AI Business Professional flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Machine learning

A subset of AI where algorithms learn patterns from data instead of following explicit hand-coded rules, improving on a task through exposure to examples.

Supervised learning

A training approach that uses labeled input-output pairs so the model learns to map inputs to known correct outputs such as categories or numeric values.

Hallucination

Fluent but incorrect or fabricated output from a generative model. It occurs because models predict plausible continuations rather than verifying facts against ground truth.

Token and context window

A token is a text unit an LLM processes, roughly a word fragment. The context window is the maximum number of tokens the model can consider at once.

Training versus inference

Training adjusts model parameters using data and compute. Inference applies an already-trained, fixed model to new inputs to produce predictions at runtime.

Fine-tuning versus RAG

Fine-tuning further trains a pretrained model on task-specific labeled examples, changing its weights. Retrieval-augmented generation instead supplies relevant external documents at query time without altering weights.

Human-in-the-loop review

A workflow design where low-confidence or high-risk model outputs are routed to a person for verification before action, using confidence thresholds to set the routing boundary.

Straight-through processing rate

The percentage of transactions completed end to end without human intervention, calculated as automated transactions divided by total transactions times one hundred.

Process mining

Analysis of system event logs to reconstruct actual workflow paths, revealing bottlenecks, deviations, and rework before automation is designed or deployed.

RPA versus AI automation

RPA executes deterministic rules across screens and systems. AI adds classification, extraction, and prediction on unstructured inputs. Combined, they handle end-to-end document workflows.

Cycle time reduction

The percentage decrease in time to complete a process, computed as baseline cycle time minus new cycle time, divided by baseline cycle time, times one hundred.

Use-case prioritization scoring

Ranking AI opportunities on business value, technical feasibility, data readiness, and risk. High-value cases with poor data are sequenced after data remediation.

Data quality dimensions

Six attributes used to assess data fitness: accuracy, completeness, consistency, timeliness, validity, and uniqueness. Model output quality is bounded by these inputs.

Data leakage in modeling

When information from the test set or target leaks into training data, producing inflated performance estimates that do not hold in production.

Class imbalance

A condition where one label dominates the dataset, making accuracy misleading. Precision, recall, F1, or AUC plus resampling or class weights are used instead.

Data lineage

A record of a dataset's origin, transformations, and movement across systems, supporting auditability, debugging, and regulatory traceability of AI inputs.

Data dictionary

Documentation defining each field's name, type, units, and allowed values, preventing conflicting metric definitions across teams that use the same data.

Structured versus unstructured data

Structured data fits tables with defined schemas. Unstructured data includes free text, images, and audio without a fixed schema. Semi-structured data uses tags or keys such as JSON.

Prompt components

The elements of an effective prompt: role, task, context, input data, constraints, and desired output format. Missing elements increase variance in model responses.

Few-shot prompting

Including example input-output pairs in a prompt so the model imitates the pattern. It typically improves format adherence compared with instructions alone.

Temperature setting

A parameter controlling output randomness. Lower values produce more deterministic, repeatable responses; higher values increase diversity and variation across runs.

Retrieval-augmented generation

A technique that retrieves relevant documents and supplies them in the prompt so answers are grounded in source material, reducing hallucination and enabling citations.

Prompt injection

An attack where untrusted input, such as retrieved text or user content, contains instructions that override the intended task. Such content must be treated as data, not instructions.

Output schema constraint

Specifying a response structure such as JSON, a table, or numbered list so model output is machine-parseable and can be validated programmatically.

Algorithmic bias sources

Bias enters through training data, labeling practices, feature selection, and deployment context. Detection requires measuring outcomes across subgroups, not just aggregate accuracy.

Data minimization principle

Collecting only the data needed for a defined purpose and not repurposing it without a lawful basis, reducing privacy risk and regulatory exposure.

AI security threat types

Common risks include data poisoning, adversarial inputs, model extraction, prompt injection, and leakage of sensitive information through model outputs.

Model card

A document describing a model's intended use, training data, measured performance, and limitations, supporting transparency, comparison, and audit.

AI governance controls

Accountability mechanisms including risk classification, approval gates before deployment, ongoing monitoring, incident response, and periodic review of models in production.

AI initiative ROI

Net benefit divided by total cost of ownership, times one hundred. Total cost includes data preparation, integration, compute, change management, and ongoing monitoring.

Build versus buy decision

Choosing to develop, purchase, or partner for AI capability based on whether it is core to competitive advantage, data sensitivity, time-to-value, and internal talent.

Pilot-to-production gap

The failure of many AI pilots to reach scaled deployment, typically caused by weak data readiness, poor system integration, unclear ownership, and insufficient change management.