Glossary
The terms we use, in plain English.
40 terms you will meet in our diagnostic, our Audit reports and our research, each with what it means and why it matters to someone running operations.
A to Z
Glossary
Jump to: A · B · C · D · E · F · G · H · I · L · M · O · P · R · V · Z
A
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Agent | Software that uses an AI model to decide and carry out steps toward a goal, such as reading an email, looking up an order and drafting a reply. | Agents can take on coordination work, but only safely when what they may do is bounded and checked. |
| Agentic workflow | A business process where one or more agents carry out steps, with rules, checks and people around them. | This is the unit we diagnose and build: one workflow, end to end, not a chatbot bolted on. |
| AI depth levels | A four-step scale for how far AI changes a workflow: Level 1 task automation, Level 2 workflow augmentation, Level 3 workflow redesign, Level 4 re-architecture. | It tells you whether you are looking at a quick fix or a change to how the work is organized, and sets expectations for cost and effort. |
| AI-native | Designed from the start around AI agents, deterministic checks and human exception handling, rather than adding AI to an unchanged process. | AI-native operations can grow capacity without growing headcount at the same rate. Bolted-on AI usually can't. |
| AMEA | Autonomous Micro-Enterprise Architecture: our working paper on the runtime architecture, where every agent action is checked before it executes. See Research. | It is the blueprint behind how we build. |
| AMEE | Autonomous Micro-Enterprise Economics: our working paper on why coordination cost can flatten as agents take over coordination work. See Research. | It explains the economic case, and its limits. |
| AMEF | Autonomous Micro-Enterprise Finance: our working paper on capital efficiency, valuation and a five-point technical due-diligence checklist. See Research. | Useful if you invest in, or are valued as, an AI-native business. |
| AMEG | Autonomous Micro-Enterprise Governance: our working paper on rules enforced at runtime, audit trails and mapping to the EU AI Act. See Research. | It shows how governance can be part of the system, not a policy document. |
| AMEI | Autonomous Micro-Enterprise Interface: the umbrella framework that ties the papers together and covers how an AI-native business meets its market. See Research. | It connects the architecture to customers, sales and growth. |
| AMES | Autonomous Micro-Enterprise Security: our working paper on the security boundary around agents, with independent checks mapped to known threats. See Research. | It is our answer to prompt injection and credential theft. |
| Audit trail | A tamper-evident record of every input, decision and action, with who or what made it and when. | When someone asks "why did it do that?", you have an answer. Regulators and auditors will ask. |
| Automation coverage | The share of a workflow's cases that run start to finish without a person touching them. | A realistic coverage target matters more than a headline automation claim. |
B
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Baseline | The measured state of a workflow before any change: workload, cycle time, exceptions and verification effort. | Without a baseline, no one can show whether a change helped. The Audit measures it. |
C
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Confidence level | How strongly the available evidence supports a finding in the diagnostic Snapshot. | A High leverage signal with low confidence means "worth measuring", not "proven". |
| Coordination tax | The time and cost spent passing work between people and systems: chasing, checking, re-keying and waiting. | It grows with headcount and is often the largest hidden cost in an operation. |
| Cycle time | How long a single case takes from start to finish. | Customers and cash flow both feel cycle time directly. |
D
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Deterministic orchestration | Running agents inside fixed, rule-based control flow, so the sequence of steps and allowed actions is set in advance. | It stops agents drifting off-task in open-ended loops and keeps cost predictable. |
E
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| EU AI Act high-risk system | An AI system used in an area the EU AI Act lists as high-risk, such as hiring or credit decisions. These carry extra obligations, including risk management, logging, transparency and human oversight. | Most operational workflows are not high-risk, but some are. Knowing early avoids a rebuild later. This is not legal advice. |
| Evidence intake | The structured 10-section record a client completes at the start of the Operational Leverage Audit. It is locked on submission, and corrections become amendments. | It makes the Audit's baseline traceable to what you told us. |
| Exception rate | How often a case has to leave the normal path and go to a person, and why. | Exceptions are where cost and risk concentrate. A high rate can be the opportunity or the blocker. |
F
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Friction map | A view of where a workflow loses time or quality across five stages: Demand/Input, Processing, Decision, Execution, Verification. | It shows where to intervene first, instead of automating the wrong step. |
| FTE | Full-time equivalent: workload expressed as the number of full-time people it would take. | It turns scattered hours across a team into one comparable number. |
G
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Guardrails | Automated checks on what goes into and comes out of an AI model: format, policy, safety and factual consistency. | They catch bad outputs before they reach a customer or a system, and avoid paying for silent retries. |
H
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Human exception handling | A defined path that hands a case to a named person when an agent cannot complete it within its rules. | Agents should never guess their way through an edge case. |
I
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| IVT | Invalid traffic: ad impressions or clicks that are not from genuine human interest, such as bots and fraud. | In advertising, IVT wastes budget and damages trust. We have built real-time IVT detection. |
L
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Leverage signal | The diagnostic's rating of an opportunity: High, Medium or Investigate. | It ranks where to look first. It is not a savings forecast. |
M
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Model Context Protocol (MCP) | An open standard for exposing business tools and data to AI agents through a common interface. | It reduces custom integration work and makes switching models easier. |
| Model routing | Sending each request to the cheapest model that can handle it, with fallback if one fails. | It keeps compute cost bounded and avoids dependence on a single provider. |
O
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| OpenRTB | The IAB Tech Lab standard for real-time bidding between buyers and sellers of digital advertising. | It is the plumbing of programmatic advertising and where much of our production work sits. |
| Operational leverage | Producing more output from the same team and cost base. | It is the outcome we diagnose: more capacity without matching headcount. |
| Operational Leverage Audit | Our paid, fixed-fee engagement on one workflow. It produces a measured baseline, future-state design, verification architecture, benefit model and a go / no-go recommendation. | It answers whether a workflow is worth changing before you commit to building anything. |
| Operational Leverage Snapshot | The output of the free diagnostic, on screen and as a PDF: top opportunities, friction map, AI depth, what we know and don't know, and what to measure next. | A free first read on one workflow. It does not promise savings. |
| Orchestration | Coordinating the steps, agents, tools and people in a workflow so work moves in the right order. | Orchestration, not the model, decides whether an AI system is reliable. |
P
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Pilot | A fixed-scope build of one Audit-identified workflow, run end to end in your environment and measured against the Audit baseline. | It proves results on real work before you scale. |
| Prompt injection | An attack where hidden instructions in an email, document or web page try to make an AI agent do something it shouldn't. | It is why agents should never hold keys or act without independent checks. |
R
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Retrieval (RAG) | Retrieval-augmented generation: fetching only the relevant documents or data for a request and giving them to the model, with sources attached. | Answers can be traced to source, and smaller prompts cost less. |
| Revenue per core employee | Revenue divided by the number of core staff running the business. | It shows whether the business is growing faster than its headcount. |
V
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Verification burden | The time people spend checking work, whether their own, a colleague's or a system's. | Heavy manual checking is often where AI-native redesign frees the most time, if the checks can be made explicit. |
| Verification gate | A rule-based check every proposed action must pass before it commits to money, data or a customer. | It is what makes autonomy safe enough to trust. Anything that fails is stopped and handed to a person. |
Z
| Term | Plain-English meaning | Why it matters to an operator |
|---|---|---|
| Zero-trust boundary | A security design where agents never hold payment or system keys, and every action is checked independently, whatever the agent says. | Even a fooled or faulty agent cannot move money or change systems on its own. |
The AME papers are working papers. Their performance figures come from synthetic workloads (100,000 synthetic transactions), not client deployments.
Next step
See these terms applied to your own workflow.
The free diagnostic takes 10 short steps, about 8–12 minutes, and returns an Operational Leverage Snapshot with leverage signals, a friction map and the indicated AI depth for one workflow.