立即咨询
全部文章
术语表 · 318 个术语

软件工程术语表

来自我们文章的术语，为高管和业务团队而非工程师撰写——阅读释义、查看示例，并用右侧问题与开发团队沟通。

术语 是什么 通俗示例 高管应问开发团队的问题
A/B Test Comparing two ways of working by running each one with a similar group. One team uses the new workflow while another keeps the old one. How do we know the two groups are comparable, and how do we keep customers from being affected?
Abstraction Layer A middle layer between our system and outside services, so we can switch providers easily. Switching AI model providers by changing a single layer. If we switch providers, how many places in the system have to change?
Acceptance Criteria The conditions that confirm a piece of work is finished. They must be observable or testable and agreed before delivery, which heads off a vague “it works well” that each side reads differently. The system exports files in the agreed format. Can this criterion be tested?
Access Control Controls who can see which data. One team sees only the manuals for its own site. How are permissions inherited, and how are they revoked when someone changes roles?
Access Review A regular review of who should still have access to what. Checking every quarter that accounts no longer in use have been closed. How often do we review access?
Accessibility Designing so that a wide range of people can use the product. It covers color, type, keyboard use, screen readers and plain language, and it is designed in from the start and tested with real users, never bolted on later. Support for screen readers and large text. When AI changes the interface, does it still meet the standard?
Action Something the system actually does, as opposed to only replying with text. The system creates a quotation and emails it to the customer. Which actions can be undone, and which cannot?
Adaptive UI A screen that changes with the context according to set rules. The parts that adapt should be limited and have defaults, so users can still predict where the main features are and how the product behaves. New users see extra guidance. Which parts can adapt, and which must stay fixed?
Adoption Analytics Data on how people actually use the system. Tracking which workflows people come back to, beyond counting logins. Are we measuring usage, or the value that usage creates?
Agent AI software that takes a goal and carries out several steps until the job is done, instead of answering one question at a time. The agent reads a request, checks the data, then drafts a document for a person to approve. What can this agent do on its own, and where does it stop?
Agent as Tool Having an agent handle a subtask while a manager agent stays in control. The specialist agent is wrapped so it can be called like a tool with clear inputs and outputs, which cuts down on uncontrolled back-and-forth between agents. A research agent sends its findings to the manager. How are the partial results checked before they are combined?
Agentic Product A product in which AI plans the steps and uses tools within set limits. The AI plans or carries out several steps under tools, rules and monitoring, so it does far more than a chat screen that produces text. An agent prepares an appointment and asks a person to confirm it. Who owns the outcome and sets the boundaries?
AI Agent AI software that takes a goal, uses data or tools, and works through several steps. An agent reads a piece of work, checks it against a checklist and passes it to a reviewer. Which data and tools does the agent use, and where are its limits?
AI Literacy The basic understanding of AI that every employee should have. Knowing which data must never be entered, and what to check before using the output. Does everyone know the basic rules about data yet?
Alert A notification that something is wrong. Every alert should have someone responsible for acting on it. An alert when spending goes over the set budget. If an alert fires and no one acts on it, how would we know?
Alert Budget A cap on how many alerts go out each day, so every one of them means something. No more than 5 alerts per shift. If alerts go over budget, which ones does the system drop first?
Analytics Tools that collect and read how people behave on a website or app, showing where they come from, what they do and where they drop off. Seeing that 70% of people leave a product page before placing an order. What are we measuring, and which numbers do we actually use to make decisions?
Anomaly Detection Finding values that differ from the normal pattern. Spotting vibration that is out of line with other readings under the same load. Which kinds of anomalies actually lead to action?
API A standard channel that lets systems exchange data. Pulling customer data from the CRM into a brief. Which systems can we connect to, and are there limits on permissions or data volume?
App Intent An app capability the system can call using natural language. It is a channel through which outside apps or systems trigger defined tasks, so the input, permissions and expected result must be spelled out clearly. Asking it to open the next inspection job. Which actions should be open to being called?
Approval Sign-off from someone with authority before work continues. A discount above the cap waits for a manager's approval. What needs approval, and if no one approves, how long does it sit?
Approval Flow The approval path a request follows before a transaction goes through. The flow must name the approver by value, risk or exception, and record the reasons and time of each decision for later audit. A supervisor confirms the company's shopping cart. Which spending limits need whose approval?
Approval Workflow A defined approval path that sets who must review what, and when. Low-risk work needs sign-off from just one supervisor. Do we route approvals by risk, or use a single path for everything?
Asset Library A searchable library of creative assets that can be reused. Approved images and copy are stored for other teams to use. Can we find old assets, and how do we know they are still usable?
Audit Log A record of who, or which system, did what and when. Tracing which version of the data a proposal used. How far back do we need to be able to look?
Authentication Confirming who a user is before letting them in, for example with a password, a code sent by text message or a face scan. A patient enters a phone number and a code received by text to open an appointment slip. Which user groups need two-step verification, and how do people recover an account if they forget the password?
Authorization Rules that set which data each user can view or edit once they have signed in. Patients can open only their own appointments, and staff can open records only for the branch they belong to. Are permission rules checked on the server every time data is fetched, and who tests them?
Automated Test A set of scripted checks that confirm key functions still work correctly, and that run on their own every time the code changes. A test simulates two people reserving the same item at once and checks that only one of them succeeds. Have the cases that broke in the live system been added to the test suite?
Automation Having the system repeat steps according to set conditions. Creating tasks after a meeting without retyping anything. If the system gets something wrong, how does it stop and roll back, and who gets notified?
Automation Champion Someone on a frontline team who drives and looks after that team's automations. An employee who knows which task should be automated next. Does the team have someone who owns its automations?
Autonomy How much freedom the system has to act. It should be set separately for each action, because one system might reply automatically yet still need a person's approval before changing data or making a transaction. Handling standard cases without waiting at every step. What evidence must be in place before we give it more freedom?
Backlog A list of work or ideas waiting to be done, in order of priority. Screened ideas waiting for the next trial round. Who prioritizes the backlog, and on what criteria?
Backup A copy of the data stored apart from the main system, used to recover when the real data is damaged or lost. The order database is copied to another location every night and kept for thirty days. Where are the backups stored, who can access them, and when did we last test a restore?
Barcode A striped code on a product that is scanned instead of typed, which cuts number-entry errors and speeds up stock counts. Warehouse staff scan incoming goods instead of typing 13-digit codes. How will we handle items that don't already have a barcode?
Baseline The figures measured before the system changes. Average time to close a case before using AI. Does the pre-trial data represent normal work?
Bottleneck The point that slows the whole process because it can only handle so much work. Every job waits for one person to approve it. Where is the bottleneck right now, and how are we measuring it?
Brief The starting document that sets out the problem, the target audience and what must be delivered. A one-page document every department refers to during a product launch. Is every team working from the same brief?
Business Brief A short document that sets out a project's problem, users, scope and success measures. A one-page document sent to every developer to base their quote on. Is every vendor quoting on the same brief?
Business Case The business reasoning for a project, with its costs, expected results and risks. The document used to decide whether to approve the budget. What are the key assumptions in this business case?
Business Rule A business condition, written down separately from the code, that the system uses to make decisions. Discounts above 15 percent need a manager's approval. If the rule changes, who can update it and how long does it take?
Capacity The most work a team or system can take on in a given period. The team can handle 200 items a week. If work goes up 30 percent, what do we need to add?
Capstone A real project used to prove a skill. Cutting the time it takes to produce a report while the quality still passes. Which piece of work proves the skill can be used in practice?
Capture Value Turning the time or cost saved into a real business result. Using the time saved to take on more work, instead of letting it slip away. What are the saved hours being used for?
Career Path The growth path for a role, which turns learning into an actual position. People who complete a capstone join a cross-department project. Which position does the new skill lead to, and who approves it?
Change Management Getting people and processes ready for a new system. Training staff and updating SOPs before go-live. Who is responsible for making sure people actually use it, beyond the system being finished?
Chatbot A system that handles recurring customer questions in place of a person, and hands off to staff when a question goes beyond its set limits. It answers price and stock questions at 2 a.m., then passes the customer to an admin when they are ready to order. In which cases must the bot hand off to a person right away?
CI/CD An automated assembly line that tests code and puts it live using the same steps every time, in place of manual work. Every time the code changes, the system runs the test suite on its own, and only if it passes is the code sent to the test server. How long does it take from finishing a code change to it going live, and which steps are still manual?
Citation Showing where an answer came from. Showing the manual's name, version and page. Can users open the original document and see its version?
CMMS A system for managing maintenance work. Creating a work order from an alert. How will an alert become a work order without creating duplicate jobs?
CMS A back-office system that lets the team edit content, images and prices on the website themselves, without calling a developer. The marketing team puts up a new promotion on its own in ten minutes. What can the team change on its own, and what still needs a developer?
Code Review Having another engineer read code before it is merged into the system, to catch mistakes and keep to one standard. Billing code written by AI is read by a second engineer, who asks about stacked discounts before it goes live. Who reviews the code that touches money and customer data, and does that happen every time?
Computer Vision AI that analyzes images or video. Detecting marks on a part. Has it been tested under real lighting, angles and speeds?
Confidence Score How confident the model is in its result. Low scores are sent to QC for checking. At what score does a case go to a person for review, and why?
Context The surrounding information a system needs in order to answer correctly. Knowing what this customer has already bought. Does the system know enough context to answer correctly yet?
Control Plane A central system that knows which automations exist, who looks after each one, and how much access and budget each uses. A register of agents with their permissions, costs and a stop button. Do we have one place today that can tell us everything that is running?
Conversion Event An event that counts as a business result. Choose events that reflect real outcomes, such as a booked appointment or a complete submission, in place of simply counting chat openings and clicks. An appointment is booked or all documents are submitted. Are we measuring conversations, or actual results?
Conversion Rate The share of visitors who become customers or complete the target action. What percentage of people who click on a recommendation actually buy. From which point to which point are we measuring, and how do we remove repeat users?
Cost Control Mechanisms that limit and track what an AI system spends. Setting a monthly budget for each agent, with an alert as it nears the limit. If spending spikes unexpectedly, who finds out and how quickly can they stop it?
Cost per Transaction The average cost of handling one item of work. The cost of issuing one invoice. Which costs does this figure include, and which does it leave out?
CPQ A system for configuring products, pricing them and producing quotations. CPQ puts options, prices and exceptions under one set of rules, which reduces offers that can be sold but cannot be delivered. Choosing a package and having the price calculated by the rules. Where do the pricing rules come from?
CRM A system that keeps customer data and contact history in one place. Short for Customer Relationship Management. Sales staff can see everything that has been discussed with this customer. Is the master copy of customer data in the CRM, or still in each person's files?
CRM Integration Connecting data with the customer relationship system. A good connection sets the direction of data flow, who owns each record and how conflicts are resolved, which takes more than sending data across once. Proposals and next steps are logged automatically. Which data must never be overwritten?
Cross-functional Team A team that brings together people from several functions to work toward one goal. A launch team with people from product, marketing and sales working together. Who owns this team's overall results?
Customer Journey The whole path a customer travels, from first hearing about you through buying and using the service. From seeing an ad, to asking the price, buying, and reporting a problem. Where along this path do we lose the most customers?
Cutover The moment of switching from the old system to the new one for real. On Saturday night, orders stop for two hours while outstanding balances are moved, and the new system opens on Sunday morning. If cutover night goes wrong, up to what time can we roll back, and who makes the call?
Cycle Time The total time from when work comes in until it is delivered, waiting time included, as opposed to only the time spent working on it. A quotation takes 40 minutes of actual work, but the cycle time is 2 days. Do we measure cycle time automatically from the system, or have people enter it by hand?
Dashboard A screen that brings data together to support decisions. A supervisor sees all work at risk of missing its SLA on one page. What does the user need to decide after looking at this screen?
Data The information a system works with, including customer data, prices, stock and work history. Product prices and stock status that the system pulls in to answer customers. Where does the system's data come from, and who keeps it accurate?
Data Access Permission that sets which data sets a system or person can reach. The agent can read prices but cannot see employee data. Which data can this system access, and when are its permissions reviewed?
Data Classification Sorting data into tiers by sensitivity, to set how each tier may be stored, sent and used. The cost price table is classified as confidential, so it may be used only with internal AI. Which types of document have no tier yet, and who decides when it is unclear?
Data Freshness How current the data is, and whether it is recent enough to base decisions on. The figures on screen were last updated 10 minutes ago. How many hours old can data be before it is no longer acceptable?
Data Mapping A table showing which field in one system matches which field in another. The sales team's product codes are matched one by one to the warehouse's product codes. Who owns this table, and who adds new products to it?
Data Migration Moving data from the old system into the new one, including cleaning it up and checking it after the move. Moving eight thousand customer records into the new system after merging the duplicates. After the move, how do we check that the data is complete and the outstanding balances match the old system?
Data Model The structure that sets what data exists and how the pieces relate to one another. A position has several tasks, and each task has its own time and risk. If we want to add a new dimension later, will we have to rebuild the system?
Data Pipeline The route that carries data from its source to where it is used. Sensor → database → dashboard How does the system cope when data is missing or arrives late?
Data Readiness Whether the data a system needs is ready, in both completeness and accuracy. Whether customer records have every field filled in, or are still entered inconsistently. If the data isn't ready, what should we do first?
Data Residency Requirements on which country or region data must be stored and processed in. A client contract states that project data must stay on servers in Thailand. Which countries is our data processed in, including when it is sent to AI?
Data Retention The policy on how long data is kept and when it is deleted. Customer conversations are kept for the period that the law and company policy require. What do we keep and for how long, and who approves this policy?
Decision A point where one path must be chosen, with consequences that follow. Approving a discount or turning it down. Who can make this decision, and where is it recorded?
Decision Intelligence Using data and systems to improve the quality of decisions. It brings data, models and the decision process together, while the person responsible still applies judgment and answers for the outcome. Preparing scenarios before choosing capacity. Which decisions does the system help us make, as opposed to what it merely displays?
Decision Log A record of what was chosen and why. It should keep the options, evidence, assumptions, owner and review date, so the organization learns from real results without relying on memory. Comparing actual results with the original assumptions. Who can access and edit the log?
Decision Record A record of what was decided, on what data, and by whom. Being able to look back and see why that path was chosen that month. Could we explain a decision we made three months ago?
Decision Rights A clear statement of who can decide which matters on their own, and which matters need approval. A supervisor can approve discounts up to a set cap, and anything above it goes to an executive. Who can make this decision, and where are the limits?
Decommission Retiring an old system or method according to a plan. Shutting down the old Excel file once the new workflow is stable. When will we switch off the old way, and is there a rollback plan?
Dependency Scan Checking the ready-made code packages a system uses for versions with publicly disclosed vulnerabilities. The tool flags a vulnerability in the package that handles file uploads, so the team updates it before going live. Does this scan run every time the code changes, and who receives the alerts?
Disaster Recovery The plan and systems for restoring service after a major incident, such as a data center outage or data being encrypted by an attack. When the cloud provider goes down across a whole region, the team brings the system up from a copy in another region. Which kinds of incidents does this plan cover, and who is needed to set it in motion?
DLP Tools that detect sensitive data and stop it from being sent outside the organization. Short for Data Loss Prevention. The system raises an alert when someone pastes a large number of national ID numbers into an outside website. Which data tiers do the rules cover, and how often do they raise false alarms?
Dynamic Profile A set of instructions or tools that changes with the user's status. The profile holds context that changes over time, so it needs a known source, freshness, permissions and a way for users to correct anything the system has misunderstood. Turning on the planning tool when the user picks trip mode. Who defines the profile?
Edge Computing Processing data close to the machines instead of sending everything up to the cloud. Analyzing images at the production line to reduce delay. Which tasks need an immediate response, and why process them on site?
Embedding Turning content into numbers so it can be searched by meaning. A search for “machine running hot” finds documents about high temperatures. Does search actually work for Thai-language data and specialist terms?
ERP The back-office system that brings accounting, stock, purchasing and production together. Short for Enterprise Resource Planning. Issuing a purchase order automatically updates stock and the accounts. How will the new system connect to our existing ERP, and who looks after that connection?
Error A result that differs from what it should be. Errors must be counted and sorted by type so they can be fixed at the right point. The system pulls the price of the wrong model into a document. What is the error rate right now, and which kind of error happens most often?
Escalation Passing a case to a person when it goes beyond the system's limits. Hand-offs should be triggered by checkable rules, such as risk, low confidence or a lack of authority, with a named receiving team and its SLA. Electrical problems go straight to a technician. Do the risk conditions cover everything they need to?
Evaluation Testing how accurately a system answers, using a set of examples whose correct answers are already known. Running 200 questions and counting how many it gets right. Who decides which answers are correct, and how often do we test?
Evaluation Set A set of examples used to test quality again and again. It must include normal cases, missing data, exceptions and attacks, along with answers or criteria that experts accept. Normal cases, risky cases and cases with incomplete data. Does it cover the exceptions we actually see?
Event Tracking Recording the key events in a system. Logging when work is received, approved and closed. Which events do we need in order to measure outcomes, beyond usage?
Exception An item that falls outside the normal criteria, so a person has to decide in place of the system. A discount request above the cap is passed to the manager. What percentage of our work is exceptions right now?
Exception Queue A queue that collects unusual cases for a person to decide. Invoices with mismatched totals are sent to a review queue. Which exceptions matter most, and what is their SLA?
Exception Rate The share of items the system cannot finish on its own and that need a person to look at. 10 percent of requests are sent to a person to decide. What is the exception rate right now, and what is causing it?
Exit Plan A plan for leaving a provider or system without disrupting the business. Keeping our data and prompts on our side so we can move. If we stop using this vendor, what can we take with us?
Experiment Dashboard A screen that brings together the results of every experiment so they can be compared. Seeing every idea that has been trialed, with its results, in one place. Are experiment results kept together, or scattered across teams?
Explainability Making the reasons and evidence behind a system's result visible. A useful explanation shows which data and rules shaped the recommendation and states its limits, instead of offering a plausible-sounding reason made up afterward. Opening up the KPI sources and the calculation. Is the explanation good enough to check, or does it just sound good?
Failover Switching automatically to a backup machine or system when the main one stops working. The main database goes down and the system switches to the backup within a minute, with nothing for customers to do. Has the switch ever been tested on the live system, and is any data lost during it?
Failure Mode A way in which a system or process tends to break or give wrong results. The system gives wrong answers when customer data is incomplete. Do we know yet how many ways this system can fail?
Fallback The backup method when AI or a tool fails. The fallback must preserve the work and its context, for example by switching to a manual flow or handing off to a person. Showing a “system error” message is not enough. Sending the work to a human queue without losing any data. Has the team rehearsed the fallback?
False Accept When the system lets a defective item or a wrong case through. A defective part reaches the customer. Which kind of error costs our business more?
False Alarm An alert that turns out not to be a real problem, often enough that people stop paying attention. Ten alerts a day, and only one of them is real. What is the false alarm rate, and who adjusts the thresholds?
False Reject When the system rejects a good item for no reason. Good items are thrown out, which drives up costs. Have we set the threshold based on the cost of both kinds of error?
FAQ A set of questions customers often ask, with standard answers. Short for Frequently Asked Questions. Questions about returns that get the same answer every time. Do the FAQ answers match what the team actually tells customers?
Feature Flag A switch that turns a feature on or off without releasing a new version. Turning on a new feature for a small group of customers first. If a feature has problems, how quickly can we turn it off, and who does it?
Feed A file or data stream sent to a destination on a regular cycle to keep product information up to date. Sending the product list to outside channels every hour. If the feed fails, how long does the destination keep showing old data?
Feedback Information coming back from users or real results, used to improve the system. An employee marks an answer as unusable and explains why. Do we feed people's corrections back into the system, or let them disappear?
Fine-tuning Training an AI model further on our own data so it answers with the organization's own wording and specialist knowledge. Training on real chat history so the bot answers in the same tone as the sales team. Which data set is used for training, and how is customer data protected?
FinOps Managing system costs so they are visible and controllable by department or by piece of work. Setting a monthly budget for each agent, with an alert as it nears the limit. What does a single run of the work cost, and who is responsible for it?
Flow The path that work or data travels from one point to another. From a customer's request through approval to delivery. Where along this path does work get stuck most often?
Gate A checkpoint that work must pass before moving to the next step. A document must pass a data check before it goes to the customer. If work fails the checkpoint, what does the system do next?
Go Criteria Criteria agreed in advance for what it takes to move forward. Ten customers must confirm before full development begins. Were the go criteria set before we started, or agreed afterward?
Golden Set A standard set of examples used to retest the system every time it changes. 50 sample cases run before each new version is released. Do we have our own test set, or do we use the vendor's?
Governance The rules on who decides what, how approvals work and how things are checked. A small working group that approves rollouts every month. Who has the authority to stop or expand a project?
Ground Truth The correct answers used as the standard for measuring a system's results. A set of cases whose answers experts have already confirmed. Who decides which answer is correct?
Guardrail A rule that checks or stops the system's work. A guardrail can be a rule applied before the work, a check on the result afterward, or a condition that stops the work and calls in a person, so guardrails need several layers designed around the risk. Never quote a price that is not in the price table. Who owns the rules, and when are they reviewed?
Guided Generation Keeping AI output inside a set structure. The system gives the AI a template, options and rules before it produces anything, which keeps quality consistent and leaves room for people to check the points that matter. Returning the plan as a list of days and activities. Does the schema handle cases where data is missing?
Handoff Passing work from AI to a person along with its context. A good handoff carries the facts, the evidence, what has already been tried and why it is being passed on, so the person taking over can decide right away. Sales receives a summary without having to ask the customer again. Which conditions require an immediate handoff to a person?
Handoff Contract An agreement on exactly what information must be attached when work is passed on. Passing a case to the next team along with the customer's details and what has already been tried. Have we set out the minimum information required for a handoff?
Handover Handing over a system, its documents and the know-how to the team that will look after it. Handing over the manual, the system architecture and all access rights. After handover, who looks after it, and on what terms?
Headless CMS A CMS that keeps content management separate from the web pages, so one set of content can go to the website, the app and the screens in each branch. Change a price in one place and it updates on the website and the app at the same time. How many channels does one set of content need to appear on?
Human Approval A point where a person must approve before the system continues. A discount above the cap waits for a manager to approve it. Which actions must never happen without a person's approval?
Human Oversight People keep meaningful oversight of the system's results, beyond having their names on paper. Spot-checking the system's results every week and recording the findings. Do the people providing oversight really have enough time and information?
Human Review Having a person check results before they are used, at points where the risk is high. Checking a price proposal before it goes to a major customer. What percentage do we check, and how do we choose what to check?
Human Touch The number of times a person has to handle one item of work. Down from 6 touches to 1, at the approval step only. How many times does a person touch each item?
Human-in-the-loop Having people check or decide at the key points. AI writes the draft, but the account owner is the one who presses send. Do people need to check every case, or only the risky ones?
Idempotency Making sure a repeated command does not produce a repeated result. This matters when a system retries a command after a timeout, because it prevents duplicates such as creating a purchase order or taking a payment twice. A retry does not create two POs. Is every important action protected against duplicates yet?
Incident An event in which the system goes wrong badly enough to affect real work. The response steps should be prepared in advance. The system sends the wrong email to 200 customers. If something happens, who is responsible, and how do we tell customers?
Integration Connecting a new system with the factory systems already in place. Sending events into the CMMS or MES. What is the data agreement between the systems, and who looks after each side?
Intent The goal a user wants to achieve. In design, the system should tie each intent to the data it needs, the next step and a definition of success, instead of only giving question categories a name. The user wants to book a site survey, beyond searching for the word “survey”. Do we know the main intents from real data, or are we guessing?
Inventory A list of what exists, such as tasks, data or systems. Taking stock of how many automations the company currently runs. Is our list complete, and who keeps it updated?
IoT Devices or sensors that send data over a network. Reading a machine's temperature every minute. Who owns each sensor, and how is it checked and calibrated?
Iteration Improving in short rounds based on measured results. Revising an offer and testing it again the following week. How short are our improvement cycles?
Job Description A document that sets out a position's scope of work, responsibilities and decisions. Used as the starting point for working out which parts of the job a system could take on. Has this document been updated since the process changed?
Kill Switch A button that stops the system immediately when a problem is found. Stopping every agent that sends messages to customers when an error is found. Who has the right to press stop, and how is unfinished work handled afterward?
Knowledge The company's know-how that the system can draw on to answer questions. Manuals, fixes for known problems and past cases. Which documents does the system's knowledge come from?
Knowledge Base An organized, searchable library of knowledge. SOPs, sample work and frequently asked questions in one place. Who signs off on the content, and how is outdated material retired?
Knowledge Card A short unit of knowledge the system can use to give a precise answer. One card for each question customers ask often. Is our knowledge broken into pieces that can be found, or still sitting in long files?
Knowledge Curator The person who keeps the knowledge library accurate, current and easy to search. Checking which manuals are still valid and which should be retired. Who is responsible for the accuracy of the knowledge library?
Knowledge Layer The layer that gathers the company's knowledge so systems can draw on it. A library of approved manuals, rules and cases. Where does the system's knowledge come from, and who approves it?
KPI A performance measure tied to a goal. Measuring the time to close a case, instead of the number of emails. Is this number tied to an outcome, and does everyone define it the same way?
Label The answer attached to an image, used to train and test a model. Marking an image as Good or Scratch. How closely do the reviewers agree, and how are wrong labels corrected?
Landing Page A single web page designed to test an offer and measure the response. A page explaining a new service, with a sign-up button. What are we measuring on this page, and what result counts as a pass?
Latency The wait between a request and the response. It must be measured across the whole journey, beyond the model's own response time, because searching for data, calling APIs and syncing all add to the time the user waits. Voice guidance must respond fast enough for work on the floor. How many seconds of delay make it unusable?
Lead Someone who has shown interest in a product and may become a customer. A person who fills in a form asking for a quote. Within how many hours is each incoming lead followed up?
Lead Time The total time from when a customer or requester starts waiting until they get the result. From a customer asking for a price to receiving the quotation. Do we measure from when the customer starts waiting, or from when we start work?
Leading Indicator A signal that comes before the result. It lets you act before the end result happens, but it must be shown to have a relationship you can actually base decisions on, beyond simply moving sooner. Work waiting for approval, before the SLA is missed. Does this signal really come first?
Learning Platform A system that organizes lessons and tracks people's development. Employees see a learning path based on their role. How does the system connect learning with real work?
Least Privilege Granting the least access needed. Only the permissions needed for that task and for that period are given, which limits the damage when a command is wrong or data is used beyond its scope. The content agent can read data but cannot send email. Are permissions reviewed when roles change?
LLM A large language model that reads and writes human language. It is the engine behind AI assistants and chatbots. Reading a customer's email and summarizing it as a work order for the team. Where does the model run, and is our data sent outside?
Load Test Simulating many users on the system at once, to measure how much it can handle and where it gets stuck. Simulating five thousand customers buying at the same moment, one week before a campaign. At how many users does the system start to slow down, and which part becomes the choke point?
Log An automatic record of what the system did and when, used to look back when there is a problem. Tracing where the system got the data for this answer. How long do we keep logs, and who can see them?
Manual Work done by hand without a system to help, which is often where time is lost and mistakes happen. Keying data from emails into the system one item at a time. Which work is still done by hand, and how many hours a month does it take?
Manual Fallback A way to do the work by hand when the system is unavailable, so the business doesn't stop. If the system goes down, the team switches to a backup procedure it has rehearsed. If the system were unavailable for a day, how would we keep working?
Memory Data the system uses to remember context from one session to the next. Memory should keep only what helps the work and has a clear expiry, and it should keep preferences, facts and conversation history separate to control risk. Remembering dietary restrictions, with consent. Can users view, edit and delete it?
Message Queue A holding area for data between two systems. Items wait in line to be sent and are not lost even if the destination goes down for a while. The ERP is down for an hour of maintenance, orders wait in the queue, and they flow in once it is back. How long can the queue get before the business feels it, and how would we know?
Meta Tag Short text in a web page's code that tells Google and social media what the page is about. It affects the headline people see in search results. Writing the title and description to match what customers actually search for. Does every important page have its title and description written yet?
Metadata Information attached to a document that describes it. Recording the machine model, the owner and the expiry date. Which fields do we need in order to find and control documents?
Middleware Go-between software that takes data from one system, converts its format, then passes it on to another. The middleware takes orders from the app, converts the product codes, then sends them into the ERP. If the middleware stops working, where do the pending items go, and who is notified?
Mobile App An app installed on a phone that can use the camera and location, and can work offline. Field technicians take photos and log jobs from their phones. Does it need to work without a signal?
Model The AI engine that does the processing. There are many versions at different prices, each strong in different areas. Simple tasks use a small model, and hard tasks use a larger, more expensive one. Do we choose the model to fit the task, or use one model for everything?
Model Abstraction A layer that separates the product from the model provider. It keeps the product logic apart from the provider, which makes it easier to switch models, control costs and test alternatives with less impact on the system. Switching models without rebuilding the workflow. How are differences in quality and cost tested?
Model Card A document summarizing what a model is used for, what its limits are and how it was evaluated. A summary of the intended uses and the cases where it should not be used. Do we have a document like this for the systems we use?
Model Drift A model's quality drops as real-world conditions change. New lighting or packaging throws the results off. Which events should trigger retesting or retraining?
Model Monitoring Tracking whether a model is still performing well. An alert when sensor data changes pattern. Who is notified when model or data quality drops?
Model Routing Deciding which model or method should handle each type of task. Simple tasks use rules, and only hard tasks go to a large model. Do we choose the model to fit the task, or use one for everything?
Monitoring Keeping watch on whether a system is still working well after it goes live, beyond checking it once at handover. Reviewing the error rate every week. If quality starts to slip, how many days until we know?
Multimodal Able to take in and understand several kinds of data. Each medium has different strengths: images can provide evidence, while voice helps when hands are busy. The system should pick what fits the task instead of adding every medium simply because it can. Reading an image of an error along with a spoken description. Which formats are good enough for real use?
Multimodal Prompt Sending several kinds of media for the model to consider together. The prompt should state the role of each image, audio clip and piece of text, and what to do when the evidence is unclear, instead of leaving the model to guess. A photo of the equipment with a spoken question. Which media are necessary, and do we have consent for them?
MVP The smallest product that can prove real value with customers. Launching the service in one city only, to measure demand. What can we cut while still proving the value?
Next.js A web framework that renders pages in advance, so the site loads fast and Google can index it fully. Product pages open almost instantly on a 4G phone. Why was this technology chosen, and how easy is it to find people to maintain it?
NLU A system's ability to understand what a customer wants from a typed sentence, even with typos or different wording. Understanding that “got it in white?” and “any white ones left?” are the same question. What does the system do when it isn't sure what the customer is asking?
Observability Being able to see what a system is doing and why it produced a given result. Tracing which data set this agent pulled before it gave a wrong answer. When something goes wrong, how long does it take us to find the cause?
OCR Reading text from images or scanned documents and turning it into text the system can search and calculate with. Photograph a purchase order and the system fills the data into a table by itself. How accurately can it read handwriting or documents photographed at an angle?
Offline-first Designed to work even without the internet. The app must be built from the start to create and edit data offline, show the sync status, and handle cases where the data on the two sides conflicts. Recording an inspection and syncing it later. How are conflicts resolved during a sync?
Omnichannel Letting customers talk to us on any channel, with the same data visible everywhere, so they never have to repeat their story. A customer messages on LINE to continue a conversation left unfinished on the website, and the full history is still there. Has customer history from every channel been brought into one place yet?
On-device AI Running AI on the device itself. It suits work that needs speed, privacy or offline use, but model size, battery life and the capabilities of each device model have to be taken into account. Summarizing notes without sending the audio to the cloud. Which device models support it?
On-premise Installing a system on machines at the company's own premises, instead of renting it in the cloud. The server that runs the model sits in the server room at head office. Who looks after the machines, updates the system and takes responsibility when hardware fails?
Operating Model How an organization brings its people, processes and systems together to deliver work. Moving from working by department to working by value stream. If systems can do the work, how should the team structure change?
Opportunity Cost The value lost by putting talented people on work a system could do. An analyst spends half a day combining files. If we gave this time back to the team, what would they do with it?
Orchestration Coordinating several steps and systems so they work together. The coordination layer must know the status, dependencies and failures of each step, so it can stop, retry or hand off to a person without restarting the whole process. Opening a new branch by following the dependencies between tasks. Who is responsible for the whole flow from end to end?
Orchestrator The controller that sequences the agents' work and combines it. The orchestrator owns the plan and the overall status, but it should not hold every permission itself. Each tool still checks the permissions for its own task. A manager agent calls on specialist agents. Who owns the final outcome?
Outcome The end result the business actually wants, as distinct from the amount of work completed. The customer gets a quotation and can make a decision, beyond the system merely producing a document. Which business result does this system improve?
Outcome Owner The person responsible for the end result, beyond their own part of the work. The person who can explain why the customer is still waiting, even though every department says its part is done. Who is measured on the overall result, as opposed to their own department's work?
Output What a system produces, such as documents, messages or recommendations. A draft reply to a customer email written by the system. What criteria must the output meet to count as usable?
Owner The person truly responsible for something, as distinct from everyone involved with it. Pricing data is owned by the finance department. Who owns this, and do they have the authority to change it?
P50 The median: half of the jobs are faster than this and half are slower. It answers the question “how long does it usually take?” A P50 of 25 minutes means half of the jobs finish faster than that. Is there a big gap between our median and our average, a sign that there are extreme cases?
P90 The value that 90 percent of jobs come in at or under. It shows the slow cases, which an average can hide. If the P90 for issuing a quotation is 3 hours, 9 out of 10 quotations are done within 3 hours. Do we look at the average or the P90, and what do the customers who wait longest experience?
Parallel Run Running the old and new systems side by side for a while, to compare results before retiring the old one. Accounting keeps keying entries by hand for two more weeks while the connector runs, and compares the totals every day. What criteria tell us we can stop running both, and who makes the call?
Payback Period How long it takes for the returns to cover the money invested. Saving this much a month means the investment pays back in so many months. Which assumptions is this figure based on?
Penetration Test Hiring experts to try to break into the system for real, under an agreement, to find weaknesses before attackers do. An outside tester tries to reach patient data without an account, then reports what they managed to do. When was the last test, what did it cover, and have all the weaknesses it found been fixed?
Percentile A way of reading data by sorting it from lowest to highest and seeing the value at a given percentage position. Sorting a month's job times and looking at the 90 percent mark. Is the reported figure an average or a percentile?
PII Information that can identify a person, which needs special care. A customer's name, phone number and document numbers. Which systems is personal data sent out to?
Pilot A trial project with a limited scope, to prove the idea before expanding it. Trialing it with one customer service team for two months. Which figure has to reach what level for this pilot to count as a success?
Policy A policy enforced on the system that sets what it may and may not do. Personal data must never be sent outside the system. Is this policy enforced in the actual system, or just written down?
Policy Layer A layer of rules enforced outside the prompt, to control what the system can do. Personal data is never sent outside the system, however the instruction is worded. Are the rules enforced by the system, or only written into the prompt?
POS The point-of-sale system that records sales and deducts stock at the moment of sale. A sale in the store immediately reduces the stock shown on the website. Does our current POS let other systems pull data from it?
Predictive Maintenance Using data from machines to predict when they will fail, so repairs happen before the production line stops. The system warns three weeks ahead that a bearing is near the end of its life. How much does each sudden machine stoppage cost us today?
Preference A user's likes or restrictions. Preferences change with the situation and should not be treated as hard rules, so the system must let users review, edit and clear what it remembers. No products that need permanent installation. Can customers see and delete their preferences?
Private LLM A language model running on infrastructure the company controls itself, so the data put into it never goes to an outside provider. An assistant that searches the company's contracts runs on machines in the company's own server room. What do the hardware and upkeep cost per year, and how much does answer quality differ from outside services?
Problem Framing The ability to turn a broad request into clear tasks, criteria and constraints. Turning “help me write this” into a brief with measurable criteria. Are our briefs clear enough for the system to get the work right?
Process The sequence of steps the company actually uses to get work done. The steps from receiving a purchase order to shipping the goods. Does the written process match what people actually do?
Process Map A picture of the real sequence of work, with the person responsible and the time taken at each step. A diagram of the work from receiving an order to shipping it. Is this map based on real data, or on interviews alone?
Process Owner The person responsible for a process from start to finish, beyond the stretch that sits in their own department. One owner for the whole path from a customer's price request to collecting payment. Who owns this process, and do they have the authority to make changes across departments?
Product Feed Product data sent for systems to use. A good feed has codes, specifications, units, prices, stock and regular update times, so every channel refers to the same set of data. Prices, stock, specifications and product codes. Who keeps the data up to date?
Product Information Management A central system that holds the master copy of product data for every channel. Change a price in one place and the website and marketplaces update to match. If data in two places doesn't match, which one does the system treat as correct?
Production The environment real users work in, as distinct from a trial system. The system employees use for their daily work. What conditions must be met before the system is ready for production?
Prompt The instruction or brief we write for AI to act on. The quality of the brief has a big effect on the quality of the answer. Asking it to summarize a customer email and spelling out what the summary must include. If the prompt needs changing, who can change it, and is it tested first?
Prompt Injection When text from an outside source tricks AI into doing something it shouldn't. A file sent in by a customer has hidden instructions in it. How do we guard against rogue instructions hidden in outside data?
Proof of Concept A small trial to prove that an idea is technically possible. Testing whether the system can really read documents in this format. If it passes, what is the next step, and how long will it take?
Prototype An early model for testing an idea before full development. A skill dashboard page trialed with one department. What do we need to learn from the prototype before investing in full development?
PWA A website that works like a mobile app. It opens straight from the browser, with nothing to download from an app store. Warehouse staff scan items with their phones without installing an app. Does it need to work when the signal drops?
QC Checking quality before delivery. Short for Quality Control. Staff inspect each piece before it is packed into a box. Are the pass and fail criteria written down clearly, or do they live in people's heads?
QR Code A square code scanned with a phone camera, used to open a web page or identify a product or document instantly. A QR code on a shelf shows that shelf's real stock when scanned. Who will do the scanning, and with what device?
Quality Gate A checkpoint that must be passed before work moves on. Checking prices and references before a proposal is sent. Is passing judged by rules or by experts, and how is the evidence recorded?
Queue Work waiting to be processed or decided by a person, lined up in order of priority. High-risk work is pushed to the top of the queue. What criteria set the queue order, and who can change it?
RAG Having AI search the organization's own information before it answers. Answering from approved SOPs, with links to the documents. Which sources does the system search, and how does it keep out outdated documents?
Rate Card A table of standard prices for each type of work or service. Used to price proposals consistently every time. Do the prices in the system match the ones sales actually uses?
Rate Limit A cap on the number of tasks or requests in a given period, to keep the system from overreaching. Limiting an agent to a set number of emails per hour. When the limit is hit, does the system stop or queue the work, and who is notified?
React A widely used library for building screens. It makes interfaces respond quickly and lets their parts be reused. The shopping cart total updates instantly without reloading the page. How easily could another team take over maintaining this code?
Real-time Data that updates the moment something happens, without waiting for a processing cycle or a refresh. Customers see the actual stock left when they order, instead of yesterday's figure. How many minutes behind can the data be before it causes problems for customers?
Realtime API A low-delay connection for voice or data. It suits voice experiences or events that need an instant response, but latency, interruptions, cost and a fallback for when the network isn't ready all need planning. A voice conversation that can pull up a case's status. If the voice connection drops, which channel does it fall back to?
Recommendation Ranking options to fit the context. Recommendations need rules that remove options that won't work and a reason for the ranking, and should not rely on language similarity alone. Choosing a machine by the number of users and the budget. Do these criteria serve the customer, or the sales figures?
Reconciliation Comparing the figures in two systems to confirm they match, and tracking down the cause of any items that don't. Every morning a report compares yesterday's order count and sales totals between the app and the ERP. Who resolves the differences, and within how many hours?
Refactoring Restructuring code so it is easier to change later, while the system keeps working exactly as before. Merging a rent calculation that existed in five places into one, with no change visible to users. How will this round of restructuring make the next feature faster or cheaper?
Requirement Something the system must be able to do, agreed in writing before development. The system must produce a quotation within 2 minutes. Can this requirement actually be measured, or is it just a broad description?
Requirement Map A structure of needs and constraints. The map links each need to its solution, assumptions and acceptance criteria, so you can check that every part of the proposal has a reason behind it. Linking pain points to features and acceptance criteria. Who confirms the requirements?
Review Queue A queue of work waiting for a person to check it before it continues. Items the system isn't sure about are sent to a queue for a supervisor to look at. Who watches this queue, and what happens if no one does?
Rework Doing work again because of mistakes or missing information. A quotation has to be corrected because the price was wrong. How do we count rework, and what percentage is it now?
Risk Tier Ranking work by risk to decide how much control it needs. Work that goes to customers is ranked higher than internal summaries. What criteria do we rank by, and who approves the highest tier?
ROI The return compared with the money invested. How much is saved, compared with the cost of development and upkeep. Over how many years, and on what assumptions, is this ROI calculated?
Role-based Access Setting permissions by role. Employees see general lessons, but HR sees assessment results. Who should be able to view, edit, approve or download each type of data?
Role-based View Different screens for different roles. Executives see the overview, and teams see their own work. How much data does each role need to see to do its job?
Rollback Taking the system back to the previous version when a new one has problems. A new version stops quotations from being issued, so the team rolls back to the old version within five minutes. When did we last rehearse a rollback, and what happened to data created in the meantime?
Root Cause The real cause of a problem, beneath the symptoms you can see. Defects rose because of how the machine was set up after a lot change, not because of the inspectors. Are we fixing the symptom or the cause?
Row-Level Security Database rules that set exactly which rows of data each user can read or edit. The appointments table returns only the rows that belong to the patient who is logged in. Which tables still lack these rules, and why?
RPO How much data, measured in time, you can afford to lose when restoring from a backup. Backing up every fifteen minutes means that a restore loses no more than the last fifteen minutes of orders. If we had to restore right now, what time would the latest recovered data be from?
RTO The longest the system may be down, counted from the incident until it is back in use. The store sets a target that the checkout page must be back within thirty minutes. Have we ever actually met our RTO in a drill?
Rule A clearly written rule so the system decides the same way every time. Amounts over 100,000 must go to a supervisor. Who can change this rule, and does a change take effect immediately?
Runbook A step-by-step guide for handling incidents anticipated in advance, naming who does the work and who makes the decisions. When the checkout page goes down, the runbook says who checks what, who posts an announcement on the company page and what message to use. When was this runbook last used in a drill, and has everyone on the on-call roster read it?
Safety Stop A condition that makes the system stop immediately, for safety. Automated messages stop when the error rate goes over the threshold. What makes the system stop on its own, and who is notified?
Sandbox A trial space kept separate from the live system. Training an agent without touching production data. How is the trial data kept separate from production, and how is it cleaned out?
Scenario A picture of the results under a set of assumptions. A scenario is a calculation under openly stated assumptions, used to compare options and see how sensitive the results are. It is not a forecast. If sales grow 10%, how many people will we need? Who confirms the assumptions?
Scenario Modeling Simulating results under several sets of assumptions to compare options. Modeling headcount if 20, 40 or 60 percent of the work is automated. Which real data are the assumptions based on, and who confirms them?
Schema Markup Extra information in a web page that tells Google which parts are prices, reviews or products, so search results show more detail. Search results show prices and review stars under the store's name. Do all our product pages carry this information yet?
Scope The agreed boundaries of what will and will not be done in this round. This round covers one product group only, in Thai only. What is out of scope, and what would it take to add it?
Secret Management How a system's secret credentials, such as database passwords and payment service keys, are kept outside the code, with only a limited set of people able to see them. The messaging service key sits in a secrets vault, and the web page code does not contain it. If a secret leaks, how many minutes does it take us to replace it, and who does it?
SEO Making a website easy to find on Google without paying for ads, through both the site's structure and its content. A customer searches for “custom wooden table” and finds our store on the first page. Which search terms actually bring in customers who are ready to buy?
Service Account An account a system or agent uses to do its work, kept separate from employees' accounts. The agent uses its own account to access systems, instead of sharing an employee's. Whose account does each system use, and can its access be revoked immediately?
Service Blueprint A diagram that shows both what customers see and the behind-the-scenes work that makes it happen. Seeing side by side what the front end does and what the back office has to support. Do we have the full picture of the work behind the scenes, or are we looking only at the screens?
Session Context Information the system remembers within a case. The context should keep what the user has confirmed apart from what the AI has inferred or predicted, and be kept only as long as privacy allows. Remembering the model and the steps the customer has already tried. How long is it kept, and who can see it?
Session Memory Remembering the context within one conversation, so the user isn't asked the same thing again. Customers don't have to restate their conditions every time they reply. How long do we keep the context, and when is it deleted?
Shadow AI Employees using AI tools the company has not approved and cannot see. An employee photographs a contract on a personal phone and has an AI app summarize it. How do we know which AI tools employees are using right now, and for what?
Shadow Mode Letting the system make recommendations without yet controlling the real work. Comparing alerts with the technicians' own judgment. How long does the trial need to run, and under which conditions, before real use?
Single Source Brief A single starting document that every department refers to. All messaging and offers are drawn from the one brief. If the brief changes, how will the work already produced be brought up to date?
Sitemap A file that tells Google which pages a website has, helping new pages get into search results faster. A new product goes up and Google finds it within a day. How long does it take a new page to appear in search results?
SKU A product code for one item that sets it apart from every other item. Short for Stock Keeping Unit. The same shirt in different colors and sizes counts as different SKUs. Is price and stock data linked to every SKU yet?
SLA The agreed service time. An SLA should measure time that matters to the person being served, separate time spent waiting for information from time spent working, and say what happens when the deadline is missed. IT picks up a request within 4 hours. Who is notified when an SLA is about to be missed?
SOP A standard procedure manual the team follows so everyone does the work the same way. Short for Standard Operating Procedure. Goods-receiving steps that every shift performs the same way. When the system changes how the work is done, has the SOP been updated to match?
Source Where data or an answer came from, used to check whether it can be trusted. The answer cites page 12 of the manual. Which document did this answer come from?
Source Code Ownership The agreement on who owns the code and documents that are developed. The company owns the code and keeps it in its own systems. If we change developers, what do we take with us?
Source of Truth The main data source that everyone treats as correct. Prices are read from the ERP, never from old files. Where does the master data live, and who has the right to change it?
SSO Signing in once with a company account to get into many systems. Short for Single Sign-On. Employees sign in to the AI assistant with their company email account, and when someone leaves, their access closes across every system at once. When an employee moves departments, how quickly does their access to the assistant change to match?
Staging A copy of the live system used to try out new things before customers get them. The sales team tries the pre-booking feature on staging for a week before it goes live. How does staging differ from the live system, and who approves going live?
State The current status of a piece of work, showing which step it has reached. The request is in the “awaiting approval” state. Does every piece of work have one status that everyone sees the same way?
State Machine Rules for how work moves from one status to the next. A state machine defines the statuses and allowed transitions clearly, which stops work from skipping steps or getting stuck in a state with no owner. Draft → Review → Approved Can a wrong status be reversed?
Statement of Work A document setting out the scope of work, the deliverables and the acceptance conditions. Stating what will be delivered this round and when it counts as finished. Are the acceptance conditions written down clearly yet?
Straight-through Processing Letting standard items flow through to completion without anyone touching them. Requests that meet the criteria are approved and recorded automatically. What percentage of items currently finish on their own with no one touching them?
Structured Data Data formatted to a standard so machines can read and understand it, as opposed to loosely written text. Listing the price and stock status in a standard format on the product page. Do our important pages carry all the machine-readable data they should?
Structured Output AI output in a fixed data format. A fixed format lets software check for missing fields and pass the data on, which lowers the risk that comes with pulling facts out of loosely written text. Returning a list of SKUs and reasons in separate fields. What does the system do when data is incomplete?
System Integration Connecting several systems so data flows from one to the next. An incoming email creates an item in the approval system. Which system owns the data, and what happens when the connection fails?
Task A single, clearly separable piece of work, used as the unit for planning and measuring. Breaking the sales admin role into 12 tasks. Have we broken the work into smaller units yet?
Task Completion A piece of work running all the way to its goal without stalling partway. A request is closed without needing to be reopened. What percentage of work finishes in one pass?
Task Inventory A list of a position's tasks, with the time, frequency and risk of each. Breaking the sales admin role into 12 tasks and assessing each one. Do we have a real list of tasks, or are we estimating by feel?
Task Success The share of tasks users complete as they intended. Customers who start a quote request and actually receive the quote. Are we measuring whether people get things done, beyond whether they log in?
TCO The total cost over the whole life of a system (total cost of ownership). Including development, cloud, support and review time. Does it include the upkeep of integrations and models, and the people who review the work?
Technical Debt The burden that builds up from code written quickly to get something out the door, which makes each later change slower and riskier. The rent formula was written in five separate places, so a single price increase means fixing all five. Which part of the system carries the most debt, and how much does it slow down changes?
Telemetry Status and usage data that a system sends continuously. Tracking error counts and API spending. Which data helps us fix the system, and which goes beyond what we need?
Template A document or message pattern that the system fills in with data automatically. A quotation that fills in the customer details and prices by itself. Who can change the template, and is there version control?
Threat Model Working out in advance who might attack the system, by which route, and what damage they could do, to decide what to protect first. The team maps how patients, former employees and outsiders could each reach medical records. What are the top three risks to our system, and who is responsible for each?
Threshold The limit value that triggers an alert. Temperature above the limit for 10 minutes straight. Is the threshold set from risk or from the average, and who approves it?
Throughput The amount of work a system or team completes in a given period. The number of quotations issued per week. If the workload doubled, could the system handle it?
Time-series Data Data arranged in time order. Vibration readings every second. Do the timestamps, units and operating states line up?
Tool An outside tool or system that AI calls on to do real work. Checking the stock system before answering a customer. Which tools can the system call, and how is that limited?
Tool Call AI calling another system to do real work, beyond replying with text. Calling an API to check stock before answering a customer. Which tools can the system call, and how is that limited?
Tool Calling Letting AI request a system function. The model asks to use the tool, but the software must check the parameters, permissions and results before anything actually happens. Checking the appointment calendar before suggesting a time. Does every tool need approval?
Total Cost The full cost, including development, usage and ongoing upkeep. Including model fees, hosting and the team's time spent maintaining it. What will the second year cost?
Touch Time The time someone actually spends working on a task, excluding waiting time. Checking a document takes 12 minutes. How big is the gap between touch time and lead time?
Traceability The ability to trace a result back to the physical part it came from. Finding images by serial and lot number. Which data do we need to be able to trace a result back to?
Tracing Recording the sequence of the system's reasoning and tool use. A good trace links the input, model, prompt, tools, output, time and cost, so the team can track down causes and build regression tests. Seeing at which step a flow failed. Is trace data kept secure, and how long is it kept?
UAT The period when real users try the system before launch, to confirm it works for the real job, beyond passing the development team's own tests. Three admins take real orders for a week before launch. Who signs off that the system is ready to use?
UI The part of a system that users see and click. Short for User Interface, as distinct from UX, which covers the whole experience. Buttons, menus and tables on the screen. Who is this screen designed for, and when do they use it?
Unit Cost The cost of doing one piece of work once, used for comparison with the cost of staff. The cost of answering one customer query. What does the unit cost include, and how does it change with volume?
UX What it is like for users to work with the system, and whether it is easy or hard to get things done. Short for User Experience. Customers finish a form in 3 fields instead of 12. How do we measure whether users are getting things done more easily?
Validation Proving an assumption with evidence from real users. 12 customers agree to pay a deposit. Which kind of evidence do we actually treat as reliable?
Value Hypothesis An assumption about why the thing we are about to build has value, and how we will prove it. We believe customers will pay if the wait drops from 2 days to 2 hours. How will we know if this assumption is wrong?
Value Stream The end-to-end path of work that creates value for the customer, across departments. Lead-to-cash, from a customer's first interest to payment. With whom does this value stream start and end?
Vector Database A database that searches by meaning instead of matching words, so AI can find relevant documents even when different words are used. Asking “what waterproof products do you have?” brings up products described as “moisture-resistant”. Which set of documents does the system pull its answers from, and how is that set updated?
Vendor An outside provider or seller of a system. A company that sells us software or AI model services. If we stop using this vendor, how do we get our data out?
Vendor Lock-in A situation where it is hard to change providers because the system is tied too closely to one of them. All the prompts and data sit in the vendor's system. If we stop using this vendor, what can we take with us?
Version A release of a document, rule set or system, used to show what was in use at a given time. This month's manual differs from last year's. Do we know which version of the rules the system was using at the time?
Version Control Keeping track of versions of the work, so you know which is the latest and can go back. Going back to a previous version of the copy. If the wrong version is released, how quickly can we roll back?
Versioning Keeping multiple versions in a way that can be audited. Every change should have a number, an author, a time and a record of what changed, so you know which document or rule was in use on the day a decision was made. Knowing which month's prices a proposal used. Which version is the approved one?
Wait Time Time when work sits idle with no one working on it. A document waits two days for approval but takes 10 minutes to check. What percentage of our total time is actual working time?
Web Application Software used through a browser, with nothing to install. A back-office system the team opens from any computer. Which devices does this system need to run on?
Webhook An automatic signal sent when an event happens. The CRM notifies the system the moment a lead changes status. If a signal is sent twice or never arrives, how does the system prevent duplicate work?
WIP Work that has started but is not yet finished. Short for Work In Progress. The more of it piles up, the more money is tied up in it. Orders sitting part-made on the production line. How many items are in progress right now, and how long have they been there on average?
Work Item One unit of work in the system, with a clear status and owner. A single request that moves through several steps. Does every piece of work keep one identifier along its whole path?
Work Order An order for work that states what must be done, who is responsible and its status. A repair order a technician picks up from the system and closes when the job is done. Is the work order linked to the machine's data and repair history?
Workflow The sequence of work from start to result, showing who does what. Training request → practice exercise → supervisor review → skill certification Which steps should be automated, and where must a person decide?
Workflow Automation Letting the system run repetitive steps one after another on its own, instead of people copying data between apps by hand. An order from LINE goes into the stock sheet and notifies the delivery team automatically, with no retyping. How many hours a day does the team spend copying data between systems right now?
Workflow Engine A system that runs the rules and routes for work. The engine holds the rules, statuses, pending work and handoffs, so the process can be changed without burying every condition in screens or prompts. Routing approvals by spending limit. Who can change the rules?