The Agentic AI Glossary

Most of this vocabulary was coined in demonstrations and acquires a different weight once a system is deployed and answerable for what it produces. Each definition gives the production meaning.

Define an Outcome

A

29 terms

Abandonment rate Evio: The share of callers who give up before reaching a resolution or a person.
Accent robustness Evio: Whether recognition holds up across the range of accents in a real caller population rather than a test set.
Access-control-aware retrieval Compass: Retrieval that respects the permissions of the person asking, so an answer never exposes a document they could not open.
Action confirmation Compass: The explicit step where a person approves an action before the system commits it.
Adaptive learning path Atlas: A sequence that changes according to what the learner has already demonstrated, rather than one fixed order for everyone.
Admissions evaluation Prism: Assessing candidates for a programme, where fairness and consistency are examined as closely as the outcome.
Adverse impact Prism: A materially different outcome rate for a protected group, which carries legal consequence as well as reputational.
Agentic automation Forge: Automation that decides the next step from the state it observes rather than replaying a fixed script.
AI audit trail Governance: The complete record of what a system did, on what evidence, and who authorised it, retained for the period the decision can be questioned.
AI avatar Atlas: A synthetic presenter that delivers a module in video, allowing one approved explanation to reach every location.
AI interview Prism: A structured interview conducted live by an agent in voice or video, with every candidate asked comparable questions and scored against the same rubric.
AI risk register Governance: The catalogue of identified AI risks, each with a named owner and a stated control, maintained as a live document.
Anchor example Prism: A worked example showing what a given score looks like, so reviewers are calibrated against something concrete.
Answer attribution Compass: Naming the source behind each part of an answer rather than the answer as a whole.
API gateway Governance: The single controlled entry point to a system, where every request is authenticated and inspected before it reaches any service.
Append-only log Governance: A record that can be added to and never amended, which is what makes a history worth presenting.
Applicant experience Loupe: How the process feels to the person completing it, which most intake programmes measure last or not at all.
Applicant tracking system (ATS)Prism: The system that manages candidates from application to offer, which should receive evaluation results only after a person has reviewed them.
Application programming interface first Forge: Preferring a supported integration wherever one exists, and reserving interface automation for where none does.
Assessment framework Prism: The structure within which rubrics and their thresholds are defined and governed.
Assessment integrity Atlas: The degree to which a score reflects the learner rather than the conditions the assessment was taken under.
Assessment item Atlas: A single question with its options and its correct answer, traceable to the source it is drawn from.
Attachment validation Loupe: Confirming that a supporting document is legible and of the type the rule asks for.
Attended and unattended automation Forge: Whether an automation executes alongside a person at their desk or on its own against a schedule.
Audit of a decision Prism: Reconstructing how a particular outcome was reached, long after the people involved have moved on.
Audit-ready submission Loupe: A submission whose completeness and validation history can be produced on request, with the documents behind them.
Automatic question generation Atlas: Producing assessment items from a source document rather than authoring them separately from it.
Automatic speech recognition (ASR)Evio: Converting spoken audio into text, the first step on which everything else in a voice system depends.
Average handle time Evio: The mean duration of a contact including wrap-up, long used as a proxy for efficiency.

B

8 terms

Barge-in Evio: Allowing a caller to interrupt the agent mid-sentence, as people do with each other.
Bias testing Prism: Examining outcomes across groups to establish whether a process is treating them differently.
Blended delivery Atlas: A mix of automated and human-led training, where the human time is reserved for what genuinely needs it.
Blind review Prism: Assessing a submission with identifying details withheld, so the judgement rests on the content.
BM25 Compass: The established keyword ranking method, still the stronger half of hybrid search for exact terms and identifiers.
Bot sprawl Forge: An estate of automations nobody owns, each quietly depending on an interface that will eventually change.
Browser automation Forge: Driving a web application the way a person would, through its pages rather than its backend.
Business rule validation Loupe: Applying the organisation's own policy as an automated check rather than as guidance in a manual.

C

36 terms

Calibration session Prism: A meeting where reviewers score the same submissions and reconcile their differences before the real review begins.
Call transcription Evio: The written record of a call, which is the input to every form of analysis that follows.
Canary deployment Forge: Releasing a change to a small share of traffic first, so a problem is found on a small population.
Change detection Compass: Identifying which source documents have changed or disappeared since the last ingestion, so only those are re-indexed.
Change window Forge: The agreed period in which changes may be applied, which an automation vendor has to operate inside.
Checkpoint Forge: A recorded point a task can resume from, so a failure does not mean repeating everything before it.
Chunk enrichment Compass: Adding the context a passage needs to be understood on its own, such as its document title and section path, before it is indexed.
Chunking Compass: How a long document is divided before indexing, which quietly determines whether retrieval ever finds the right passage.
Citation grounding Compass: Attaching every statement in an answer to the passage it came from, so a reader can check it.
Code switching Evio: Moving between languages inside a single sentence, which is ordinary in Indian conversation and difficult for most systems.
Cohen's kappa Prism: A statistic for agreement between two raters that accounts for the agreement chance alone would produce.
Cohort Atlas: A group of learners treated together for rollout and reporting, usually by role or by start date.
Cold transfer Evio: Passing a caller on with no context, which is where most of the frustration in handover comes from.
Compensating transaction Forge: A deliberate reversing action for a step that cannot simply be rolled back.
Competency framework Atlas: The structured description of what a role requires, against which an assessment score is interpreted.
Completeness check Loupe: Confirming that everything required is present before a submission is allowed to proceed.
Completion record Atlas: The record of who completed which module, on which version of the source, and when.
Computer use Forge: A model operating a computer directly by reading the screen and issuing clicks and keystrokes.
Concordance Prism: The share of cases where an automated score and a human reviewer arrive at the same conclusion.
Conditional logic Loupe: The rules that decide which parts of a form appear, and under what circumstances.
Confidence calibration Prism: A score that signals how certain it is, so low confidence can be routed to a person.
Consent and declaration capture Loupe: Recording what the applicant agreed to and declared, in a form that stands up later.
Consent capture on a call Evio: Recording that a caller agreed to something, in a form that can be produced later as evidence.
Containment rate Evio: The share of contacts resolved without reaching a person, which is the number most contact centres are judged on.
Content freshness Atlas: How recently training content was regenerated against its source, and how far it has drifted since.
Context assembly Compass: Selecting and ordering the retrieved material a model reads before it answers, with every source labelled and within a fixed budget of tokens.
Context engineering Compass: Deciding what a model should have in front of it for a given task, which has largely displaced prompt wording as the discipline that matters.
Context window Compass: The amount of text a model can consider at once, which bounds how much evidence an answer can rest on.
Controlled document Atlas: A policy or procedure under formal version control, where the current revision is the only one that counts.
Conversational memory Compass: What a system carries forward between turns, and what it deliberately forgets.
Course authoring Atlas: The work of turning source material into a structured course, historically the slowest and most expensive step.
Credential vault Forge: Managed storage for the secrets an automation needs, so credentials are never held in the automation itself.
Cross-field validation Loupe: Checking entries against one another, where each is individually valid and the combination is not.
Customer relationship management (CRM)Evio: The system of record for customers and their interactions, where an automated agent writes back only the fields the organisation has agreed.
Customer-managed keys Governance: Encryption keys held and controlled by the customer rather than by the vendor.
Cut-off threshold Prism: The score at which a decision changes, and the justification for placing it where it was placed.

D

26 terms

Data drift Governance: The input distribution shifting away from what a system was evaluated on, which usually precedes model drift.
Data quality score Loupe: A graded view of what arrived, so intake quality can be tracked rather than discussed anecdotally.
Data residency Governance: The requirement that data remains within a stated jurisdiction, including while it is being processed.
Dead letter queue Forge: Where messages go when they cannot be processed, so that failures are visible rather than lost.
Decision provenance Governance: The lineage of a decision, reaching back to the data and the model version in force at the time.
Decision record Prism: The complete account of a decision, from its inputs to its named author, retained for as long as the decision can be questioned.
Defensibility Prism: Whether a decision can be explained and substantiated when it is challenged, which is the only standard that matters at the point of appeal.
Deny-by-default egress Governance: Permitting outbound connections only to destinations explicitly allowed, and refusing everything else.
Desktop automation Forge: Driving applications on a desktop, including the legacy ones that were never given an interface for integration.
Deterministic replay Governance: Reconstructing exactly what a system did on a given occasion, by replaying recorded inputs against a pinned version.
Dialogue management Evio: The logic that decides what the agent does next, given the state of the conversation and the caller's intent.
Digital Personal Data Protection Act (DPDP)Governance: India's data protection statute, governing consent, purpose limitation and the obligations of those processing personal data.
Disposition code Evio: The classification applied to a call afterwards, which is what reporting and routing decisions are later drawn from.
Distractor generation Atlas: Writing the wrong answers in a multiple-choice item so that they are plausible to someone who has not understood the material.
Do Not Call register Evio: The regulated list of numbers that may not be contacted for marketing, which an outbound system has to honour.
Document classification Loupe: Identifying what kind of document has arrived before deciding how to treat it.
Document connector Compass: An authenticated link to a repository that keeps its content and its permissions in step with the index.
Document ingestion Atlas: Reading a source file and extracting its structure and content so a system can work from it directly.
Document splitting Loupe: Separating a single uploaded file that in fact contains several distinct documents.
Document store Compass: The repository holding the original files and their versions, so every cited answer can be traced back to the exact document it came from.
Document summarisation Compass: Producing a short, faithful account of a long document, used to index it more accurately and to let a reader judge it before opening it.
Document understanding Loupe: Interpreting a document's structure and meaning, which goes well beyond recognising its characters.
Document vault Loupe: Secure storage for documents applicants upload, with each file scanned and then encrypted under its own key.
Document-to-Markdown conversion Compass: Converting files of every format into structured Markdown before chunking, so headings and tables survive into the index.
Dual-tone multi-frequency (DTMF)Evio: Keypad tones, still the safest way to collect a card number or an account reference on a call.
Dynamic form Loupe: A form whose fields change according to earlier answers, so an applicant is asked only what applies.

E

20 terms

Eligibility check Loupe: Confirming that an applicant qualifies before effort is spent on the rest of the assessment.
Embedding Compass: A numerical representation of a passage of text, arranged so that similar meanings fall near one another.
Embedding service Compass: The service that converts text into numerical vectors for retrieval, ideally hosted inside the client tenancy so no passage leaves it.
Endpointing Evio: Judging that a speaker has finished, which determines how responsive a call feels more than raw model speed does.
Enterprise resource planning (ERP)Forge: The core system for finance and operations records, which automation reaches through a tightly scoped service account.
Enterprise search Compass: Finding material across an organisation's systems, the category an answering layer is measured against.
Entity extraction Compass: Identifying the named things in a question or a document, such as people, products, policies and dates, so retrieval can match on them precisely.
Entity resolution Compass: Recognising that several records refer to the same person or organisation despite differing in form.
Environment drift Forge: The accumulation of small changes in applications and configuration that eventually stops an automation working.
Error reason code Loupe: A specific, reportable statement of why something failed, attached to the failure itself.
Escalation policy Governance: The stated conditions under which a system must stop and involve a person.
EU AI Act Governance: European legislation classifying AI systems by risk and attaching obligations to each tier.
Evaluation harness Governance: The repeatable test suite a change has to pass before it reaches production.
Evidence linking Prism: Attaching each score to the passage in the submission that justifies it.
Evidence pack Forge: The bundled record of an automated case, holding its inputs and every decision with its reason, ready for an auditor.
Exception handling Loupe: The defined path for submissions that cannot be processed automatically, with ownership attached.
Exception queue Forge: The work an automation could not complete, presented to a person with the reason attached.
Execution ledger Forge: The record of every execution, with its outcome and the person who authorised it.
Explainability Governance: Presenting the basis for an output in terms the person affected by it can evaluate.
Exponential backoff Forge: Lengthening the wait between retries so that a struggling system is given room to recover.

F

9 terms

Faithfulness Compass: Whether an answer represents its sources accurately, as distinct from whether it is agreeable or well written.
Federated search Compass: Querying several repositories where they are, instead of copying everything into one place first.
Field-level validation Loupe: Checking an entry at the field where it was made, so the error surfaces to the person who can correct it.
First contact resolution Evio: The share of issues resolved on the first attempt, without a callback or a second ticket.
First-pass yield Loupe: The share of submissions accepted without rework, which is the cleanest measure of intake quality.
Fluency bias Prism: Rewarding submissions that are well written over submissions that are correct, which is the commonest distortion in manual review.
Form abandonment Loupe: Applicants starting a form and stopping, which is usually a design problem rather than a motivation problem.
Form accessibility Loupe: Meeting the standards that allow a form to be completed by people using assistive technology, which public procurement increasingly requires.
Function calling Compass: The implementation of tool calling, where the model emits a structured request matching a declared signature.

G

6 terms

Golden dataset Governance: A fixed, curated set of cases with agreed answers, used so that results remain comparable across versions.
Grant evaluation Prism: Assessing funding applications against published criteria, where every unsuccessful applicant reads the standard applied.
Graph index Compass: An index of entities and the relationships between them, held alongside the vector index for questions that depend on how things connect.
Graph traversal Compass: Following the relationships in a graph index outward from the entities in a question, to reach material that answers it through connection.
Groundedness Compass: The measurable degree to which an answer is supported by the documents retrieved for it.
Guardrails Governance: Enforced limits on what a system may do or say, applied outside the model rather than requested of it.

H

13 terms

Hallucination Compass: A fluent statement that is unsupported by any source, which is the failure mode that matters most in a regulated setting.
Halo effect Prism: One strong element lifting the assessment of unrelated elements in the same submission.
Handwriting recognition Loupe: Reading handwritten entries, which remains a genuine constraint on public sector and field-collected forms.
Hardware security module (HSM)Governance: Dedicated hardware that holds cryptographic keys and performs operations without exposing them.
Headless browser Forge: A browser without a visible window, used to execute web steps on a server.
Hit rate Compass: The share of questions for which the correct passage appeared anywhere in the retrieved set.
Human approval step Forge: A gate in a workflow where a named person authorises an action before it commits.
Human final call Prism: The point beyond which a model recommends and a named person decides.
Human handoff Evio: The moment a system stops and a person takes over, and the record of why that happened.
Human in the loop Governance: A person inside the decision path, whose approval is required before the outcome stands.
Human on the loop Governance: A person supervising a system that acts on its own, intervening by exception rather than by default.
Human resource management system (HRMS)Atlas: The system holding employee records with their roles and reporting lines, which a training platform reads to know who should learn what.
Hybrid search Compass: Combining keyword matching with meaning-based search, because each one fails on cases the other handles.

I

10 terms

Idempotency key Forge: An identifier that ensures a repeated request has the same effect as a single one, so a retry cannot pay an invoice twice.
Image captioning Atlas: Generating a written description of an image or diagram, so its content can be searched and explained.
Inference cost per task Governance: What a completed unit of work costs to process, which is the figure a finance function will eventually ask for.
Inline assistance Loupe: Guidance delivered at the field being completed rather than in a separate document nobody opens.
Intelligent document processing (IDP)Loupe: Turning submitted documents into structured, validated data rather than into images filed somewhere.
Intent detection Evio: Working out what the caller actually wants, as distinct from the words they used to ask for it.
Inter-rater reliability Prism: The degree to which two reviewers independently reach the same judgement on the same submission.
Interactive voice response (IVR)Evio: The menu-driven system a voice agent is generally deployed to replace, or to stand in front of.
ISO/IEC 42001 Governance: The management system standard for artificial intelligence, certifiable in the way information security already is.
Item difficulty Atlas: The proportion of learners who answer an item correctly, which tells you whether it is measuring anything.

J

2 terms

Jailbreak resistance Governance: How well a system holds its constraints under deliberate attempts to remove them.
Judge calibration Prism: Checking the automated scorer against human decisions, regularly and on cases it has not seen.

K

7 terms

Key-value extraction Loupe: Pulling named fields and their values out of a document, which covers most routine intake.
Know your customer (KYC)Loupe: The regulated identity and background checks an institution completes before onboarding a customer.
Knowledge base Compass: The organised body of material a system answers from, with an owner and a review cycle rather than a folder.
Knowledge check Atlas: A short question placed inside a module to confirm the point has landed before the learner continues.
Knowledge decay Compass: The rate at which an answer becomes wrong because its source changed and the index did not.
Knowledge graph extraction Compass: Reading documents to identify the entities and relationships they describe, then resolving and storing them as a graph with every fact sourced.
Knowledge retention Atlas: What a learner can still demonstrate some weeks after completion, as distinct from what they scored on the day.

L

9 terms

Large language model as judge Prism: Using a model to score against a rubric, which requires its own calibration and its own evidence.
Latency budget Evio: The total time available for a response, divided across recognition, retrieval, generation and speech.
Layout analysis Loupe: Determining the regions of a page and how they relate, before any text is interpreted.
Learner analytics Atlas: What completion and timing data can and cannot tell you about whether training worked.
Learning management system (LMS)Atlas: The platform of record for enrolment and completion reporting, which an AI training layer writes into rather than replaces.
Learning record store (LRS)Atlas: The repository where experience statements are kept, separate from the system that delivered the learning.
Lip sync Atlas: The alignment between a synthetic voice and the mouth movement of the presenter rendering it.
Listwise reranking Compass: Reordering retrieved passages by having a model compare the whole candidate set at once, instead of scoring each passage in isolation.
Long-lived task Forge: Work that spans hours or days, which has to survive restarts rather than completing inside one session.

M

15 terms

Mandatory field coverage Loupe: The share of required fields actually completed, measured per form and per population.
Mean reciprocal rank (MRR)Compass: A measure that rewards finding the correct passage near the top rather than merely somewhere in the list.
Media relay Evio: The component that carries live encrypted audio between a caller and a voice agent, passing the stream through without storing any of it.
Metadata extraction Compass: Deriving the attributes of a document, such as its owner and its type, which is what makes filtering possible.
Microlearning Atlas: Training delivered in short units that a person can complete between other work rather than in a scheduled block.
Model card Governance: A standard description of a model, covering the uses it is meant for and the limits it carries.
Model Context Protocol (MCP)Compass: An open way of exposing tools and data to a model, so an integration is declared once rather than per application.
Model drift Governance: A model's performance degrading over time as the world it was evaluated against moves on.
Model fallback Governance: The defined alternative when the preferred model is unavailable or declines.
Model routing Governance: Selecting a model per task rather than standardising on one, so the choice can change without a rebuild.
Moderation Prism: A second reading applied to a sample or to borderline cases, to confirm the standard held.
Multi-hop retrieval Compass: Answering a question that requires joining two or more documents rather than finding one passage.
Multi-turn conversation Evio: An exchange where each reply depends on what came before rather than on the latest utterance alone.
Multilingual dubbing Atlas: Delivering the same module in several languages from one approved source, with the meaning preserved.
Multilingual form assistance Loupe: Guiding an applicant through a form in their own language while the submission itself stays in the official one.

N

4 terms

Natural language understanding (NLU)Evio: Converting what a person says into structured meaning, such as an intent and the details that go with it, so a system can act on it.
NIST AI Risk Management Framework Governance: A voluntary United States framework for identifying and managing AI risk, frequently referenced in procurement.
Normalisation Prism: Adjusting for systematic differences between reviewers or panels so that scores remain comparable.
Number masking Evio: Connecting two parties without either seeing the other's number, which privacy regulation increasingly expects.

O

5 terms

Observability for AI systems Governance: Instrumentation that records what happened inside a request, not only whether it succeeded.
Optical character recognition (OCR)Loupe: Converting an image of text into machine-readable characters, the oldest step in the chain and still a common failure point.
Optical character recognition fallback Forge: Reading values as pixels when a system offers no other route to them.
Order effect Prism: The influence of position in the pile on the score a submission receives.
Outbound dialer Evio: The system that places calls from a list, with pacing and retry rules attached under a compliance policy.

P

17 terms

Pairwise comparison Prism: Judging two submissions against each other, which people do more reliably than assigning absolute scores.
Panel review Prism: Several reviewers assessing the same submissions, with a defined method for resolving disagreement.
PDF parsing Compass: Recovering readable text and table structure from PDF files, which record where characters appear on a page and very little about what they mean.
Permission trimming Compass: Filtering results against the asker's entitlements before anything is shown or summarised.
Personal data masking Governance: Removing or replacing identifying details before any content reaches a model.
Phoneme Evio: The smallest unit of sound that distinguishes one word from another, and the level speech models actually work at.
Policy version pinning Atlas: Recording which revision of a source document a module was generated from, so the two cannot quietly diverge.
Post-call summary Evio: A structured account of what happened on a call, written into the system of record rather than left in a recording.
Pre-fill Loupe: Populating fields from what the organisation already holds, which reduces both effort and error.
Private model endpoint Governance: A model served so that prompts and outputs remain inside your boundary and are not used for training.
Privileged access management (PAM)Forge: Controlling and recording the use of elevated access, including by automated actors.
Process mining Forge: Reconstructing how a process actually operates from system event logs, rather than from how it is documented.
Proctoring Atlas: Supervision of an assessment, whether by a person or by software, so that the result means something.
Progressive disclosure Loupe: Presenting a complex task in stages so the applicant is never facing all of it at once.
Prompt injection Governance: Instructions hidden in content a model reads, intended to redirect its behaviour.
Prosody Evio: The rhythm and intonation of speech, which is why a correct sentence can still sound wrong.
Public switched telephone network (PSTN)Evio: The traditional telephone network, which a voice agent reaches through a carrier rather than directly.

Q

2 terms

Query rewriting Compass: Reformulating what a person asked into what the index can actually answer.
Queue depth Forge: How much work is waiting, which is the earliest signal that capacity and demand have parted company.

R

23 terms

Ranking and rating Prism: Two different questions, one asking which is better and the other asking how good each one is.
Reasoning trace Prism: The recorded path from evidence to score, which is what makes a decision reviewable afterwards.
Recertification cycle Atlas: The interval at which a qualification or clearance has to be demonstrated again rather than assumed.
Reciprocal rank fusion (RRF)Compass: Merging the ranked results of several retrieval methods by their positions in each list, so methods with incomparable scores can be combined fairly.
Red teaming Governance: Adversarial testing before deployment, carried out by people trying to make the system fail.
Refresher training Atlas: A shortened repeat of material already covered, issued when a policy changes or a cycle comes due.
Refusal and abstention Compass: The system declining to answer when the evidence is not there, which is a feature rather than a shortfall.
Regression suite Governance: The tests that confirm a change has not broken behaviour that previously worked.
Regulatory training Atlas: Training an organisation is obliged to deliver and evidence, where being a quarter behind is an exposure.
Release registry Governance: The store of signed, versioned software releases and their bills of materials, from which a client environment pulls only verified updates.
Reranking Compass: A second pass that reorders retrieved passages by relevance, which is usually what separates a demonstration from production.
Resubmission Loupe: An applicant returning a corrected submission, and the loop a well-designed form avoids entirely.
Retrieval augmented generation (RAG)Compass: Answering from documents retrieved at the moment of the question, so the answer reflects your material rather than the model's memory.
Retrieval evaluation Compass: Measuring whether the right passages were found, separately from whether the answer read well.
Retry policy Forge: The rules governing when a failed step is attempted again, how often, and when it stops.
Reviewer fatigue Prism: The decline in consistency that follows sustained review, which is a human limit rather than a failing.
Reviewer queue Loupe: The backlog of submissions awaiting a human decision, and the order they are presented in.
Reviewer throughput Loupe: How many submissions a reviewer completes in a period, at a stated standard.
Rework rate Loupe: The share of submissions that have to go back, and the cost that follows each return.
Right of appeal Prism: The applicant's entitlement to have a decision reconsidered, and the record that makes reconsideration possible.
Robotic process automation (RPA)Forge: Software that repeats a recorded sequence of interface steps, which works until the interface changes.
Rollback Forge: Returning a system to its previous state after a change, which has to be planned rather than improvised.
Rubric Prism: The stated criteria a submission is scored against, written down before any scoring starts.

S

37 terms

Schema-constrained extraction Loupe: Requiring the output to match the structure of the field it is destined for, so nothing arrives in a shape the system cannot accept.
Score distribution Prism: The shape of the scores awarded, which reveals a rubric that is compressing or failing to discriminate.
Score drift Prism: The standard moving over a long review cycle, so that the first submission and the four hundredth are judged differently.
Scoring dimension Prism: One axis of a rubric, assessed independently of the others so that a weakness is visible rather than averaged away.
SCORM Atlas: The long-standing interchange standard that allows a course to be delivered and tracked by any compliant platform.
Screen scraping Forge: Reading values off a rendered screen, the brittle predecessor to structured extraction.
Selector Forge: The expression that identifies an element on a page, and the thing that breaks when the page is redesigned.
Self-healing selector Forge: A selector that recovers when an element moves or is renamed, instead of stopping the workflow.
Semantic deduplication Prism: Detecting and removing items that say the same thing in different words, using similarity of meaning instead of matching text.
Semantic search Compass: Finding material by what it means rather than by the words it happens to contain.
Sentiment analysis Evio: Inferring how a caller felt, useful in aggregate and much weaker on any individual call than vendors usually admit.
Service account Forge: A non-human identity with its own entitlements, used so that automated actions are attributable.
Service desk Forge: The system and function that records cases and tickets with their ownership and service levels, which automation updates and closes.
Service level agreement (SLA)Forge: The commitment an automation is held to, expressed in time and completeness rather than availability alone.
Session border controller (SBC)Evio: The device at the edge of a voice network that handles call admission and interoperability.
Session Initiation Protocol (SIP)Evio: The signalling standard that sets up and tears down a voice session between endpoints.
Session state Compass: The technical record of where a conversation or task has reached, so it can resume correctly.
SetFit intent classifier Evio: A small, fast classifier trained from a few labelled examples per intent, used to recognise common requests on a call without a large model.
Shortlisting Prism: Reducing a large field to a reviewable one, where consistency matters most because volume is highest.
Signed release Governance: A release whose authenticity and integrity can be verified before it is admitted to an environment.
Simulated execution Forge: Executing a workflow with its effects suppressed, to confirm what it would do before it does it.
Slide and deck extraction Atlas: Pulling the text and its intended order out of a presentation, which is where most enterprise training already lives.
Software bill of materials (SBOM)Governance: The itemised list of components inside a release, which is how a known vulnerability is traced to what contains it.
Spaced repetition Atlas: Returning to the same material at widening intervals so that it is retained rather than recognised once.
Speaker diarisation Evio: Separating an audio stream by speaker, so a transcript records who said which part.
Speech synthesis Evio: The formal term for generating speech, covering the voice model as well as the audio produced.
Speech to text (STT)Evio: The same task as automatic speech recognition, named the way most product teams search for it.
Standard operating procedure (SOP)Atlas: The document that states how a task is to be carried out, and the source a training module is generated from.
State machine Forge: An explicit model of the states a task can occupy and the transitions allowed between them.
Straight-through processing Forge: A transaction completing end to end without a person touching it, the benchmark in banking operations.
Streaming spoken response Evio: Generating and speaking a reply in pieces, so the listener hears the first words while the rest of the response is still being produced.
Structured evaluation Prism: Scoring against defined criteria with the reasoning recorded, as opposed to an overall impression.
Subject matter expert review loop Compass: The named people who check answers in a defined cycle, and the record of what they corrected.
Subjective answer grading Atlas: Scoring free-text answers against a rubric, so open questions can be assessed consistently without a trainer reading every response.
Submission receipt Loupe: The evidence an applicant holds that they filed, with a timestamp and a reference.
Synthetic voice Atlas: A generated voice used in place of a recorded one, which raises questions of consent and likeness that belong in the contract.
System of record Compass: The authoritative source for a given fact, which an AI layer must agree with rather than compete against.

T

21 terms

Table extraction Loupe: Recovering rows and columns from a document, which is markedly harder than extracting isolated fields.
Tamper evidence Governance: Properties that reveal whether a record has been altered since it was written.
Task mining Forge: The desktop-level equivalent, observing the steps people take to complete work.
Taxonomy Compass: The agreed structure of categories a body of material is organised by, which retrieval leans on more than most teams expect.
Tenancy isolation Governance: Keeping one organisation's data and compute separate from every other organisation's.
Tender evaluation Prism: Scoring supplier bids against published criteria, under procurement rules that expect the reasoning to be produced.
Text to speech (TTS)Evio: Generating spoken audio from written text, with the quality judged on naturalness as much as clarity.
Throughput Forge: Completed units of work per period, measured at the point that matters to the business rather than the system.
Time to competency Atlas: How long a new joiner takes to reach the standard, measured rather than estimated.
Time to first audio Evio: The delay before the caller hears anything, which is the latency figure a person actually experiences.
Time to submit Loupe: How long an applicant takes to complete a submission, measured from first entry to acceptance.
Token budget Governance: The allowance of model input and output for a task, and where the cost of a design actually accrues.
Tool calling Compass: A model invoking a defined capability, which is the point at which answering becomes acting.
Trace and span Governance: The structured record of a request and each operation inside it, which is how a slow or wrong answer is diagnosed.
Traceability Governance: The ability to follow an output back through every input and step that produced it.
Trainer dependency Atlas: The constraint created when delivery requires a scheduled person, which limits how fast a network can be reached.
Training attestation Atlas: A formal confirmation that a named person has been trained on a named policy, which an auditor can ask for.
Training audit evidence Atlas: The report an organisation produces when asked to demonstrate that its people were trained on the current version.
Training coverage Atlas: The share of a population trained on the current version of a given policy at a given date.
Transcreation Atlas: Adapting content so that it carries the same intent in another language, which direct translation often fails to do.
Turn taking Evio: Managing who speaks when, which is the hardest part of making a spoken exchange feel natural.

U

1 term

User interface automation Forge: Acting through the screen because no programmatic route exists, which is common in regulated estates.

V

6 terms

Validation at the point of entry Loupe: Catching an error while the applicant is present, rather than discovering it in review weeks later.
Vector database Compass: A store that holds numerical representations of text and returns the passages closest in meaning to a query.
Vendor evaluation Prism: Assessing suppliers against defined requirements, with the comparison recorded rather than asserted.
Vendor lock-in Governance: The cost of changing supplier later, which is decided by architecture rather than by contract.
Voice activity detection (VAD)Evio: Deciding which parts of an audio stream contain speech and which contain silence or noise.
Voice over Internet Protocol (VoIP)Evio: Carrying voice over data networks instead of dedicated telephony circuits.

W

7 terms

Warm transfer Evio: Passing a caller to a person along with the context of the conversation so far.
Web scraping Compass: Collecting content from websites and stripping away navigation and advertising, so only the substantive text reaches the index.
Weighted rubric Prism: A rubric where criteria carry different weights, which is where an organisation's judgement is actually encoded.
Word error rate (WER)Evio: The share of words a transcription gets wrong, counting insertions and deletions alongside substitutions.
Workflow agent Compass: An agent that carries out a defined business process from start to finish, with a model making the judgement calls inside fixed steps.
Workflow orchestration Forge: Coordinating steps across several systems, with the dependencies and the failure handling defined.
Write-back Compass: Returning a result into the system of record, which is where an answering layer takes on real consequence.

X

1 term

xAPI (Experience API)Atlas: A successor to SCORM that records learning events in more detail and outside a single course player.

Z

1 term

Zero data retention Governance: A commitment that inputs and outputs are not stored after a request completes.

Tell us the number you need to move

Define an Outcome