Glossary
The vocabulary of product management, defined plainly.
Definitions follow this handbook's use of each term, with chapter references for the method. Abbreviations usually appear under the full name. Familiar short forms such as RICE and DACI have their own entries.
A
AARRR (pirate metrics): Dave McClure’s five stages of a customer’s life with a product: acquisition, activation, retention, referral and revenue. In B2B each stage needs a reading for accounts and buying groups rather than individual visitors. See Chapters 05 and 08.
Acceptance criteria: The details that show whether a story was completed correctly, in Mike Cohn’s definition. They can be examples, rules, tests or brief notes. Writing them in Given, When, Then form forces each requirement into something a test or an eval can check. See Chapters 06 and 10.
Acqui-hire: Buying a company mainly for its team, used when you need a team or capability faster than you can build it. License-and-hire is the variant that licenses the technology and hires the people without buying the company. See Chapter 22.
Actionable metrics: Metrics that show clear cause and effect, usually by cohort, so a team can tell what to do next. Eric Ries set them against vanity metrics. See Chapters 17 and 21.
Activation: The moment a new user or account first gets real value, defined as a specific completed action that delivers that value. In enterprise products it is never a login, and for AI products it usually means an accepted output. See Chapters 05 and 08.
Adoptability: The fourth risk that B2B products carry beyond desirability, feasibility and viability: whether the buying group, the security review and procurement will let the product through. See Chapter 01.
Agent: An AI system in which the model directs its own process and tool use, in a loop, based on what it observes. Erik Schluntz and Barry Zhang of Anthropic distinguish agents from workflows, which follow predefined code paths. See Chapters 31 and 32.
Agent2Agent protocol (A2A): A protocol created by Google and announced in April 2025 for agents to talk to other agents across organizations. It became a Linux Foundation project in June 2025 and complements MCP, which connects agents to tools. See Chapters 14 and 31.
Agile Manifesto: The statement of values and principles for software development written in February 2001 by seventeen practitioners, among them Kent Beck, Martin Fowler, Ken Schwaber and Jeff Sutherland. See Chapter 11.
AI cost model: The artifact that tracks what one task, user or resolved outcome costs in inference, and which way that cost is trending. It belongs to the Business dimension of the Product 360. See Chapters 00 and 31.
AI literacy: A duty under the EU AI Act, applying from February 2025, to make sure the people who build and use AI systems understand them well enough. Minimal-risk systems carry no specific duties beyond it. See Chapter 33.
AI PRD: A PRD with additional requirements for an AI feature: eval targets, failure taxonomy, latency budgets and task costs. It also specifies trust and failure UX, staged autonomy, a kill switch, a gross-margin floor and regulatory classification. Chapter 04 calls these sections the AI feature addendum. See Chapters 04 and 31.
AI risk assessment: The artifact that records an AI product’s failure modes, guardrails and regulatory obligations. It belongs to the Operate dimension of the Product 360. See Chapters 00 and 33.
Ambidextrous organization: Charles O’Reilly and Michael Tushman’s model in which separate exploratory units are integrated at the senior team, so new businesses get protection while the core shares its assets. See Chapters 20 and 23.
Anchoring: The bias by which an initial number pulls later estimates toward it, even when the number is arbitrary. The book rates it robust and applies it to the first price or first ROI figure a buyer sees. See Chapters 10 and 12.
Answer engine: An AI assistant that buyers use to research and shortlist vendors before they touch your website. The book’s answer engine audit runs a fixed set of buyer questions across the major assistants each month. See Chapter 08.
Appetite: Ryan Singer’s term in Shape Up for the time a business will spend on a problem. An estimate prices a design. An appetite sets the time and shapes the design to fit. See Chapters 06 and 11.
Aspirational OKR: An OKR describing a desired future state. The playbook used here expects an average score of 0.7, with high variance and possible carryover. Committed OKRs expect 1.0. A miss requires a blameless postmortem. See Chapter 02.
Assumption map: A two-by-two that plots a team’s assumptions by importance and by the evidence behind them. Assumptions in the top-right quadrant (important, little evidence) get tested first. See Chapters 01 and 21.
Atomic network: Andrew Chen’s term for the smallest network that is stable and can grow on its own. In B2B it is often one team, one account or one integration pair. See Chapter 14.
Automation ladder: This handbook’s six-rung path from single prompts toward agents. Advance when the previous rung works reliably. This is a Product 360 synthesis. See Chapter 32.
Autonomy levels: Kevin Feng, David McDonald and Amy Zhang’s five levels, defined by the user’s role. They start with the user directing and deciding as operator. Set autonomy per action. See Chapter 31.
B
Backlog: Candidate work, organized here into three stores: an inbox for raw requests, an opportunity backlog of customer problems and a delivery backlog of committed work. See Chapters 03 and 06.
Battlecard: A one-page, sales-facing summary of one competitive situation: why you win, the competitor’s pitch in its own words, landmine questions and proof. See Chapters 13 and 27.
Beachhead: Geoffrey Moore’s method of dominating one narrow segment, delivering its whole product and earning references before expanding into adjacent segments. See Chapters 13 and 24.
Beta: A pre-release stage before general availability. GitLab’s sequence of stages runs experiment, beta, limited availability and generally available. Beta and preview terms usually carry no SLA. See Chapter 25.
Big Idea: A single sentence that states your point of view and what is at stake for the audience, written before you build a presentation. Nancy Duarte and Cole Nussbaumer Knaflic both use the term. See Chapter 35.
Bottom line up front (BLUF): The military writing convention of putting the conclusion, recommendation or request in the first lines, then the background. See Chapters 34 and 35.
Break-even volume test: A check run before any price change: how much volume can you lose, or must you gain, for the change to leave contribution flat? See Chapter 15.
Build, measure, learn: Eric Ries’s loop in Lean Startup: turn ideas into products, measure how customers respond and learn whether to pivot or persevere. The aim is to speed up the loop. See Chapters 05 and 21.
Build, partner or buy: The decision about how to acquire a capability. The book frames it as a sequence of questions, starting with whether internal resources are relevant and, if not, whether the resource can be borrowed by contract. See Chapter 22.
Bundling: Selling products as a package. The book distinguishes separate pricing (components only), pure bundling (the package only) and mixed bundling (both). A bundle can sell to buyers who would not pay the standalone price of either part. See Chapters 12 and 15.
Business case: An event artifact presenting an investment, its assumptions, scenarios and expected returns for finance review. See Chapter 13.
Business viability risk: One of Marty Cagan’s four risks: does the product work for the rest of the business, including sales, marketing, finance, legal, support and partners? The PM is accountable for it. See Chapters 00 and 04.
Buying center: Frederick Webster and Yoram Wind’s 1972 model of the roles in an organizational purchase: users, influencers, buyers, deciders and gatekeepers. See Chapter 01.
Buying group: The set of people in a customer organization who together make a purchase decision. A buying group can say no at many points and yes only at the end, after consensus. See Chapters 01 and 10.
Buying jobs: Gartner’s six jobs a B2B buying group has to complete, in no fixed order: problem identification, solution exploration, requirements building, supplier selection, validation and consensus creation. See Chapters 01 and 08.
Buying journey: The journey map of a buying group working through the buying jobs, with one swimlane per persona. The book draws it separately from the usage journey (one end user, one scenario) and connects the two at the first-value moment. See Chapter 01.
C
Calibrated trust: Trust that matches a system’s actual trustworthiness. Too much trust produces misuse, such as accepting wrong answers. Too little produces disuse, such as ignoring right ones. See Chapters 10 and 31.
Canary release: A release that exposes a new version to a small subset of users first, to reduce risk. Danilo Sato separates it from an A/B test, which checks a hypothesis over days. See Chapter 25.
Capability evals: Evals that measure progress on hard tasks. They should start low and climb, and tasks graduate into the regression suite once they pass reliably. See Chapter 31.
Capacity split: The capacity a strategy assigns to new value, platform work, quality and debt, and customer commitments. Write the allocation explicitly so urgent requests can be evaluated against it. See Chapter 03.
Category design: The discipline, from Play Bigger, holding that winning companies define a new problem, name the category that solves it and become its recognized leader. Play Bigger now calls it necessary but not sufficient. See Chapter 24.
CD3: Don Reinertsen’s sequencing rule: cost of delay divided by duration. Work with the highest cost of delay per unit of time goes first. See Chapter 03.
Censoring: In survival analysis, the situation where some customers have not yet left, so you cannot see the rest of their history. Dropping current customers from a churn analysis is the most common mistake it causes. See Chapter 09.
Champion: The person inside a customer who advocates for your product and whose reputation rides on it. The MEDDIC sales method renamed Miller and Heiman’s coach as the champion. See Chapters 01 and 09.
Charge metric: The unit you bill on, which can differ from the value metric that tracks what the customer gets. Stripe’s 2025 framework names three kinds for AI products: consumption, workflow and outcome. See Chapters 02 and 12.
Choice cascade: A. G. Lafley and Roger Martin’s definition of strategy in Playing to Win as five linked questions, starting with the winning aspiration and running through where to play, how to win, required capabilities and management systems. See Chapter 03.
Claims substantiation log: The go-to-market artifact that records the evidence behind every external claim about the product. See Chapter 24.
Cohort: A group of users or customers who started in the same period. A cohort retention curve plots the share of each cohort still active after one, two or three periods. See Chapters 05 and 09.
Cold start problem: The first of Andrew Chen’s five stages of a network: getting a network going before it has enough participants to be useful. The later stages are the tipping point, escape velocity, hitting the ceiling and the moat. See Chapter 14.
Competitive alternatives: The first of April Dunford’s five positioning components: what customers would do if your product did not exist. In B2B the answer is often a manual process, an internal tool or doing nothing. See Chapter 24.
Complement: Michael Cusumano’s term for something that makes a host platform more valuable. Being a good complement brings distribution, trust and a buying path at the cost of dependency. See Chapters 14 and 16.
Concierge MVP: An MVP in which people visibly do the work that software will eventually do, to test whether the outcome is worth having. In a Wizard of Oz test the people are hidden. See Chapters 04 and 21.
Confidence interval: The range of plausible values around an estimated effect. The book asks for the effect size with a confidence interval, not only a p-value. See Chapters 07 and 35.
Confidence Meter: Itamar Gilad’s tool for grading confidence by evidence type. Opinions earn little confidence, customer research more and observed use from an MVP or launch the most. See Chapter 03.
Conjoint analysis: A pricing research method in which respondents choose among realistic packages that vary features and price, usually across 8 to 20 choice tasks. It estimates the value of each attribute and willingness to pay. See Chapter 15.
Connector: Live, permissioned access from an AI harness to a system where evidence sits, such as the CRM, backlog, analytics or support tickets. Most are built on MCP, and every connector is also an access grant. See Chapters 19 and 32.
Context engineering: Curating everything a model sees, including its prompt. Anthropic frames the aim as the smallest set of useful context that produces the intended outcome. See Chapters 19 and 32.
Context pack: Context for AI analysis: the decision, a Product 360 summary, metric definitions and canonical values, segment definitions and a source map. Include data rules and known quality issues. See Chapters 19 and 32.
Continuous discovery: Teresa Torres’s practice of treating discovery as a weekly habit rather than a phase: the team building the product talks to customers every week in pursuit of a desired outcome. See Chapter 01.
Contract churn: Churn in which the customer does not renew. It is observed and unambiguous and happens at renewal dates. Compare usage churn. See Chapter 09.
Controlled experiment: An experiment that randomly splits users between a control, which sees the current experience, and one or more treatments, which see the change. Because the split is random, a difference beyond chance is caused by the change. See Chapters 05 and 07.
Core interaction: In Platform Revolution, the single most important exchange a platform exists to enable. It has three components: the participants, the value unit and the filter that matches them. See Chapter 14.
Cost of delay: Don Reinertsen’s term for the economic cost of delivering later. WSJF and CD3 use it to sequence work. See Chapters 03 and 11.
Cost per task: The inference and infrastructure cost of one completed AI task. The book puts a cost-per-task check in the Definition of Done for features that call models. See Chapters 11 and 31.
Counter-positioning: One of Hamilton Helmer’s 7 Powers: a newcomer adopts a new, superior business model that the incumbent will not copy, because copying would damage its existing business. It is often temporary. See Chapter 16.
Curse of knowledge: Experts overestimate what others understand because they struggle to imagine lacking their own knowledge. Chip and Dan Heath popularized the term. See Chapters 34 and 35.
Customer acquisition cost (CAC): Total sales and marketing cost for a period divided by the new customers won in it. State whether it is blended or paid only, and whether it is fully loaded. See Chapter 08.
Customer advisory board (CAB): A small group of senior customers who advise a company on strategy. See Chapter 27.
Customer development: Steve Blank’s method for new ventures, based on the observation that a startup searches for a business model while a large company executes a known one. A new product line inside a company is in search mode too. See Chapter 21.
Customer Effort Score: A survey measure of how much effort a customer had to spend to get something done, introduced in research finding that effort drives loyalty in service interactions more than delight does. See Chapter 09.
Customer health score: A banded rating per account (green, yellow, red) that combines several signals so customer success and product teams can see which accounts need attention. Because it is a prediction, test it against past renewals. See Chapters 09 and 27.
Customer lifetime value (CLV): The present value of the margin a customer will generate over the relationship. The book models it by segment and channel, after cost to serve and with ranges. Compare lifetime value (LTV). See Chapter 09.
Customer ROI calculator: An external tool, built for the economic buyer and the champion, that lets a customer test a value claim with their own numbers. Every default cites its evidence and can be overwritten. See Chapter 12.
Customer satisfaction (CSAT): A survey that asks how satisfied a customer was with a specific interaction, usually right after it. See Chapter 09.
Cycle time: A flow metric: the time from start to finish for finished items. It is a lagging indicator. Work item age is its leading counterpart. See Chapter 11.
D
DACI: A decision-rights framework, popularized by Atlassian’s Team Playbook, that names four roles for a single decision: Driver, Approver, Contributors and Informed. The book treats it as the lightest fix that works. See Chapters 28 and 29.
Dark patterns: Harry Brignull’s 2010 term for interfaces designed to trick people into doing things they did not mean to do, such as sneaking, forced action or hard-to-cancel flows. He now prefers “deceptive patterns.” See Chapter 10.
Data contract: A written agreement between the owner of a dataset and the people who depend on it, covering schema, data quality expectations, service levels, roles and support. See Chapter 18.
Data network effect: The claimed loop in which more customers give more data, which improves the product, which brings more customers. The book calls it the most overclaimed effect in B2B and AI products, following Andrei Hagiu and Julian Wright’s critique. See Chapters 14 and 16.
Data protection impact assessment (DPIA): The assessment GDPR Article 35 requires where processing is likely to result in a high risk to people. It describes the processing and its purposes, assesses necessity, proportionality and risks, and sets out the measures to address them. See Chapter 33.
Decision log: A running record of decisions, with date, options, decision owner, evidence and outcome. This handbook also records reversibility and a review date. See Chapters 00 and 29.
Decision memo: A document that leads with the recommendation, then covers the frame, the options, the evidence and confidence, the trade-offs, the reversibility and the kill criteria. See Chapters 17 and 34.
Decision quality: How well a decision used the information and reasoning available at the time. Assess outcome quality separately: sound decisions can still have poor results. See Chapter 17.
Decision-rights register: A standing table of recurring decision types, such as a pricing change, a deprecation or a new autonomous action, with the door type, framework, roles and forum assigned in advance. See Chapter 28.
Decoy: A product option priced to make another option look good, also called the attraction effect. In 2014, Shane Frederick and colleagues found the effect appeared mostly in stylized numeric tables and questioned its practical significance. See Chapter 12.
Definition of Done: In the Scrum Guide, a formal description of the state of the Increment when it meets the quality measures required for the product. The book adds evals and cost-per-task checks for AI features. See Chapters 06 and 11.
Definition of Ready: Explicit, visible criteria, usually based on INVEST, that a story must meet before it enters an iteration. It is not part of the Scrum Guide. See Chapter 06.
Deployer: Under the EU AI Act, an organization that uses an AI system under its own authority. In a typical B2B stack the enterprise customer is the deployer. Compare provider. See Chapter 33.
Deprecation: The state of a feature or API that is no longer in active development and may be removed later, though it still works. The book separates it from end of servicing, removal and retirement. See Chapter 30.
Design partner: A customer who helps shape a product before general availability, sharing workflows and data and giving scheduled feedback in return for early access and influence. See Chapters 22 and 25.
Design sprint: A five-day method, developed by Jake Knapp at Google and GV, that compresses a double diamond into a week: map, sketch, decide, prototype and test with customers. See Chapter 21.
Design thinking: A human-centered approach to solving complex problems that came to business largely through IDEO and Stanford’s d.school. IDEO frames its target as the overlap of desirability, feasibility and viability. See Chapter 21.
Desired outcome: The root of Teresa Torres’s opportunity solution tree: the metric the team is trying to move. In Tony Ulwick’s usage, desired outcomes are the measurable results customers want at each step of a job. See Chapter 01.
DHM: Gibson Biddle’s model of product strategy: how will the product delight customers in hard-to-copy, margin-enhancing ways? See Chapter 03.
Differentiation value: In economic value estimation, the money value of every way your product differs from the customer’s next best alternative, positive or negative. See Chapter 12.
Direction, Delivery and Durability: The three groups of Product 360 dimensions. Direction covers customer, strategy and business. Delivery covers build, measure and go-to-market. Durability covers operate, lifecycle and leadership. See Chapter 00.
Disagree and commit: Jeff Bezos’s practice for decisions with conviction but no consensus: say you disagree, then commit fully to the decision. It applies to the boss as well. See Chapters 17 and 29.
Discount rate: The rate used to convert future cash flows into today’s money, usually the company’s required return or hurdle rate. Finance sets it, not the PM. See Chapters 12 and 13.
Discounted cash flow (DCF): A valuation method that discounts each year’s expected cash back to today and adds the results. Net present value is its summary number. See Chapters 12 and 13.
Disruptive innovation: Clayton Christensen’s term for an offering that starts simple and cheap, at the low end of a market or in a new market of nonconsumers, and moves upmarket until it displaces incumbents. The book’s disruption test applies the theory as Christensen stated it. See Chapters 16 and 20.
Door type: A decision’s reversibility: one-way doors are hard to reverse, two-way doors easier. Record the type to choose a proportionate decision process. See Chapters 17 and 28.
DORA metrics: The software delivery measures from the DORA research program: three for throughput (change lead time, deployment frequency, failed deployment recovery time) and two for instability (change fail rate, deployment rework rate). The book treats them as context for the PM, not as PM goals. See Chapter 06.
Dual-track agile: A way of working in which discovery (fast, cheap learning) and delivery (building production-quality software) run in parallel. Desirée Sy described the practice, Marty Cagan popularized the name and Jeff Patton clarified it. See Chapters 07 and 11.
Durable artifacts: Artifacts that change rarely, usually once a year or at a major strategic shift, such as the vision, the strategy one-pager, personas and positioning. See Chapter 00.
E
Early access program (EAP): A release stage in which accounts whose admin opts in get a feature in private or public preview, under preview terms, with no SLA and best-effort support. For B2B AI features, who opts in is a governance decision. See Chapter 25.
Economic buyer: The buying-group role that owns the budget and signs. Their goal is to hit a business target within budget, and their typical fear is being unable to prove the ROI to the CFO. See Chapters 01 and 12.
Economic value estimation (EVE): Thomas Nagle and Georg Müller’s method for estimating a price ceiling. Add your product’s differentiation value to the next best alternative’s reference value. MIT teaches the same concept as economic value to the customer (EVC). See Chapter 12.
Empowered product team: Marty Cagan’s term for a cross-functional team that is given problems to solve and measured on outcomes. He contrasts it with delivery teams and feature teams, which are accountable for output. See Chapter 36.
Enablement kit: Seller materials for a product or major release: positioning, pitch narrative, discovery guide, objections, battlecards, demo script and environment, and pricing guidance. See Chapter 27.
End-of-life (EOL): The final product phase. Its public announcement names the successor and milestone dates: end of sale, last shipment or term, end of support and decommissioning. See Chapter 30.
End-of-sale: The lifecycle milestone after which a product takes no new orders, though renewals of existing subscriptions may continue. In the book’s retirement gates it is R2. See Chapter 30.
Engagement: The frequency, intensity or depth of use over a period. Kerry Rodden and colleagues recommend reporting it as an average per user, never as a total, because totals rise with headcount. See Chapters 05 and 09.
Envelopment: A platform entering another market by bundling the target’s functionality with its own and using existing customer relationships. Thomas Eisenmann, Geoffrey Parker and Marshall Van Alstyne described it in 2011. See Chapter 14.
Error analysis: Hamel Husain and Shreya Shankar’s method for starting AI evaluation: collect representative production traces, have a domain expert write notes on what went wrong and group the notes into a failure taxonomy. See Chapters 05 and 31.
EU AI Act: The European Union’s AI regulation, in force since August 2024. It classifies systems by intended use and assigns duties by role. Categories cover prohibited practices, high-risk systems, transparency-only systems and minimal risk. Check current obligation dates with counsel. See Chapters 31 and 33.
Evals: Repeatable tests of an AI system’s outputs against defined criteria, with pass-rate targets. The book treats them as the executable form of acceptance criteria and as the gate for every model, prompt or retrieval change. See Chapters 11 and 31.
Event artifacts: Artifacts created for a specific decision or moment and then archived, such as PRDs, PR/FAQs, launch plans, experiment readouts, business cases and sunset plans. See Chapters 00 and 37.
Exchange model: Allan Cohen and David Bradford’s idea, in Influence Without Authority, that you get cooperation from people you do not manage by offering something they value in return. They call the things people value currencies. See Chapter 36.
Expansion: Revenue growth from existing customers through more seats, usage or products. It is why NRR can exceed 100% and why simple LTV formulas break. See Chapters 08 and 09.
Exploitation and exploration: James March’s two kinds of organizational learning. Exploitation refines what the company already does. Exploration searches for what it could do. Both compete for the same people and money, and exploitation pays back sooner. See Chapter 23.
Extended support: Paid support for a product past its standard end of support. The book treats it as a bridge for customers who need more time to migrate and as a lever to encourage migration. See Chapter 30.
F
Failure taxonomy: A named list of the ways an AI product fails, built by clustering notes from error analysis. It becomes the structure of the eval suite and often the team’s backlog. See Chapter 31.
Fat tail: A distribution in which extreme outcomes are far more common than a normal curve predicts. Bent Flyvbjerg’s work shows that software investments have fat-tailed cost overruns. See Chapter 13.
Feasibility prototype: Just enough code to retire a technical risk, such as a new technology, an algorithm or a performance question. It may have no interface, and engineers build it. See Chapter 04.
Feasibility risk: One of Marty Cagan’s four risks: can engineering build it with the time, skills and technology available? The tech lead is accountable for it. See Chapters 00 and 04.
Feature factory: John Cutler’s term for a team that ships features without measuring impact. His 2016 signs include celebrating launches, skipping iteration and disconnecting work from core metrics. See Chapter 07.
Feature flag: A switch that changes system behavior without a deploy. Pete Hodgson sorts flags (toggles) into release, experiment, ops and permissioning types, with very different lifespans. See Chapters 06 and 25.
Feature team: In Marty Cagan’s taxonomy, a team handed a roadmap of features with dates, accountable for output, with value and viability left to the stakeholder who asked for the work. See Chapter 36.
Fidelity: How closely a prototype resembles the final product. NN/g’s Kara Pernice treats it as three independent dials: interactivity, visuals and content. See Chapter 04.
File mode: A written grilling session. The harness proposes questions and recommended answers for a dimension. Untouched recommendations count as accepted. This mode suits factual questions and input from other people. Compare live mode. See Chapter 37.
Five Forces: Michael Porter’s model of how industry structure splits the value a product creates. The five forces are rivalry among existing competitors, the threat of new entrants, buyer power, supplier power and the threat of substitutes. For AI products, model providers often act as suppliers and general-purpose AI harnesses as substitutes. See Chapter 13.
Five-question check: This book’s test before trusting a number, adapted from Barr Moses’s five pillars of data health: freshness, quality, volume, schema and lineage. See Chapter 18.
Flow metrics: The four measures the Kanban Guide requires: work in progress, throughput, work item age and cycle time. See Chapter 11.
Force multiplier: Marty Cagan’s framing of product operations: it helps product management and design work better while preserving their decision authority. See Chapter 28.
Four forces of progress: Bob Moesta and Chris Spiek’s explanation of switching. The current situation’s push and a new solution’s pull must outweigh anxiety and existing habits. See Chapter 01.
Four risks: Marty Cagan’s four risks of any product idea: value, usability, feasibility and business viability. The team owns all four together, with the PM accountable for value and viability. See Chapters 00 and 04.
FP&A (financial planning and analysis): The finance function that owns plans, forecasts, cost allocation and the conventions of the business case model. The PM owns the assumptions and their evidence. FP&A owns the model’s mechanics and the discount rate. See Chapter 13.
Fundamental rights impact assessment (FRIA): An assessment some deployers owe under EU AI Act Article 27 before first use of a high-risk system. These include public bodies and deployers using AI for credit scoring or life and health insurance pricing. See Chapter 33.
Funnel: An ordered set of steps with the conversion rate between each pair. It shows where people drop out. Recordings, tickets and conversations tell you why. See Chapters 05 and 08.
G
General availability (GA): The release stage at which a product is fully tested, open to all users, production-ready and backed by SLA and support obligations. See Chapters 25 and 26.
General Data Protection Regulation (GDPR): The EU data protection law. Purpose limitation and data minimization govern personal data in prompts, logs and training. Article 35 requires a DPIA when processing is likely to create high risk to people. See Chapters 19 and 33.
General-purpose AI model (GPAI model): Under the EU AI Act, a model with a separate track of obligations from the AI systems built on it. In a typical B2B stack the model lab is the GPAI model provider. See Chapter 33.
Generative AI Profile: NIST AI 600-1, published in July 2024, which tailors the NIST AI Risk Management Framework to generative AI and lists twelve risks it creates or worsens, including confabulation. See Chapter 33.
Go/no-go review: The launch meeting at which a named decider confirms that every function’s readiness criteria are met before release. The decider depends on the launch tier. See Chapter 25.
Goals, Signals, Metrics: The process that accompanies the HEART framework: name the goal, find the user signals that show progress toward it, then choose metrics that track those signals. See Chapter 05.
Good, better, best: A three-tier packaging pattern in which each tier adds capability for a higher price. See Chapters 12 and 15.
Grilling session: The Capstone’s method: a harness asks one question at a time, proposes evidence-based answers, challenges weak reasoning and records the decisions. See Chapter 37.
Gross margin: Revenue minus the direct cost of delivering it, as a share of revenue. AI features carry inference cost, which makes gross margin a product variable the PM has to watch. See Chapters 08 and 31.
Gross revenue retention (GRR): Starting recurring revenue minus contraction and churn, divided by starting recurring revenue. It can never exceed 100%. See Chapters 09 and 27.
Growth-share matrix: BCG’s model for managing products as a portfolio, sorting businesses by market growth and relative share into stars, cash cows, question marks and a low-growth, low-share quadrant. Practitioners turn it into four postures: invest, maintain, harvest and divest. See Chapter 29.
Guardrail metric: A metric that must not get worse while you push another one. In experiments it protects against a treatment that lifts the OEC and breaks something else. In a metric tree it is a counter-metric with a threshold. See Chapters 02 and 05.
H
Hackathon: A short event in which employees build prototypes on a theme. Hackathons are cheap and fast at surfacing problems, feasibility and talent. Their weakness is continuity, so the book requires a receiving owner before one starts. See Chapter 22.
Hallucination: A fluent, confident AI output that is false, such as an invented root cause. Investigate recurring failures by segment and severity. Add confirmed cases to evals. See Chapters 06 and 29.
Harness: The AI environment you work in: a chat application, desktop agent, coding tool or assistant built into company tools. Its parts include projects, connectors and skills. See Chapters 19 and 32.
Hazard: In survival analysis, the chance that a customer who has made it to a given period leaves during that period. The survival curve tells the cumulative story. The hazard shows where risk concentrates. See Chapter 09.
HEART: A user-experience metrics framework developed at Google by Kerry Rodden, Hilary Hutchinson and Xin Fu: happiness, engagement, adoption, retention and task success. See Chapter 05.
High-risk system: Under the EU AI Act, a system in an Annex I product-safety category or an Annex III use area. High-risk systems carry the heaviest obligations. See Chapter 33.
Human in the loop: A person reviews or approves AI output before it takes effect. Design the review to resist automatic approval and measure whether reviewers catch errors. See Chapter 31.
Hybrid pricing: A pricing model that combines a fixed component, such as a platform fee or seats, with a usage or outcome component. The book’s comparison of pricing models lists it as the fit for most B2B AI products. See Chapter 15.
I
ICE: Sean Ellis’s prioritization score for growth experiments: impact times confidence times ease, each on a relative 1 to 10 scale. It is fast, but all three numbers are opinions. See Chapter 03.
Incrementality: The part of an outcome that would not have happened without an activity, measured only against a counterfactual group that did not get it. See Chapter 08.
Inner loop: In Fred Reichheld’s Net Promoter System, the loop in which feedback reaches front-line employees directly and quickly and they follow up with the customer. See Chapter 27.
Innovation accounting: Eric Ries’s way of measuring progress on a new venture in three milestones: establish the baseline, tune the engine and decide whether to pivot or persevere. See Chapters 05 and 21.
Innovation ambition matrix: Bansi Nagji and Geoff Tuff’s model, published in HBR in 2012, that classifies bets by where to play and how to win into core, adjacent and transformational. Their benchmark was roughly 70/20/10 in spend. See Chapters 20 and 23.
Innovation placemat: A one-page innovation plan for a single bet or a small portfolio of related bets, small enough for a sponsor to read in five minutes. See Chapter 20.
Input metrics: Three to five complementary factors that most directly drive the North Star and that a team can influence through the product. Amplitude suggests four lenses for finding them: breadth, depth, frequency and efficiency. See Chapters 02 and 29.
Inside view: Daniel Kahneman and Dan Lovallo’s term for a forecast built from the details of the project at hand. It is reliably optimistic. Compare outside view. See Chapter 13.
Insight: A statement of why something happened and what follows from it, built on a finding. The book’s insight brief adds the evidence and buying-group context that a design-thinking point-of-view statement leaves out. See Chapter 10.
Insight brief: A one-page format for carrying an insight to decision-makers. It follows the Pyramid Principle: lead with the governing point and the decision it informs. See Chapter 10.
INVEST: Bill Wake’s 2003 checklist for a well-shaped story: independent, negotiable, valuable, estimable, small and testable. See Chapter 06.
ISO/IEC 42001: The first certifiable management system standard for AI, published in 2023. It asks an organization to run AI with a policy, roles, risk and impact assessments, controls, audit and continual improvement, much as ISO/IEC 27001 does for information security. See Chapters 31 and 33.
J
Jagged frontier: Ethan Mollick’s term for the uneven boundary of AI capability: strong at some tasks and weak at others that look equally hard to a person. You find the line only by using it. See Chapter 32.
Job map: Tony Ulwick and Lance Bettencourt’s universal sequence of steps any job can be broken into: define, locate, prepare, confirm, execute, monitor, modify and conclude. See Chapter 01.
Job story: Alan Klement’s alternative to a persona-based user story that separates the situation and motivation from the solution: when [situation], I want to [motivation], so I can [expected outcome]. See Chapter 01.
Jobs to be done (JTBD): The idea that customers hire products to make progress in a situation. The book uses two schools: Moesta, Spiek and Klement’s focus on progress and switching, and Ulwick’s treatment of the job as a measurable process. See Chapter 01.
Join key: The identifier linking the same account or user across telemetry, CRM, billing and support. Consistent joins make cross-system B2B analysis possible. See Chapter 18.
Journey funnel model: The book’s model that records, for each journey stage, the buyer’s question, the persona who owns it, the touchpoints and the measures, including stages you cannot observe. See Chapter 08.
Journey map: A visualization of the process a person goes through to accomplish a goal. Sarah Gibbons of NN/g lays it out in three zones: the lens, the experience and the insights. See Chapter 01.
K
Kanban: In the Kanban Guide, a strategy for optimizing the flow of value through a process by visualizing the workflow, actively managing items and improving the workflow. It suits teams with interrupt-driven demand. See Chapter 11.
Kano model: Noriaki Kano and colleagues’ 1984 classification of features by how their presence and absence affect satisfaction: must-be, performance, attractive and indifferent. Run it per persona. See Chapter 03.
Key results: In John Doerr’s form of OKRs, the specific, time-bound measures of progress toward an objective, either met or not. A typical set has three to five per objective. See Chapter 02.
Kill criteria: Conditions agreed before a bet, pilot or price change starts that determine when it stops. See Chapters 20 and 23.
Kill line: This book’s state-and-date form for kill criteria: we stop [bet] if by [date] we have not shown [state], unless [named exception], and [forum] decides. Every funded bet carries one. See Chapter 23.
Kill switch: A global control that disables an AI system’s actions immediately. For an agent that acts, the book treats it as part of the minimum viable product. See Chapters 06 and 31.
L
Latency budget: The maximum acceptable response time for an AI feature, stated as time to first token and end to end. It is a section of the AI PRD. See Chapter 31.
Launch tiers: A three-level sizing of launches by customer and business impact, not engineering effort, that sets the announcement, readiness work and go/no-go decider. Tier 1 covers new products, SKUs or prices. Tier 3 covers fixes and minor changes. See Chapter 25.
Lean Startup: Eric Ries’s method, set out in 2011 and built on Steve Blank’s customer development, in which validated learning is the unit of progress and teams run build, measure, learn loops. See Chapter 21.
Lean UX hypothesis statement: Jeff Gothelf and Josh Seiden’s format linking a feature to two outcomes: we believe that [business outcome] will be achieved if [user] attains [benefit] with [feature]. See Chapter 01.
Lifetime value (LTV): In the standard SaaS formula the book cites, average revenue per account times gross margin, divided by the churn rate. With NRR above 100% it runs toward infinity, so cap the lifetime. Compare customer lifetime value. See Chapter 08.
Limited availability (LA): A release stage that is public but offered on some platforms only or at reduced scale. It is fully supported and documented and held to the same security requirements as GA. See Chapter 25.
Live mode: A grilling session run as a conversation, one question at a time, suited to judgment-heavy dimensions such as strategy, business and leadership. Compare file mode. See Chapter 37.
Living artifacts: Artifacts updated weekly, monthly or quarterly, such as roadmaps, dashboards, OKRs, risk registers and decision logs. Review them on their assigned cadence. See Chapter 00.
LLM-as-judge: Using a model to grade another model’s outputs. The book’s rule is one narrow judge per failure mode, validated against human labels. See Chapter 31.
Loss aversion: The tendency for losses to loom larger than equivalent gains. The book rates the evidence real but conditional and applies it to framing migrations, renewals and downgrades. See Chapter 10.
M
Marketing mix modeling (MMM): A top-down attribution method that relates outcomes to aggregate spend by channel over time, controlling for seasonality and trend. It needs no user-level data but is slow and struggles with small B2B conversion volumes. See Chapter 08.
Marketing qualified lead (MQL): A lead whose intent is inferred from signals such as downloads, event registrations and page visits. It is followed by the sales accepted lead (SAL), the sales qualified lead (SQL) and the opportunity. See Chapter 08.
Message house: A practitioner tool for organizing messaging: a roof with one umbrella message, three pillars of supporting messages and a foundation of proof. It has no single recognized originator. See Chapter 24.
Messaging: The external, persona-specific words that express positioning: a core value proposition, the customer’s pains, a benefit for each and proof points. See Chapter 24.
Metric dictionary: A governed definition for each recurring metric, including owner, type, thresholds and version. Resolve conflicting definitions with the metric owners. See Chapters 18 and 19.
Metric review: A quarterly review, separate from the weekly business review, of whether metrics still guide useful decisions. Check owners, actions, versioned definitions and normal ranges. See Chapter 29.
Minimum detectable effect (MDE): The smallest effect an experiment can reliably detect at its sample size. With only hundreds of accounts it is often larger than any plausible result, which means the test should not start. See Chapter 05.
Minimum viable 360: The spine of about twelve artifacts most products need. Create additional artifacts when a decision requires them. See Chapters 00 and 37.
Minimum viable product (MVP): Eric Ries’s term for the version of a new product that lets a team collect the most validated learning about customers with the least effort. It is the fastest route to learning, which is different from the smallest product you can ship. See Chapter 21.
Moat: A lasting advantage protecting returns from competitors through pricing power or lower cost. This handbook tests moats with Hamilton Helmer’s 7 Powers. See Chapter 16.
Model and vendor register: The artifact listing which AI models and versions are in use, for which features, with their deprecation dates. It belongs to the Lifecycle dimension of the Product 360. See Chapters 00 and 30.
Model Context Protocol (MCP): An open standard, introduced by Anthropic in November 2024, for connecting AI applications to external data, tools and workflows. Most connectors are built on it, and a product’s MCP server is a priced surface. See Chapters 14 and 19.
Model deprecation runbook: A process for model retirement: inventory dependencies, monitor notices, set an early internal deadline, evaluate replacements, then canary and migrate with a rollback path. See Chapter 30.
Mom Test: Rob Fitzpatrick’s three rules: discuss the customer’s life, ask about specific past behavior and listen more than you talk. Commitment of time, reputation or money provides the signal. See Chapter 01.
MoSCoW: A prioritization method from the DSDM agile framework that sorts scope within a timebox into must have, should have, could have and won’t have this time. DSDM’s guidance keeps must haves to about 60% of estimated effort. See Chapter 03.
Multi-homing: Participants joining competing networks at the same time because it is cheap to do so. Andrei Hagiu separates multihoming costs from switching costs. See Chapter 14.
Multi-touch attribution (MTA): A bottom-up attribution method that assigns credit across the touchpoints it can observe before a conversion. It is fast and granular but correlational, and it sees only tracked touches. See Chapter 08.
N
Net present value (NPV): The sum of a project’s future cash flows discounted to today at the hurdle rate, minus the upfront investment. A positive NPV means the investment is expected to return more than the hurdle. See Chapters 12 and 13.
Net Promoter Score (NPS): Fred Reichheld’s 2003 measure. Customers rate recommendation likelihood from zero to ten. Subtract the percentage of detractors (zero to six) from promoters (nine and ten). Report by account and role alongside retention. See Chapter 09.
Net revenue retention (NRR): Starting recurring revenue plus expansion, minus contraction and churn, divided by starting recurring revenue. The book treats it as a product metric because it summarizes whether the product keeps delivering value after the sale. See Chapters 09 and 27.
Network effect: In NFX’s definition, the property that each new user makes the product more valuable to every other user. The book separates it from scale effects, learning effects and switching costs, which also make a product stronger as it grows but behave differently. See Chapter 14.
News embargo: An arrangement to share news with a journalist in advance on condition that it is not published until a set time. It is usually a matter of trust rather than contract. See Chapter 26.
No decision: A lost deal in which the buyer chose no vendor and stayed with the current way of working. In mid-market software it is often the top loss reason. See Chapter 13.
North Star metric: The single metric that best captures the core value a product delivers to its customers, as Sean Ellis and Morgan Brown describe it. It sits on top of an input-metric tree. See Chapter 02.
Now, next, later: Janna Bastow’s roadmap format: current work, less-defined upcoming work and larger future problems. Organize items around problems and make uncertainty explicit. See Chapter 03.
O
Objection handling: How the product answers concerns that might block a purchase, such as support, onboarding, integrations, security reviews and certifications. April Dunford separates it from value, because it is rarely why anyone buys. See Chapters 24 and 27.
Objectives and key results (OKRs): A goal system in which an objective states what you want to achieve and key results measure progress toward it. It grew from Peter Drucker’s management by objectives, was adapted by Andy Grove at Intel and brought to Google by John Doerr. The book writes OKRs as behavior changes, not features. See Chapter 02.
One-way door: Jeff Bezos’s Type 1 decision: consequential and irreversible or nearly so. It deserves slow, careful deliberation. Compare two-way door. See Chapters 17 and 29.
OP1 and OP2: The two passes of Amazon’s annual operating plan as Colin Bryar and Bill Carr describe it. Teams write OP1 plans as narratives in the autumn, and OP2, produced in January after review against fourth-quarter actuals, becomes the plan of record. See Chapter 29.
Open innovation: Henry Chesbrough’s 2003 argument that firms can use outside ideas as well as their own, and outside paths to market as well as internal ones. Inbound brings outside knowledge in. Outbound takes unused inside knowledge out. See Chapter 22.
Opportunity assessment: Marty Cagan’s ten-question, one-page gate, first published in 2006, that a team answers before committing real discovery time. It starts with exactly what problem the product will solve. See Chapter 01.
Opportunity scoring: Tony Ulwick’s method for ranking customer outcomes rather than features: importance plus any positive gap between importance and satisfaction. See Chapters 01 and 03.
Opportunity solution tree: Teresa Torres’s structure for discovery. The desired outcome is the root, customer opportunities branch from it, solutions branch from opportunities and assumption tests sit under each solution. See Chapters 01 and 03.
Outcome: In Josh Seiden’s definition, a change in human behavior that drives business results. It sits between output (what the team makes) and impact (the business result). See Chapter 02.
Outcome-based pricing: Charging only when a defined result happens, such as a resolved ticket. It is closest to customer value and riskiest for the seller, and it needs an unambiguous, measurable outcome definition before the price. See Chapters 12 and 15.
Outcome-Driven Innovation (ODI): Tony Ulwick’s approach to jobs to be done, which treats the job as a process with measurable steps and desired outcomes at each step. See Chapter 01.
Outer loop: In Fred Reichheld’s Net Promoter System, the loop in which the company deals with changes the front line cannot make alone: products, processes and policies. Feature requests should reach the roadmap through it. See Chapter 27.
Outside view: Daniel Kahneman and Dan Lovallo’s term for a forecast that starts from how similar projects actually turned out. Anchor on it first. Compare inside view and reference class forecasting. See Chapter 13.
Overall evaluation criterion (OEC): The quantitative measure of what an experiment is trying to achieve, chosen to predict long-term value. Ron Kohavi and colleagues recommend agreeing a single OEC before the test. See Chapters 05 and 07.
P
p-value: The answer to a narrow question: if the change had no effect, how surprising would data like these be? It is not the probability that the hypothesis is true, so always ask for the effect size with a confidence interval as well. See Chapter 07.
pass@k: The probability that at least one of k attempts succeeds. It suits tasks where a user can pick the best of several options. See Chapter 31.
pass^k: The probability that all k attempts succeed. A task that succeeds 75% of the time succeeds three times in a row only about 42% of the time, so pass^k is the honest measure for agents. See Chapters 05 and 31.
Payback period: How long it takes until cumulative cash in exceeds cash out. Executives like it because it is simple, but it ignores everything after payback, so show it alongside NPV. CAC payback is the customer-level version. See Chapters 08 and 13.
Peeking: Checking an experiment repeatedly and stopping the moment it shows significance. Because significance assumes a sample size fixed in advance, peeking drives the false positive rate far above its nominal level. See Chapters 05 and 07.
Penetration pricing: Launching at a low price to win share fast, build scale or network effects and make the market less attractive for imitators. Compare skimming. See Chapter 12.
Performance engine: Vijay Govindarajan and Chris Trimble’s term for the existing operation of a company, built for repeatable, efficient work. A dedicated innovation team needs a deliberate partnership with it. See Chapters 20 and 23.
Persona: A description of a specific kind of person with specific goals, used to turn “the customer” into people a team can design for. The book uses proto-personas, qualitative personas and statistical personas, and in B2B includes the people who buy rather than use. See Chapter 01.
Pipeline: Marshall Van Alstyne, Geoffrey Parker and Sangeet Paul Choudary’s term for a business that creates value through a linear series of activities: design, make, sell and support. Compare platform. See Chapter 14.
Pivot: A change of hypothesis when experiments show the current one is wrong. In Lean Startup, every milestone ends in a decision to pivot or persevere. See Chapter 21.
Platform: A business that creates value by facilitating interactions between external producers and consumers, rather than making most of what is exchanged. Most B2B software companies are hybrids: a pipeline core product with a platform of integrations, partners and APIs around it. See Chapter 14.
Pocket price waterfall: The walk from list price through on-invoice discounts to invoice price, then through rebates, free months, free services, payment terms and credits to the pocket price the company actually keeps. See Chapter 15.
Portfolio review: A review that asks where the next dollar should go across product lines: what to invest in, maintain, harvest or divest. See Chapters 23 and 29.
Positioning: April Dunford’s definition: the decision about how a product is different and better than the alternatives, for a specific set of customers. Her method works through five components in order: competitive alternatives, differentiated capabilities, value, target customers and market category. See Chapters 10 and 24.
Positioning statement: Geoffrey Moore’s template: for a named target customer with a stated need, the product is a member of a named category that delivers a key benefit, unlike the primary alternative. See Chapters 02 and 24.
Post-launch review: A review held at a point named before launch, often 30, 60 or 90 days, that judges each assumption validated, invalidated or unclear against the success criteria. See Chapters 05 and 25.
Power (7 Powers): Hamilton Helmer’s term for the conditions that create the potential for persistent differential returns. Every Power needs both a benefit and a barrier. See Chapter 16.
Power (statistical): The chance that an experiment detects a real effect of the size you care about. An underpowered test often misses real effects, so a “no significant difference” result from it tells you little. See Chapter 07.
Power-interest grid: A stakeholder map, usually attributed to Aubrey Mendelow, with axes for power and interest. Its quadrants suggest managing closely, keeping satisfied, keeping informed or monitoring. See Chapter 36.
PR/FAQ: Amazon’s Working Backwards document, described by Colin Bryar and Bill Carr. A press release names the product, customer and benefit. External FAQs cover price, operation and support. Internal FAQs cover competition, economics and risks. See Chapters 04 and 34.
Pre-mortem: Gary Klein’s technique of imagining that a project has already failed and listing the reasons why. The book uses it for big bets and for every keynote demo. See Chapters 07 and 26.
Price elasticity of demand: The percentage change in quantity demanded divided by the percentage change in price, as an absolute value. Above 1, demand is elastic and a price rise lowers revenue. Below 1, it is inelastic. See Chapter 15.
Private preview: In GitHub’s release stages, a feature under heavy development, available by invitation, possibly undocumented and without an SLA or support. See Chapters 25 and 26.
Product 360: This book’s model of a product as nine dimensions (customer, strategy, business, build, measure, go-to-market, operate, lifecycle and leadership), each answered by an owned, current artifact. It also serves as the context an AI harness works from. See Chapters 00 and 37.
Product 360 index: The single page that links every artifact of a Product 360, names its owner and shows when it was last updated. See Chapter 00.
Product hypothesis: A statement of a product bet split into customer, problem and solution hypotheses, so each part can fail on its own. See Chapter 01.
Product life cycle (PLC): The pattern of an offering’s sales over time: an S-shaped rise followed by a plateau and a fall, in four classic stages of introduction, growth, maturity and decline. Theodore Levitt made it a management tool in 1965. The book warns that the curve can mislead. See Chapter 30.
Product operations: The function that makes good product decisions cheaper to reach. Melissa Perri and Denise Tilles describe three pillars: data and insights, customer and market insights, and process and governance. See Chapter 28.
Product Owner: The Scrum accountability for maximizing the value of the team’s work, developing the Product Goal and ordering the Product Backlog. The book separates it from the product manager role. See Chapter 11.
Product-led growth (PLG): In OpenView’s definition, a go-to-market strategy in which the product itself is the primary driver of acquisition, conversion and expansion. Blake Bartlett coined the term in 2016. See Chapter 08.
Product-qualified lead (PQL): A user or account whose behavior in the product signals readiness to buy. Start with a simple threshold and refine it toward the behaviors that best predict conversion. See Chapter 08.
Product requirements document (PRD): The decision record for a feature: problem, customer, scope, success measures and requirements a prototype cannot show. The prototype demonstrates behavior. The PRD records decisions. See Chapters 04 and 31.
Product trio: Teresa Torres’s term for a product manager, a designer and a software engineer jointly responsible for a shared outcome, who interview customers, map opportunities and test solutions together. See Chapters 01 and 04.
Product vision: The positive change a product brings about, and for whom, over several years. It changes only when the purpose itself proves wrong, and it can be written as a statement, a board or a narrative. See Chapter 02.
Progressive delivery: Rolling a change out in stages behind feature flags (internal users, design partners, an opt-in beta, customer cohorts, everyone), each with exit criteria that stop the rollout when they fail. See Chapter 06.
Prompt injection: An attack in which content read by a model, such as a customer email, web page or document, contains instructions the model then follows. See Chapters 32 and 33.
Prompt library: A team’s shared, versioned set of prompts for recurring work, each with an owner, the inputs it needs, the context to attach, what good looks like and how to check it. See Chapter 32.
Proto-persona: Jeff Gothelf’s persona built from a team’s assumptions in a short workshop, with no new research. It is used to expose disagreement and create hypotheses to test. See Chapter 01.
Provider: Under the EU AI Act, the organization that develops an AI system, or has one developed, and places it on the market under its own name. The SaaS company that builds a product on a model is usually the provider. See Chapter 33.
Public preview: In GitHub’s release stages, a feature ready for broader distribution and documented, but still without an SLA or support. See Chapters 25 and 26.
Pyramid Principle: Barbara Minto’s structure for business writing: one governing thought at the top, with supporting ideas grouped beneath it, so the reader gets the point before the detail. See Chapter 34.
Q
Qualitative persona: A persona built from 5 to 30 interviews coded into themes. NN/g calls it the best fit for most teams and most product decisions. See Chapter 01.
Quarterly business review (QBR): A quarterly review of whether bets and goals are on track. The book uses the term for both an internal review and a customer-facing review of the value a customer got. See Chapters 18 and 29.
Quote to cash: The sequence of business processes that turns an offer into revenue, from offer configuration, pricing and quoting through contracts, orders, billing, revenue recognition and renewal. An offer is ready when every stage can handle it on launch day. See Chapter 24.
R
RACI: A responsibility framework that assigns Responsible, Accountable, Consulted and Informed roles to tasks and deliverables. DACI assigns roles to a decision instead. See Chapter 28.
RAPID: Bain & Company’s decision-rights framework: Recommend, Agree, Perform, Input and Decide. Its Agree role is a sparing veto for people who must sign off, such as Legal or Security. See Chapters 28 and 29.
Red-teaming: Deliberately attacking an AI system to find ways it can fail or be misused before customers do. The book requires at least one expert and one AI-driven approach before beta, and every confirmed finding becomes an eval case. See Chapter 33.
Reference class forecasting: Bent Flyvbjerg’s method for taking the outside view: pick a class of comparable past projects, find the distribution of their outcomes and position your project within it. See Chapter 13.
Reference value: In economic value estimation, the price of the customer’s next best alternative: what they would actually choose if you did not exist. See Chapter 12.
Regression evals: Evals that protect what already works. They should sit near 100%, and a drop is an incident. See Chapters 28 and 31.
Relaunch: Repositioning a product when its market or capability has changed enough. April Dunford names three triggers, and the book adds two for AI-era products. See Chapter 10.
Release stages: The labels that tell customers what they can rely on, such as private preview, public preview, limited availability and general availability. The book treats each label as a contract. See Chapters 25 and 26.
Resulting: Annie Duke’s term for judging a decision by its outcome. It teaches teams to avoid risk and to rewrite history. See Chapter 17.
Retention: The share of users or accounts from one period still active in a later one. Read it by cohort. See Chapters 05 and 09.
Retention triangle: A table with one row per cohort and one column per period since start, each cell showing the share retained. Read it down columns, along rows and along diagonals. See Chapter 09.
RICE: Sean McBride’s Intercom score: reach times impact times confidence, divided by effort. It omits timing, so assess urgent items with WSJF as well. See Chapters 03 and 11.
Risk register: A list of what could go wrong, each item with an owner, a plan, a leading indicator and an escalation threshold. For AI products, write each risk as cause, event and consequence. See Chapters 20 and 29.
S
Sample ratio mismatch (SRM): An observed ratio of units between experiment groups that differs from the designed ratio, a sign that assignment, logging or eligibility is broken. No result should go out without an SRM check. See Chapters 05 and 07.
Scaled Agile Framework (SAFe): A framework for coordinating many agile teams, which groups them into Agile Release Trains that plan together in Planning Intervals of typically 8 to 12 weeks. It prioritizes with WSJF. See Chapter 11.
Scenario: A version of a business case that changes several linked assumptions together to tell a coherent story, such as a downside in which slower adoption also slows expansion. Compare sensitivity analysis. See Chapters 12 and 13.
SCQA: An opening for business writing: Situation (what the reader already knows), Complication (what changed), Question (what the complication raises) and Answer. Barbara Minto’s Situation, Complication, Question form earns the reader’s attention before the answer. See Chapter 34.
Scrum: A delivery framework defined by the Scrum Guide, with three accountabilities (Product Owner, Scrum Master and Developers) and work organized in fixed-length Sprints of one month or less. See Chapters 06 and 11.
Sean Ellis test: A survey that asks users how they would feel if they could no longer use the product. Ellis found that startups struggling for traction almost always had fewer than 40% answering “very disappointed.” See Chapter 05.
Semantic layer: The mechanism between raw data and every tool that uses it, in which each metric is defined once (measure, filters, time grain, dimensions) and served everywhere. It makes a single source of truth practical. See Chapters 18 and 19.
Sensitivity analysis: Changing one assumption at a time to find the two or three that flip a decision, such as an NPV from positive to negative. Put your evidence effort there. See Chapters 12 and 13.
Serviceable addressable market (SAM): The part of TAM your product can actually serve today, given its integrations, geographies, compliance posture, channels and price point. See Chapter 13.
Serviceable obtainable market (SOM): The part of SAM you can realistically win in a stated period, given sales and partner capacity, win rates and competition. The book builds it from go-to-market capacity rather than an assumed share. See Chapter 13.
Shape Up: Ryan Singer’s method, published by Basecamp, built around fixed time and variable scope: shaped pitches, a betting table, uninterrupted cycles, a circuit breaker and hill charts. See Chapter 11.
Single source of truth: One governed place for each metric definition and value, used by every report. See Chapter 18.
Six-page narrative: Amazon’s practice, described in Jeff Bezos’s 2017 shareholder letter, of replacing slide presentations with narratively structured six-page memos read silently at the start of a meeting. See Chapters 29 and 34.
Skills: Folders of instructions, scripts and resources that an AI harness loads only when a task calls for them. Anthropic introduced Agent Skills in October 2025 and published the format as an open standard in December 2025. See Chapters 19 and 32.
Skimming: Launching at a high price to sell to the buyers who value the product most, then stepping the price down to reach the next segment. Compare penetration pricing. See Chapter 12.
SMART goals: George Doran’s 1981 form for objectives: Specific, Measurable, Assignable, Realistic and Time-related. Later versions substitute Achievable, Relevant and Time-bound. The book notes its limits for innovation bets. See Chapter 20.
SMT lockup: Gibson Biddle’s pairing of each Strategy (a hypothesis) with a Metric (a proxy metric) and Tactics (the projects and experiments that carry it out). If the tactics ship and the metric does not move, the hypothesis was wrong. See Chapter 03.
Sorting effect: The pattern in which retention rate rises with tenure even when no individual customer becomes more loyal, because customers likely to leave go early and the survivors are mostly unlikely to leave. Peter Fader and Bruce Hardie describe it. See Chapter 09.
Stage-Gate: Robert Cooper’s model, first in print in 1988, that puts a gate before each stage of development. Gatekeepers who own the resources review deliverables against criteria and decide Go, Kill, Hold or Recycle. See Chapters 23 and 30.
Statistical persona: A persona built from qualitative research plus a survey of 100 or more respondents, ideally 500 or more, clustered into segments. It suits pricing, packaging and sizing decisions where segment size matters. See Chapter 01.
Story points: A relative estimate of the size of a story. Ron Jeffries traced them to early Extreme Programming teams that renamed their “ideal days” as points. The book notes they lose meaning when coding agents shrink build time. See Chapters 06 and 11.
Strategy kernel: Richard Rumelt’s three-part core of a good strategy: a diagnosis, a guiding policy and coherent actions. See Chapter 03.
Strategy one-pager: The handbook’s nine-field strategy artifact. It covers diagnosis, winning aspiration, where to play, how to win, guiding policy and refusals. It also records coherent actions, capabilities and systems, necessary assumptions, and outcome metrics. See Chapters 00 and 03.
Stream-aligned team: In Team Topologies, a team that delivers customer value along one value stream and owns its outcomes. Product areas should map to streams so each team can own a metric end to end. See Chapters 02 and 04.
Sunset decision memo: The document that recommends retiring a product, stating the recommendation in one sentence with dates and answering why now, who is affected, what the options are and how customers will migrate. See Chapter 30.
Survival curve: A curve showing the share of a starting group still with you after each period. It starts at 100% and can only stay flat or fall. Edward Kaplan and Paul Meier’s 1958 estimate is a standard way to draw one when many customers have not left yet. See Chapter 09.
Sustaining innovation: In Clayton Christensen’s theory, innovation that makes good products better for an incumbent’s most demanding customers, whether incremental or a breakthrough. See Chapters 16 and 20.
Switching costs: The cost to a customer of leaving one product or platform for another. As one of Hamilton Helmer’s 7 Powers, it means customers stay because leaving is painful. The book keeps it separate from network effects. See Chapters 14 and 16.
Synthetic users: AI-generated respondents used to prepare research. Treat their output as hypotheses and validate it with real customers. See Chapters 01 and 21.
System 1 and System 2: The two modes of thinking Daniel Kahneman popularized: System 1 is fast, automatic and associative. System 2 is slow, effortful and deliberate. The book uses them as a model, not as literal brain systems. See Chapter 10.
System card: The product team’s record of an AI system: models, uses, data flows, tools, permissions and autonomy. Include eval results and known limitations. See Chapter 33.
T
Team charter: A one-page statement of why a team exists, what it owns, how it works and how it is measured. It settles ownership and decision rights before anyone needs them. See Chapter 36.
Team Topologies: Matthew Skelton and Manuel Pais’s model of four team types (stream-aligned, platform, enabling and complicated-subsystem) and three interaction modes (collaboration, X-as-a-service and facilitation). See Chapters 04 and 07.
Technical debt: Ward Cunningham’s 1992 metaphor for shipping code that is not quite right: a little debt speeds development if it is paid back promptly, and every minute spent on that code is interest. Martin Fowler’s quadrant classifies debt as deliberate or inadvertent and prudent or reckless. See Chapter 06.
Technical evaluator: The buying-group role that runs the security, IT or architecture review. Their goal is to keep the environment safe and supportable, and their fear is a risk they will be blamed for. See Chapters 01 and 10.
Technical product manager: In this book’s definition, a PM whose customers, product surface or dominant risks are technical, so they must reason about feasibility, architecture and operational tradeoffs directly instead of delegating them. See Chapter 07.
Ten types of innovation: Larry Keeley and colleagues’ Doblin framework, which groups successful innovations into ten types across configuration, offering and experience. The strongest enterprise moves often sit outside the offering, in profit model, network or product system. See Chapter 20.
Thinking aloud: A usability method in which participants say what they are thinking while they use the system. Jakob Nielsen calls it perhaps the single most valuable usability method. See Chapter 04.
Three horizons: Mehrdad Baghai, Stephen Coley and David White’s model of business maturity. Horizon 1 defends and extends the core. Horizon 2 builds emerging businesses. Horizon 3 creates future options. See Chapters 20 and 23.
Throughput: Items finished per unit of time. When AI shortens coding, measure verified outcomes alongside completed work. See Chapters 06 and 11.
Time to value: How long it takes a new customer to realize the product’s value. It is one of the standard product-led growth metrics and a core customer success measure. See Chapters 08 and 27.
Total addressable market (TAM): The revenue available if every possible buyer with the job bought a solution like yours. It answers how big the market could ever be. See Chapter 13.
Total cost of ownership (TCO): Every cost of acquiring, deploying, operating and eventually retiring a solution over its life. Subscription is often the smaller part. See Chapter 12.
Total Economic Impact (TEI): Forrester’s structure for business value studies, with four components: benefits, costs, flexibility (the value of future options) and risk (an adjustment that reduces benefit estimates). See Chapter 12.
Tracking plan: The contract between product and engineering that lists every event, exactly when it fires, its properties and types, the question it answers, the metric it feeds and its owner. See Chapter 05.
Transparency note: A document, added for Tier 1 AI launches, that explains how an AI system works, its capabilities, intended uses, limitations and performance, and the choices deployers make that affect its behavior. See Chapter 25.
Two-way door: Jeff Bezos’s Type 2 decision: changeable and reversible. It deserves a light, fast process, and starting or extending a pilot is a typical example. Compare one-way door. See Chapters 17 and 28.
U
Usability risk: One of Marty Cagan’s four risks: can users figure out how to use the product? The designer is accountable for it. See Chapters 00 and 04.
Usage churn: Churn in which the customer still pays while use fades: seats go idle, workloads move or the champion leaves. With usage pricing it reduces revenue with no renewal date to warn you. Compare contract churn. See Chapter 09.
User story: A need expressed as “As a [user], I want [something], so that [reason].” Rachel Davies introduced the form at Connextra. Mike Cohn popularized it. Include intent and use job stories when the situation matters. See Chapter 06.
User story map: Jeff Patton’s planning tool that lays stories out along a backbone of user activities, then slices them into releases, with the walking skeleton as the top slice. It replaces the flat backlog. See Chapter 06.
V
Value map: The half of the value proposition canvas that describes your offer: products and services, pain relievers and gain creators. See Chapter 12.
Value metric: The unit a price scales with, such as per user, per environment, per API call or per resolved ticket. Kyle Poyar proposed five tests for a good one, starting with whether it correlates with the value different customers get. See Chapter 15.
Value proposition canvas: Alexander Osterwalder and his Strategyzer co-authors’ tool, introduced in 2014, with two halves filled in separately and then compared: a customer profile (jobs, pains and gains) and a value map. See Chapter 12.
Value risk: One of Marty Cagan’s four risks: will customers buy the product, or choose to use it? The PM is accountable for it. See Chapters 00 and 04.
Value versus effort: A two-by-two that sorts ideas into quadrants such as quick wins and big bets. It has no single originator, makes a good first cut in a workshop, but both axes are guesses. See Chapter 03.
Van Westendorp: A pricing research question set that asks at what price a product would seem so cheap buyers doubt it, a bargain, getting expensive and too expensive to consider. See Chapter 15.
Vanity metrics: Eric Ries’s term for metrics that make you feel good but do not tell you what to do, such as cumulative sign-ups or total page views. They usually only go up. See Chapters 17 and 21.
Venture client: A model, associated with Gregor Gimmy and BMW, in which a company buys and uses a startup’s product for a strategic purpose without taking equity. See Chapter 22.
Verification standard: Written checks for AI-generated work before people rely on it. A PM’s verification standard functions as a small evaluation system. See Chapter 32.
Vision narrative: The long form of a product vision: a page or two describing the customer’s world once the product has done its job, in the tradition of Amazon’s Working Backwards practice. See Chapter 02.
Voice-of-customer program (VoC): A system connecting surveys, tickets, calls and telemetry to product decisions through a shared theme taxonomy and routes to responsible teams. See Chapters 09 and 28.
W
Walking skeleton: The smallest end-to-end version of a system, built first. It is the top slice of a user story map. See Chapter 06.
Waterfall: The sequential model of requirements, design, implementation, verification and maintenance in strict order, usually traced to Winston Royce’s 1970 paper. It still fits some fixed-commitment work, often with an iterative team inside. See Chapters 07 and 11.
Weekly business review (WBR): Amazon’s weekly metrics forum, described by Colin Bryar and Bill Carr, in which owners walk input metrics and explain exceptions. The book designs its metric pack around inputs first. See Chapters 18 and 29.
Weighted shortest job first (WSJF): SAFe’s prioritization rule: relative cost of delay divided by relative job duration. Cost of delay combines user and business value, urgency, and risk reduction or opportunity enablement. See Chapters 03 and 11.
Win-loss analysis: Interviewing buyers after deals close, both wins and losses, to find out what actually drove the decision. A neutral interviewer gets the real reasons more often than the seller does. See Chapters 13 and 27.
WIP limit: A cap on the work in progress in a stage or a whole system. By Little’s Law, holding throughput steady and halving WIP halves average cycle time. See Chapter 11.
Wizard of Oz: A test in which the participant uses an interface that appears automated while a hidden person produces some or all of the responses. It tests whether customers use and trust an experience before it is built. See Chapters 04 and 21.
Words of estimative probability: Sherman Kent’s 1964 proposal to assign numeric ranges to uncertainty phrases. It addressed differing interpretations of terms such as “probable” and “serious possibility.” See Chapter 35.
Workflow: In Erik Schluntz and Barry Zhang’s distinction, an AI system that orchestrates models and tools through predefined code paths. In pricing, per workflow means charging per completed multi-step task. See Chapters 15 and 31.