confuse it with LoRa (with a lowercase a), a long-range radio technology used in the Internet of Things.
Common phrases
LoRA adapter · train a LoRA · LoRA rank · QLoRA · merge the LoRA weights
More about this term →RLHF
abbreviation
Meaning
Short for reinforcement learning from human feedback: a training stage in which people rank model answers, a reward model learns their preferences, and the LLM is then tuned to score higher.
Where you use it
Part of post-training, after pretraining and supervised fine-tuning. It played a big role in making chat models more helpful and polite. Related methods include RLAIF, where AI feedback replaces some human feedback, and DPO, which learns from preferences without a separate reward model.
Examples
After RLHF, our testers rated the model's answers as much more helpful.
We worked with medical experts to rank answers for the RLHF data.
Common phrases
RLHF stage · reward model · human preference data · preference tuning
More about this term →knowledge distillation
noun
Meaning
Training a smaller 'student' model to copy the outputs of a larger 'teacher' model, so that it keeps much of the quality while being faster and cheaper to run.
Where you use it
Often just called distillation; the verb is distill, as in 'we distilled the 70B model into a 3B model'. Used to cut inference costs or to run models on phones. Licenses matter here, because some model licenses forbid using outputs to train other models.
Examples
Knowledge distillation gave us a model ten times smaller with only a 2% drop in accuracy.
We used knowledge distillation to build an on-device version of the classifier.
Common phrases
teacher model · student model · distill a model · distillation loss
More about this term →image segmentation
noun
Meaning
A computer vision task that labels every pixel of an image with a class or an object, giving exact shapes instead of rough boxes.
Where you use it
Semantic segmentation labels each pixel by class (road, sky, person); instance segmentation also separates individual objects (person 1, person 2). Used in medical imaging, satellite images and self-driving cars. The output is usually called a mask.
Examples
We use image segmentation to measure the exact size of each tumor on the scan.
Labeling data for image segmentation takes much longer than drawing boxes.
Common phrases
semantic segmentation · instance segmentation · segmentation mask · segmentation model
More about this term →bounding box
noun
Meaning
A rectangle drawn around an object in an image and stored as coordinates, showing where the object is. It is the standard output of object detection.
Where you use it
Annotators draw bounding boxes, and models predict them together with a class and a confidence score. Box quality is measured with IoU (intersection over union). In code it is often shortened to bbox.
Examples
The bounding boxes are drawn too loosely, so the model learns a lot of background.
Each bounding box is saved as x, y, width and height.
Common phrases
draw a bounding box · predicted bounding box · bounding box coordinates · bbox
More about this term →diffusion model
noun
Meaning
A generative model that learns to remove noise step by step, so it can create new images, audio or video by starting from pure noise and gradually refining it.
Where you use it
The main technology behind most text-to-image generators. Engineers talk about sampling steps, guidance scale, image-to-image editing and inpainting. Generation is slower than with single-pass models because it takes many steps.
Examples
The marketing team uses a diffusion model to show our products in different settings.
Cutting the sampling steps from 50 to 20 made the diffusion model much faster with little loss in quality.
Common phrases
text-to-image diffusion model · sampling steps · guidance scale · latent diffusion
More about this term →named entity recognition
noun
Meaning
An NLP task that finds the names of people, organizations and places, as well as dates, amounts and similar items, in a text and labels each one with its type.
Where you use it
Shortened to NER. Used to pull structured data out of contracts, invoices, news and medical notes, and to hide personal data. Custom entity types, such as product codes or drug names, often need extra labeled data.
Examples
Named entity recognition pulls the company names and amounts out of each invoice.
Our named entity recognition model confuses some Turkish surnames with place names.
Common phrases
NER model · entity types · extract entities · named entities
More about this term →corpus
noun
Meaning
A large, organized collection of texts or speech recordings used to train language models or to study how a language is used.
Where you use it
Engineers build, clean, filter and deduplicate a corpus, and describe it by its size in tokens, its languages and its sources. The word is more common in NLP and linguistics than in other areas of ML.
Examples
We built a Turkish legal corpus of about two billion tokens.
Half of the web corpus was duplicates, so we removed them before training.
Usage note
The plural is irregular: one corpus, two corpora. 'Corpus' is not 'corpse' (a dead body), although both come from the Latin word for 'body'.
Common phrases
text corpus · training corpus · build a corpus · parallel corpus · web corpus
More about this term →explainability
noun
Meaning
How well people can understand why a model made a particular prediction or decision, often with the help of tools that show which inputs mattered most.
Where you use it
In research it appears as XAI (explainable AI). It matters in credit, insurance, hiring and medicine, where people may need a reason for a decision. Common methods include SHAP and LIME. Often used alongside interpretability, which focuses on how a model works inside.
Examples
For loan decisions explainability matters a lot, because customers can ask for the main reasons behind a rejection.
The explainability report shows that the postcode had too much influence on the score.
Common phrases
model explainability · explainability tools · explainable AI · feature attribution
More about this term →guardrail
noun
Meaning
A rule or check placed around a model to stop harmful, off-topic or unsafe inputs and outputs, for example a filter that blocks personal data in answers.
Where you use it
Usually plural: teams add, test and tune guardrails on both the input and the output side. Guardrails can be simple rules, classifier models or a second LLM that checks the first. The word comes from the metal barriers along roads.
Examples
The guardrails block any answer that contains a credit card number.
After launch we retuned the guardrails, because they were also blocking normal medical questions.
Common phrases
add guardrails · safety guardrails · input and output guardrails · guardrail model
More about this term →red teaming
noun
Meaning
Testing a model or AI system by deliberately trying to make it fail or behave harmfully, in order to find weaknesses before real users or attackers do.
Where you use it
Done by internal teams, outside experts or automated tools, before and after launch. The people involved are red teamers, and their findings feed into guardrails and further training. The term comes from security exercises in which one team plays the attacker.
Examples
Red teaming showed that the assistant would reveal its system prompt if asked in French.
We planned two weeks of red teaming before the public launch.
Common phrases
red teaming exercise · red-team a model · red teamers · automated red teaming
More about this term →alignment
noun
Meaning
Making an AI system's behavior match human intentions and values, so that it is helpful and honest and avoids causing harm. Also the research field that studies this.
Where you use it
Used both for practical work (RLHF, safety training, evaluations) and for long-term research on more capable systems. People say a model is well aligned or misaligned, and talk about alignment teams and alignment research.
Examples
Alignment work made the model more willing to admit when it doesn't know something.
She left product engineering to join the alignment research team.
Usage note
In general and engineering English, alignment also means putting things in a straight line, or the alignment of text in a document.
Common phrases
alignment research · alignment team · well aligned · misaligned model · safety and alignment
More about this term →MLOps
noun
Meaning
The practices and tools for running machine learning reliably in production: versioning data and models, automating training and deployment, and monitoring models over time.
Where you use it
Built on DevOps ideas. Used for a team (the MLOps team), a role (MLOps engineer) and a set of tools: experiment tracking, model registry, feature store, CI/CD for models and monitoring. For LLM systems some people say LLMOps.
Examples
Without proper MLOps, nobody knew which model version was in production.
Our MLOps pipeline retrains the model every Monday and deploys it if the tests pass.
Common phrases
MLOps engineer · MLOps pipeline · MLOps platform · model registry
More about this term →data drift
noun
Meaning
A change over time in the data a deployed model receives, compared with the data it was trained on, which can slowly make its predictions worse.
Where you use it
Teams monitor for data drift, set alerts and retrain when it grows too large. A related idea is concept drift, where the relationship between the inputs and the correct answer changes, as when customer habits shift after a big event.
Examples
Data drift started when the app launched in a new country with different customers.
The data drift dashboard shows that the average order size has doubled since January.
Common phrases
detect data drift · monitor for data drift · concept drift · drift alert
More about this term →quantization
noun
Meaning
Storing a model's weights, and sometimes its activations, with fewer bits, for example 8-bit or 4-bit numbers instead of 16-bit, to make it smaller and faster.
Where you use it
Used to run large models on fewer or smaller GPUs, on laptops and on phones. There is usually a small loss in quality, which teams measure before shipping. People say a model is quantized to 4 bits and name formats such as INT8.
Examples
After 4-bit quantization the model fits on a single GPU, and accuracy dropped by less than a point.
Quantization made inference almost twice as fast on our servers.
Usage note
The UK spelling is quantisation. In Turkish teams people often say 'modeli kuantize ettik' informally.
Common phrases
4-bit quantization · post-training quantization · quantized model · quantization error
More about this term →data cleaning
noun
Meaning
The process of finding and fixing problems in data, such as wrong formats, typos, duplicates and missing values, so that it can be trusted for analysis.
Where you use it
A daily task for analysts and data scientists, often said to take most of a project's time. You do, automate or document data cleaning; the verb form is 'clean the data', and 'clean up the data' is more informal.
Examples
Most of last week went on data cleaning; the date column had three different formats.
Let's automate the data cleaning steps so we don't repeat them every month.
Usage note
'Data cleansing' means the same thing and is common in business and data management texts.
Common phrases
do data cleaning · data cleaning step · data cleaning script · clean up the data
More about this term →missing values
noun
Meaning
Empty cells or fields in a dataset where no value was recorded, often shown as NULL, NaN or a blank, which need to be handled before analysis.
Where you use it
You check for, count, drop, fill or impute missing values. Analysts often report the share of missing values per column and explain why data is missing, because the reason affects which method is safe to use.
Examples
About 8% of the income column has missing values, mostly from the old sign-up form.
Before you drop rows with missing values, check whether they all come from one region.
Usage note
NULL in SQL and NaN ('not a number') in Python libraries both mark missing values, but they behave differently in comparisons and calculations.
Common phrases
handle missing values · fill missing values · drop missing values · check for missing values
More about this term →outlier
noun
Meaning
A value that is very far from most of the other values in a dataset, for example one order of 50,000 dollars among orders of around 40 dollars.
Where you use it
You detect, investigate, remove or cap outliers. A careful analyst first asks whether an outlier is an error or a real and important case. Common in EDA, data cleaning and when explaining surprising averages.
Examples
The average order value jumped because of a few outliers from wholesale customers.
Don't just delete the outliers; check whether they're data entry errors first.
Common phrases
detect outliers · remove outliers · extreme outlier · outlier detection
More about this term →exploratory data analysis
noun
Meaning
The first stage of an analysis, in which you summarise and plot a dataset to understand its structure, spot patterns and problems, and form questions to test.
Where you use it
Usually shortened to EDA, said letter by letter. You do EDA at the start of every project, often in a notebook, using summary statistics, histograms and scatter plots before any modelling.
Examples
During exploratory data analysis we noticed that half the customers had no age recorded.
I'll share my exploratory data analysis notebook before we decide on a model.
Common phrases
do exploratory data analysis · EDA notebook · initial EDA · findings from EDA
More about this term →data visualization
noun
Meaning
The practice of showing data as charts, graphs and maps so that people can see patterns, comparisons and changes more quickly than in a table.
Where you use it
Used both for the skill and for a single chart ('a visualization'). Often shortened to data viz or dataviz in speech. You create, design or present a visualization. British texts spell it 'visualisation'.
Examples
Good data visualization makes the drop in March obvious without any explanation.
She runs the company's internal training on data visualization.
Common phrases
data visualization tool · create a visualization · interactive visualization · data viz
More about this term →KPI
abbreviation
Meaning
Short for key performance indicator: one of a small number of metrics that a company or team has chosen to measure progress towards its main goals.
Where you use it
Said letter by letter, plural KPIs. You set, track, hit or miss KPIs; they appear in dashboards, quarterly reviews and performance targets. Managers often ask analysts how a KPI is trending.
Examples
Our main KPI this quarter is the 30-day retention rate.
The team hit all its KPIs except customer satisfaction.
Common phrases
set KPIs · track KPIs · hit a KPI · KPI dashboard
More about this term →A/B test
noun
Meaning
An experiment in which users are randomly split into two groups that see different versions of something, so you can measure which version performs better.
Where you use it
Said 'A-B test'. You run, launch, set up and analyse an A/B test, and it also works as a verb: 'we A/B tested the new checkout'. Version A is usually the current one (the control) and B the new one. Very common in product, marketing and growth teams.
Examples
We ran an A/B test on the sign-up button, and the green version won.
Let's not launch the redesign to everyone yet; let's A/B test it for two weeks first.
Usage note
Tests with more than two versions are often called A/B/n tests; testing several page elements in combination is called multivariate testing.
Common phrases
run an A/B test · A/B test results · set up an A/B test · A/B testing platform
More about this term →statistical significance
noun
Meaning
A sign that a result is unlikely to be explained by random chance alone, usually judged by whether the p-value falls below a threshold chosen in advance, such as 0.05.
Where you use it
Used in A/B test reports and research. A result reaches or fails to reach statistical significance; the adjective 'statistically significant' is even more common. Analysts remind stakeholders that significant does not automatically mean large or important.
Examples
The new layout increased clicks, but the difference didn't reach statistical significance.
We'll wait until the test reaches its planned sample size before checking statistical significance.
Usage note
In everyday English 'significant' means 'important', but in statistics it only means 'unlikely to be due to chance'. In a very large sample even a tiny effect can be statistically significant.
Common phrases
reach statistical significance · statistically significant result · test for statistical significance · level of significance
More about this term →data pipeline
noun
Meaning
A series of automated steps that moves data from its sources, transforms it and delivers it to a destination such as a warehouse, dashboard or model.
Where you use it
Data engineers build and maintain pipelines; analysts notice when one breaks or runs late. You build, schedule, monitor and fix a pipeline, and people say a pipeline 'failed', 'ran' or 'is down'. In context often just 'pipeline'.
Examples
The dashboard is empty because the data pipeline failed overnight.
We're building a new data pipeline to bring in app events every hour.
Common phrases
build a data pipeline · pipeline failure · maintain a pipeline · pipeline runs daily
More about this term →ETL
abbreviation
Meaning
Short for extract, transform, load: the process of taking data out of source systems, cleaning and reshaping it, and loading it into a warehouse.
Where you use it
Said letter by letter. Used as a noun ('the ETL') and before other nouns ('ETL job', 'ETL pipeline'). Many modern teams use ELT instead: they load raw data first and transform it inside the warehouse.
Examples
The nightly ETL job loads yesterday's transactions into the warehouse.
The numbers look off because someone changed the ETL logic last week.
Common phrases
ETL job · ETL process · ETL pipeline · ETL tool
More about this term →mean
noun
Meaning
The value you get by adding all the numbers in a group and dividing the total by how many numbers there are; the usual meaning of 'average'.
Where you use it
Statisticians say 'mean'; business people usually say 'average'. You calculate, compare or report a mean. Because outliers and skewed data pull it easily, analysts often report it together with the median.
Examples
The mean session length is eight minutes, but most sessions are under two.
Can you add the mean and the median to the summary table?
Usage note
This noun has nothing to do with the verb 'mean' (to have a meaning) or the adjective 'mean' (unkind); the context always makes it clear.
Common phrases
calculate the mean · sample mean · mean value · mean and standard deviation
More about this term →median
noun
Meaning
The middle value when all the numbers in a group are sorted from smallest to largest; half the values are below it and half are above it.
Where you use it
Preferred over the mean for skewed data such as income, house prices or response times, because outliers barely affect it. You report, compare or calculate the median; 'median salary' and 'median time to …' are common phrases.
Examples
We report median delivery time because a few very late orders distort the average.
The median customer places three orders a year.
Common phrases
median value · median income · report the median · median time
More about this term →distribution
noun
Meaning
The way the values of a variable are spread out, showing which values are common, which are rare and how far apart they are; often viewed as a histogram.
Where you use it
You look at, plot, describe or compare a distribution. Analysts describe its shape with words such as normal, skewed, wide, narrow or long-tailed. Named examples include the normal distribution and the Poisson distribution.
Examples
Let's look at the distribution of order values before we pick a threshold.
The distribution of session lengths has a very long right tail.
Usage note
In business English 'distribution' also means getting products to shops or customers; the context makes the meaning clear.
Common phrases
normal distribution · plot the distribution · skewed distribution · distribution of values
More about this term →correlation
noun
Meaning
A relationship between two variables in which they tend to change together, often measured by a coefficient from -1 to 1, where 0 means no linear relationship.
Where you use it
You find, measure or check for a correlation; it can be positive, negative, strong or weak. Analysts often remind each other that 'correlation is not causation'. The verb is 'correlate with'.
Examples
There's a strong correlation between app usage and renewal, but we can't say one causes the other.
The correlation between page speed and bounce rate is negative: faster pages lose fewer visitors.
Common phrases
strong correlation · positive correlation · correlation coefficient · correlation between
More about this term →raw data
noun
Meaning
Data in the form in which it was first collected or exported, before anyone has cleaned, combined or summarised it.
Where you use it
You collect, store, export or keep raw data. Teams often keep an untouched raw copy so they can rerun the cleaning steps. 'Raw' contrasts with 'clean', 'processed' and 'aggregated'. Treated as uncountable: 'the raw data is…'.
Examples
Can you send me the raw data, not just the summary table?
We keep the raw data for 90 days in case we need to reprocess it.
Common phrases
raw data file · keep the raw data · export raw data · raw event data
More about this term →data source
noun
Meaning
Any place that data comes from, such as a database, an app, a spreadsheet, a survey, an external API or a file sent by a partner.
Where you use it
You connect to, combine, document or question a data source. Analysts often have to explain why two data sources give different numbers. In BI tools, 'data source' also means the connection a dashboard reads from.
Examples
The two data sources disagree because one counts refunds and the other doesn't.
Please list every data source you used in the appendix.
Common phrases
connect a data source · external data source · primary data source · combine data sources
More about this term →duplicate
noun
Meaning
A record that appears more than once in a dataset, for example the same customer or order stored twice, which can make counts and totals too high.
Where you use it
You find, check for or remove duplicates; the process is called deduplication, often shortened to 'dedupe'. Also used as an adjective ('duplicate rows') and a verb ('the join duplicated the orders').
Examples
Revenue looked too high because the table had duplicates for every refunded order.
Always check for duplicates after a join.
Usage note
The noun and adjective end in /-kət/, but the verb ends in /-keɪt/: 'a duplicate' /ˈduːplɪkət/, 'to duplicate' /ˈduːplɪkeɪt/.
Common phrases
remove duplicates · check for duplicates · duplicate rows · duplicate records
More about this term →data quality
noun
Meaning
How accurate, complete, consistent and up to date data is, and therefore how far it can be trusted when people make decisions.
Where you use it
You check, monitor, measure or improve data quality. Teams set up data quality checks or tests that send an alert when values look wrong. 'Data quality issues' is a polite way to say that the data has problems.
Examples
We found some data quality issues in the March numbers, so please don't share them yet.
A daily data quality check now flags any day with zero orders.
Common phrases
data quality issues · data quality check · improve data quality · poor data quality
More about this term →standard deviation
noun
Meaning
A measure of how spread out values are around the mean: a small value means most values are close to the average, a large one means they vary a lot.
Where you use it
Often written as SD, std or the Greek letter sigma (σ). You calculate or report the standard deviation, usually next to the mean. 'Two standard deviations above the mean' is a common way to describe unusual values.
Examples
Delivery times have a mean of three days and a standard deviation of half a day.
Anything more than three standard deviations from the mean is flagged for review.
Common phrases
calculate the standard deviation · high standard deviation · one standard deviation · mean and standard deviation
More about this term →sample
noun
Meaning
A smaller group selected from a larger population or dataset and studied in order to draw conclusions about the whole.
Where you use it
You take, draw or select a sample, and a good sample is random and representative. Also a verb: 'we sampled 10% of users'. The whole group being studied is called the 'population'.
Examples
We took a random sample of 5,000 support tickets and labelled them by hand.
The survey sample isn't representative; it's mostly people from big cities.
Common phrases
random sample · take a sample · representative sample · sample data
More about this term →trend
noun
Meaning
The general direction in which a value moves over a period of time, such as a steady rise in sales, ignoring short-term ups and downs.
Where you use it
A trend can be upward, downward, flat, long-term or seasonal. You spot, identify or follow a trend, and a metric 'is trending up'. Analysts warn that one or two data points are not yet a trend.
Examples
The downward trend in sign-ups started well before the price change.
One good week isn't a trend; let's see what happens over the next month.
Common phrases
upward trend · downward trend · long-term trend · spot a trend · trending up
More about this term →time series
noun
Meaning
A set of measurements of the same thing recorded in order at regular time intervals, such as daily revenue or hourly website visits.
Where you use it
Singular and plural are the same: 'a time series', 'several time series'. Analysts plot, smooth, decompose and forecast time series, looking for trend, seasonality and anomalies. Before a noun it is often hyphenated: 'time-series data'.
Examples
We built a time series of daily orders going back to 2019.
This time series has a strong weekly pattern, with dips every Sunday.
Common phrases
time series analysis · time-series data · forecast a time series · daily time series
More about this term →segment
noun
Meaning
A group of customers or users who share certain characteristics, such as country, device or buying behaviour; as a verb, to divide data into such groups.
Where you use it
You analyse, compare or target a segment, or break results down by segment. Analysts often segment results to check whether an average hides different behaviour in different groups. The process is called segmentation.
Examples
The new pricing works well for small businesses but not for the enterprise segment.
Let's segment the results by device; mobile users might behave differently.
Usage note
The stress changes with the part of speech: the noun is SEG-ment, while the verb is usually seg-MENT.
Common phrases
customer segment · segment the data · break down by segment · high-value segment
More about this term →hypothesis
noun
Meaning
A specific, testable statement about what you expect to happen and why, written down before an experiment or analysis starts.
Where you use it
You form, state, test, support or reject a hypothesis. Product teams often use the pattern 'If we…, then…, because…'. The plural is hypotheses /haɪˈpɑːθəsiːz/.
Examples
Our hypothesis is that a shorter form will increase completed sign-ups.
The data didn't support the hypothesis, which is still a useful result.
Common phrases
form a hypothesis · test a hypothesis · support the hypothesis · reject the hypothesis
More about this term →control group
noun
Meaning
The group in an experiment that does not get the change being tested, so its results show what would have happened without that change.
Where you use it
Also called 'the control'. In an A/B test the control group sees the current version and the treatment group sees the new one. You assign users to a control group, compare against it or hold one out.
Examples
Conversion in the treatment group was 4.1%, compared with 3.6% in the control group.
We kept 10% of users in a control group that never received the emails.
Common phrases
assign to the control group · treatment and control groups · compare against the control · holdout group
More about this term →p-value
noun
Meaning
The probability of seeing a result at least as extreme as the one observed if there were really no effect, that is, if the null hypothesis were true.
Where you use it
Written with a hyphen and a lowercase p, often in italics. A small p-value, commonly below 0.05, is taken as evidence against the null hypothesis. You calculate, report or look at the p-value.
Examples
The p-value was 0.03, so the result is significant at the 5% level.
Don't stop the test early just because the p-value dipped below 0.05 on day two.
Usage note
A p-value is not the probability that the hypothesis is true, nor the probability that the result happened by chance alone; this is one of the most common misreadings in data work.
Common phrases
low p-value · calculate the p-value · report the p-value · p-value threshold
More about this term →sample size
noun
Meaning
The number of observations, people or units included in a sample or an experiment; larger samples usually give more reliable estimates.
Where you use it
You calculate, increase or reach the required sample size. Before an A/B test, analysts estimate the sample size needed to detect an effect of a given size. Often written as n, as in 'n = 1,200'.
Examples
With this sample size we can only detect a lift of 5% or more.
The sample size is too small to say anything about users over 65.
Common phrases
calculate the sample size · small sample size · required sample size · sample size calculator
More about this term →conversion rate
noun
Meaning
The percentage of users who complete a desired action, such as buying or signing up, out of all the users who had the chance to do it.
Where you use it
One of the most watched metrics in e-commerce and marketing. You track, improve or boost the conversion rate, and people say a page or campaign 'converts well'. Always define the denominator: visitors, sessions or sign-ups.
Examples
The checkout conversion rate dropped from 3.2% to 2.7% after the update.
Is this conversion rate per visitor or per session?
Usage note
A change from 3% to 4% is a rise of one percentage point, but a relative increase of about 33%. Say which one you mean.
Common phrases
improve the conversion rate · checkout conversion rate · low conversion rate · conversion rate by channel
More about this term →churn
noun
Meaning
The loss of customers or users who stop using a product or cancel a subscription during a given period; also the rate at which this happens.
Where you use it
Used as a noun ('monthly churn is 4%') and a verb ('customers who churned'). You measure, predict or reduce churn; 'churn rate' and 'churn model' are common. A key idea in subscription, telecom and banking businesses.
Examples
Churn is highest in the first month, so onboarding matters most.
We're building a model to predict which customers will churn.
Common phrases
churn rate · reduce churn · predict churn · monthly churn · churned customers
More about this term →bar chart
noun
Meaning
A chart that uses rectangular bars of different lengths to compare values across categories, such as sales by region or users by country.
Where you use it
The most common chart in business reports. Bars can be vertical (often called a column chart) or horizontal, stacked or grouped. Good practice is to start the value axis at zero and sort the bars.
Examples
A horizontal bar chart works better here because the country names are long.
The bar chart makes the gap look huge because the axis starts at 90.
Usage note
A bar chart compares categories; a histogram shows the distribution of one numeric variable. They look similar but answer different questions.
Common phrases
stacked bar chart · horizontal bar chart · grouped bar chart · column chart
More about this term →scatter plot
noun
Meaning
A chart that shows the relationship between two numeric variables by drawing one dot for each observation, with one variable on each axis.
Where you use it
Used to look for correlation, clusters and outliers. You plot or draw a scatter plot, sometimes adding a trend line. Also written 'scatterplot' or called a 'scatter chart' or 'scatter graph'.
Examples
The scatter plot of price against rating shows almost no relationship.
Add a trend line to the scatter plot so people can see the direction quickly.
Common phrases
draw a scatter plot · scatter plot with a trend line · scatter plot matrix · plot X against Y
More about this term →aggregate
verb
Meaning
To combine many rows of data into summary values, such as totals, counts or averages, usually grouped by a category such as day, product or region.
Where you use it
In SQL this is done with functions such as SUM, COUNT and AVG together with GROUP BY. You aggregate data by a category or to a level ('aggregate to weekly'). The nouns 'aggregate' and 'aggregation' are also common.
Examples
Let's aggregate the events by day before we plot them.
The dashboard shows aggregated numbers, so you can't see individual customers.
Usage note
The verb ends in /-geɪt/ ('AG-ri-gate'), while the noun and adjective end in /-gət/ ('AG-ri-git').
Common phrases
aggregate by · aggregate to monthly · aggregated data · aggregate function
More about this term →ad hoc analysis
noun
Meaning
A one-off analysis done quickly to answer a specific question that has just come up, rather than a regular report or a planned project.
Where you use it
A large part of many analysts' week. You get, handle or prioritise ad hoc requests, and too many of them can block bigger projects. 'Ad hoc' is Latin for 'for this purpose' and is also written 'ad-hoc' before a noun.
Examples
Sorry for the delay; I had three urgent ad hoc analysis requests this week.
If this ad hoc analysis keeps coming up, we should turn it into a dashboard.
Common phrases
ad hoc request · ad hoc query · handle ad hoc requests · one-off analysis
More about this term →impute
verb
Meaning
To fill in missing values with estimated ones, for example with the column's median, the most common category or a value predicted from other columns.
Where you use it
A technical word used by data scientists and statisticians; the noun is imputation. You impute missing values with or using a method. Good practice is to record which values were imputed, for example in a separate flag column.
Examples
We imputed missing ages with the median for each country.
How did you impute the missing incomes, and how many rows did it affect?
Usage note
Outside statistics, 'impute' means to say that someone has a certain quality or motive, as in 'impute bad intentions'; in data work it only means filling gaps.
Common phrases
impute missing values · impute with the median · imputation method · imputed values
More about this term →data wrangling
noun
Meaning
The work of turning messy raw data into a clean, well-structured form for analysis, including reshaping, merging, filtering and fixing formats.
Where you use it
An informal but widely used term, close in meaning to 'data munging' and 'data preparation'. It is broader than data cleaning, because it also covers reshaping and combining. Tools such as pandas, dplyr and SQL are used for it.
Examples
The modelling took a day; the data wrangling took two weeks.
Most of the data wrangling happens in SQL before anything reaches Python.
Usage note
In American English 'wrangle' also means to herd horses or cattle, which is why the word suggests hard, untidy work.
Common phrases
data wrangling skills · do the data wrangling · data wrangling script · wrangle the data
More about this term →skewed
adjective
Meaning
Describes a distribution that is not symmetrical and has a long tail of values on one side, such as income data where a few people earn far more than most.
Where you use it
Data is 'right-skewed' when the long tail is on the high side and 'left-skewed' when it is on the low side. In right-skewed data the mean is usually above the median. Also used loosely to mean biased: 'the results are skewed'.
Examples
Revenue per customer is heavily skewed, so the median is more useful than the mean.
The survey results are skewed because only our most loyal users replied.
Common phrases
right-skewed · heavily skewed · skewed distribution · skewed results
More about this term →summary statistics
noun
Meaning
A small set of numbers that describe a dataset at a glance, such as the count, mean, median, minimum, maximum and standard deviation of each column.
Where you use it
Usually the first thing you calculate in EDA, and most tools have a one-line command for it, such as describe() in pandas. Also called descriptive statistics. You compute, report or check summary statistics.
Examples
The summary statistics show a maximum age of 212, so something's wrong with the data.
Start the report with a table of summary statistics for each region.
Common phrases
compute summary statistics · table of summary statistics · descriptive statistics · basic summary statistics
More about this term →sanity check
phrase
Meaning
A quick, simple test to make sure a result is reasonable before you trust it, for example checking that a total matches a figure you already know.
Where you use it
Very common and informal in data teams. You do or run a sanity check, or 'sanity-check' a number. People ask colleagues 'Can you sanity-check this before I send it?' when they want a second pair of eyes.
Examples
As a sanity check, I compared our revenue total with the finance report.
Could you sanity-check these numbers before I send them to the board?
Common phrases
do a sanity check · quick sanity check · sanity-check the numbers · pass a sanity check
More about this term →percentile
noun
Meaning
The value below which a given percentage of observations fall; for example, the 90th percentile of load time is the time within which 90% of page loads finish.
Where you use it
Often written as p50, p90 or p99, especially for response times. The 50th percentile is the median, and the 25th and 75th are the quartiles. People also say a customer is 'in the top 10th percentile', though 'in the top 10%' is clearer.
Examples
The 95th percentile of page load time is 1.8 seconds, while the median is 0.6.
Customers above the 90th percentile of spending get a personal account manager.
Common phrases
90th percentile · percentile rank · top percentile · calculate percentiles
More about this term →causation
noun
Meaning
A relationship in which a change in one thing directly produces a change in another, rather than the two simply moving together.
Where you use it
Most often heard in the phrase 'correlation does not imply causation'. Analysts establish causation with controlled experiments such as A/B tests, or with careful causal inference methods. The adjective is 'causal'.
Examples
Users who watch the tutorial stay longer, but we can't claim causation without a test.
The A/B test finally gave us evidence of causation, not just correlation.
Usage note
Do not confuse 'causal' (about cause and effect) with 'casual' (relaxed, informal); the two letters are easy to swap.
Common phrases
correlation and causation · establish causation · evidence of causation · imply causation
More about this term →confidence interval
noun
Meaning
A range of values, calculated from sample data, that is likely to contain the true value; a 95% interval comes from a method that captures the true value 95% of the time.
Where you use it
Often shortened to CI. You calculate or report a confidence interval, usually at 95%. Analysts present results as 'a lift of 3%, with a 95% CI from 1% to 5%'. A narrow interval means a more precise estimate.
Examples
The confidence interval runs from -1% to 4%, so we can't rule out that there's no effect at all.
Please show the confidence interval on the chart, not just the point estimate.
Common phrases
95% confidence interval · narrow confidence interval · wide confidence interval · report the confidence interval
More about this term →variant
noun
Meaning
One of the versions shown to users in an experiment, usually the new version that is compared with the control.
Where you use it
Teams talk about 'the control and two variants' or 'variant B'. You create, launch or roll out a variant and assign users to it; the best-performing one is the 'winning variant'. 'Treatment' means much the same.
Examples
Variant B increased add-to-cart clicks by 6% compared with the control.
We'll roll out the winning variant to all users next week.
Common phrases
winning variant · test variant · variant B · roll out a variant
More about this term →null hypothesis
noun
Meaning
The default assumption in a statistical test that there is no effect or no difference, for example that a new design does not change the conversion rate.
Where you use it
Written H0 and read 'H-nought' or 'H-zero'. You reject or fail to reject the null hypothesis; statisticians avoid saying that you 'accept' it. Its counterpart is the alternative hypothesis (H1).
Examples
With a p-value of 0.4, we can't reject the null hypothesis.
Our null hypothesis is that free shipping has no effect on order size.
Common phrases
reject the null hypothesis · fail to reject the null hypothesis · under the null hypothesis · alternative hypothesis
More about this term →cohort
noun
Meaning
A group of users who share a starting point or experience in the same period, such as everyone who signed up in January, and who are followed over time.
Where you use it
Cohort analysis compares how different cohorts behave in the weeks or months after they start, usually in a triangular cohort table. You build, compare or track cohorts. Common in product analytics, subscription businesses and medical research.
Examples
Retention in the March cohort is much better than in the February cohort.
Let's do a cohort analysis by sign-up month to see whether the onboarding changes helped.
Common phrases
cohort analysis · cohort table · sign-up cohort · monthly cohort
More about this term →funnel
noun
Meaning
The series of steps users go through to reach a goal, such as visit, sign-up and purchase; fewer people remain at each step, like the shape of a funnel.
Where you use it
You build, analyse or optimise a funnel and look at where users drop off. Common phrases are 'top of the funnel' (first contact), 'bottom of the funnel' (purchase), 'funnel conversion' and 'drop-off rate'.
Examples
The biggest drop in the funnel is between adding to cart and starting checkout.
Can you build a funnel chart for the new sign-up flow?
Common phrases
conversion funnel · funnel analysis · drop off in the funnel · top of the funnel
More about this term →histogram
noun
Meaning
A chart that shows the distribution of one numeric variable by dividing its range into intervals, called bins, and drawing a bar for how many values fall in each.
Where you use it
Usually one of the first charts in EDA. You plot a histogram and choose the number or width of the bins, which can change how the shape looks. The bars touch each other, unlike in a bar chart.
Examples
The histogram shows two peaks, so we probably have two different customer groups.
Try a histogram with narrower bins; the current one hides the detail.
Common phrases
plot a histogram · bin width · number of bins · histogram of values
More about this term →line chart
noun
Meaning
A chart that connects data points with a line to show how a value changes over time, such as monthly revenue or daily sign-ups.
Where you use it
The standard choice for time series and trends. You plot a line chart, often comparing several lines, but too many lines turn it into a 'spaghetti chart'. Also called a line graph.
Examples
A line chart of weekly orders makes the seasonal pattern easy to see.
There are twelve lines on this line chart; let's highlight just the top three countries.
Common phrases
plot a line chart · multi-line chart · line chart over time · line graph
More about this term →drill down
phrasal verb
Meaning
To move from a summary view to more detailed data, for example from yearly sales to monthly sales and then to single stores, in order to find the cause of a pattern.
Where you use it
Used about dashboards and analysis: you drill down into a number, by region or to the order level. Many dashboards let users click to drill down. The noun is 'drill-down', as in 'a drill-down view'.
Examples
If you drill down by region, you'll see the drop is all in the north.
Can the dashboard let managers drill down to individual stores?
Common phrases
drill down into · drill down by region · drill-down view · drill down to the detail
More about this term →filter
verb
Meaning
To keep only the rows of data that meet certain conditions, such as orders from 2024 or users in Turkey, and leave out the rest.
Where you use it
In SQL you filter with WHERE, or with HAVING after grouping; in dashboards users choose filters from drop-down menus. You filter by something and filter out what you don't want. Also a noun: 'add a filter'.
Examples
Filter out test accounts before you calculate the conversion rate.
The dashboard is filtered to last month by default.
Common phrases
filter by · filter out · apply a filter · date filter
More about this term →data lake
noun
Meaning
A large storage system that holds huge amounts of raw data in its original format, structured or not, until someone needs to process and analyse it.
Where you use it
Usually built on cloud storage such as Amazon S3. It is cheaper and more flexible than a warehouse but harder to query. A badly managed lake is jokingly called a 'data swamp', and a design that combines both is called a 'lakehouse'.
Examples
All the raw clickstream logs go into the data lake first.
The data lake has everything, but nobody knows which files are current.
Common phrases
store in the data lake · cloud data lake · data lake storage · lakehouse
More about this term →batch processing
noun
Meaning
Processing data in large groups at scheduled times, for example once every night, rather than handling each record the moment it arrives.
Where you use it
The opposite is stream processing, or real-time processing. Batch jobs run on a schedule, and people talk about the 'nightly batch' or 'daily batch'. Most reporting dashboards rely on batch processing, so their data is a few hours old.
Examples
We use batch processing for finance reports because they only need daily numbers.
With batch processing, today's orders won't show up until tomorrow morning.
Common phrases
nightly batch · batch job · batch processing pipeline · batch versus streaming
More about this term →baseline
noun
Meaning
A simple reference point, such as last year's figure, the current version or a very basic model, against which new results are compared.
Where you use it
Used in modelling ('beat the baseline'), experiments ('baseline conversion rate') and reporting ('compared with the baseline'). You set or establish a baseline and measure against it. A model that cannot beat a simple baseline is not useful.
Examples
Our baseline is simply predicting last week's sales for every store.
The baseline conversion rate is 2%, and we want to be able to detect a rise to 2.3%.
Common phrases
beat the baseline · baseline model · baseline conversion rate · compared with the baseline · set a baseline
More about this term →feature engineering
noun
Meaning
The work of creating new input variables from raw data to help a model learn, for example turning a date of birth into an age or counting a customer's purchases.
Where you use it
Often the part of a project that improves a model most. Feature engineering relies on domain knowledge; typical steps are ratios, time windows, encoding categories and aggregating events. Finished features are often kept in a 'feature store'.
Examples
Good feature engineering beat the fancy model: a simple ratio of returns to orders did most of the work.
Most of the feature engineering is done in SQL so that it runs in the warehouse.
Common phrases
do feature engineering · feature engineering step · feature store · domain knowledge
More about this term →target variable
noun
Meaning
The value a model is trained to predict, such as whether a customer will churn or how much they will spend; also called the label or dependent variable.
Where you use it
Often just called 'the target' and written y in formulas and code. You define, predict or encode the target variable. Defining it precisely, such as 'churn within 30 days', is one of the most important decisions in a project.
Examples
Our target variable is whether the customer cancels within 30 days of renewal.
Make sure no feature is calculated from the target variable itself.
Common phrases
define the target variable · binary target variable · predict the target · target leakage
More about this term →normalize
verb
Meaning
To rescale numeric values to a common range, such as 0 to 1, or to adjust figures so that groups of different sizes can be compared fairly.
Where you use it
In modelling you normalize features so that no single one dominates; in reporting you normalize by population or user count ('per 1,000 users'). The British spelling is 'normalise' and the noun is normalization.
Examples
We normalized the sales figures by store size, so small shops can be compared with large ones.
Normalize the features before you run k-means.
Usage note
In database design 'normalization' means something different: organising tables so that the same data is not stored many times. Make clear which meaning you intend.
Common phrases
normalize the data · normalize by population · normalized score · min-max normalization
More about this term →data dictionary
noun
Meaning
A document or table that describes every field in a dataset: its name, meaning, data type, allowed values, source and owner.
Where you use it
You write, maintain, check or update a data dictionary. New analysts are told to read it first, and good teams keep it up to date. In larger companies it often lives inside a data catalog tool.
Examples
According to the data dictionary, status 4 means 'cancelled by customer'.
Please add the new columns to the data dictionary before the release.
Common phrases
maintain a data dictionary · check the data dictionary · data catalog · field definitions
More about this term →event tracking
noun
Meaning
Recording specific user actions in an app or website, such as clicks, page views or purchases, as events with details like time, user and device.
Where you use it
Engineers set it up, while analysts decide which events and properties to track, often in a 'tracking plan'. Google Analytics, Segment, Amplitude and Mixpanel are well-known tools. Broken event tracking is a common reason for strange numbers.
Examples
We can't measure the new button yet because nobody added event tracking to it.
After the app update, event tracking for purchases fired twice.
Common phrases
set up event tracking · tracking plan · track an event · event properties
More about this term →web scraping
noun
Meaning
Collecting data automatically from websites with a program that downloads pages and extracts the parts you need, such as prices or product names.
Where you use it
You scrape a site or build a scraper, often with tools such as Beautiful Soup or Scrapy. Many sites limit scraping in their terms of use, and personal data is protected by laws such as GDPR and KVKK. An official API is usually preferred when one exists.
Examples
We collected competitor prices through web scraping until they released a public API.
The web scraping job broke because the site changed its page layout.
Common phrases
web scraping script · scrape a website · build a scraper · scraped data
More about this term →anomaly
noun
Meaning
A data point or pattern that clearly differs from what is expected, such as a sudden spike in sign-ups or a day with no orders at all.
Where you use it
Anomaly detection systems monitor metrics and send alerts. You detect, investigate or explain an anomaly, which may come from a real event, a bug or broken tracking. The plural is anomalies and the adjective is 'anomalous'.
Examples
The anomaly on 14 March turned out to be a bot hitting the sign-up page.
Our anomaly detection flagged a 40% drop in payments within an hour.
Usage note
An outlier is usually a single extreme value, while an anomaly can also be an unusual pattern over time. Many people use the two words loosely as synonyms.
Common phrases
anomaly detection · detect an anomaly · investigate an anomaly · data anomaly
More about this term →seasonality
noun
Meaning
A regular pattern in data that repeats over a fixed period, such as higher sales every December, more traffic on weekdays or a dip every summer.
Where you use it
Analysts adjust for seasonality before comparing periods, which is why they compare a month with the same month last year. Weekly and yearly patterns are the most common. Periods on the lunar calendar, such as Ramadan, move about 11 days earlier each year, which complicates comparisons.
Examples
Sales are down from November, but that's normal seasonality, not a real problem.
The forecast model needs to handle both weekly and yearly seasonality.
Common phrases
weekly seasonality · adjust for seasonality · seasonal pattern · seasonally adjusted
More about this term →statistical power
noun
Meaning
The probability that a test will detect an effect of a given size if that effect really exists; tests are commonly designed to have 80% power.
Where you use it
Mostly used when planning experiments: power depends on sample size, effect size and the significance level. A test with low power, called 'underpowered', can easily miss real effects. Closely linked to the minimum detectable effect (MDE).
Examples
With only two weeks of traffic, the test has very low statistical power.
We did a statistical power calculation and need about 30,000 users per group.
Common phrases
power analysis · statistical power calculation · underpowered test · 80% power
More about this term →heatmap
noun
Meaning
A chart that shows values in a grid using colours, where darker or warmer colours usually mean higher values, such as orders by day of the week and hour.
Where you use it
Also written 'heat map'. Common for correlation matrices, cohort tables and activity by time. Website heatmaps show where visitors click or how far they scroll. Designers prefer colour scales that colour-blind readers can also follow.
Examples
The heatmap shows that most orders come in on weekday evenings.
I'd use a heatmap for the correlation matrix instead of a table of numbers.
Common phrases
correlation heatmap · click heatmap · colour scale · plot a heatmap
More about this term →pivot table
noun
Meaning
A table that summarises a larger dataset by grouping rows and columns by categories and calculating totals, counts or averages for each combination.
Where you use it
Best known from Excel and Google Sheets, but the same idea exists in pandas (pivot_table) and in some SQL dialects (PIVOT). You create or build a pivot table; to 'pivot' data means to turn rows into columns.
Examples
I made a quick pivot table of revenue by region and month.
Finance still checks our numbers with pivot tables in Excel.
Common phrases
create a pivot table · pivot table in Excel · pivot the data · pivot by month
More about this term →granularity
noun
Meaning
The level of detail in data, meaning what one row represents, for example one row per order, per customer per day or per store per month.
Where you use it
Also called the 'grain' of a table. Data can be fine-grained (very detailed) or coarse (summarised). Analysts ask 'What's the granularity of this table?' before a join, because mixing levels leads to double counting.
Examples
The orders table is at order-level granularity, but the targets are monthly.
We store events at the finest granularity and aggregate them later.
Common phrases
daily granularity · level of granularity · finer granularity · grain of the table
More about this term →data lineage
noun
Meaning
A record of where data comes from, which steps and systems it passes through and how it changes along the way, right up to the reports that use it.
Where you use it
Used by data engineers and data governance teams, and often shown as a graph in tools such as dbt or data catalogs. You trace or document lineage, for example to see which dashboards will break if a source table changes.
Examples
The data lineage graph shows that six dashboards depend on this table.
Without data lineage, it took us two days to find where the wrong number came from.
Common phrases
trace data lineage · lineage graph · column-level lineage · document lineage
More about this term →