Glossary

The process by which a business integrates AI capabilities into its operations, products, or decisions such that they change how the business actually works — not just that an AI tool has been deployed.

The distinction matters. Most "AI adoption" is really AI deployment — the tool exists, but work hasn't changed. Real adoption means the process, the decision, or the output is genuinely different because of AI. Usage metrics don't capture this. Outcome metrics do.

AI systems that operate with greater autonomy, executing multi-step tasks, making intermediate decisions, and using tools to accomplish goals — rather than responding to a single prompt and stopping.

Agentic AI is where the hype currently lives, and where the governance gaps are largest. The value is real — but so is the risk of systems taking consequential actions without adequate human oversight. The relevant question for most businesses is not "should we use agentic AI?" but "where, specifically, is a human checkpoint required?"

The ability of a business leader to understand what AI can and can't do well enough to make sound strategic and operational decisions — without needing to be a technical practitioner.

AI fluency is not about writing prompts or understanding model architecture. It is about knowing how to evaluate vendor claims, where AI outputs require human review, when to build vs. buy, and how to hold an AI programme accountable. Most executives are further behind on this than they realise.

The execution of a predefined task or process using software, without human involvement in individual instances. Automation follows explicit rules: if X, then Y.

Automation and AI are frequently conflated, but they are different tools for different problems. Automation is deterministic — it does exactly what you tell it. AI is probabilistic — it makes judgments. Most businesses would benefit more from better automation before adding AI. The right sequence: stabilise the process → automate the stable parts → use AI for what remains that requires judgment.

An individual who provides strategic advice and expertise to a company's leadership but has no legal fiduciary responsibility and is not a director of the company.

Board advisors are often confused with Non-Executive Directors. The distinction matters: advisors advise, directors have legal duties. An advisor can be brought in quickly, compensated with small equity or a retainer, and discharged easily. A NED carries legal responsibilities and represents a more significant governance commitment.

The strategic decision between developing a software capability in-house versus purchasing a commercial solution. The decision turns on differentiation, speed, cost, and whether the capability is core to competitive advantage.

The most common mistake is building what doesn't differentiate (commodity functionality at significant cost) and buying where differentiation is critical (limiting what's possible to what the vendor allows). Build where the way you do something is a competitive advantage. Buy where the category problem is solved well enough by existing solutions.

The structured approach to transitioning individuals, teams, and organisations from a current state to a desired future state — encompassing communication, training, process redesign, and ongoing reinforcement.

Change management is the most consistently underfunded element of digital and AI programmes. It is often treated as a communications task (tell people about the change) rather than a design task (redesign the work so the change actually happens). Technology goes live; adoption requires change management.

A C-suite executive responsible for leading the digital strategy of a business — typically encompassing digital products, digital operations, data, and increasingly AI adoption. Distinct from a CTO (technology infrastructure and engineering) and a CMO (marketing).

The CDO role has evolved significantly. In many businesses it started as a marketing role (digital channels), became a technology role (digital transformation), and is now increasingly an operational role (AI-powered operations). The scope varies widely by company — what matters is clarity on what the role owns and the authority to act on it.

The set of policies, processes, and accountabilities that determine how data is collected, stored, maintained, and used across an organisation. It answers: who owns what data, who can access it, how it stays accurate, and what rules apply to its use.

Data governance is unglamorous and frequently deferred. It becomes critical when AI is introduced — because AI systems inherit and amplify data quality problems. An AI model trained on or operating against poorly governed data produces unreliable outputs. Most businesses discover this after deployment.

A fundamental change in how a business operates and delivers value, enabled by digital technology. Distinct from digitisation (converting analogue to digital) and digitalisation (using digital tools in existing processes). Transformation means the business model, operating model, or competitive position changes.

Digital transformation has become one of the most abused terms in business. Deploying new software is not transformation. Transformation happens when digital capability changes what the business can do, who it can serve, or how it competes — not just how it executes existing processes.

Digitisation is converting something from analogue to digital — a paper form becomes a PDF, a handwritten record becomes a database entry. Digitalisation is using digital tools and data to improve or automate an existing process. Neither is transformation on its own.

The distinction matters for scoping work and setting expectations. Many "digital transformation" programmes are actually digitisation projects. That is not a criticism — digitisation has real value — but calling it transformation sets the wrong expectations and often leads to disappointment when operational costs don't fall as dramatically as hoped.

A senior executive (CDO, CTO, CMO, CPO, etc.) who works with a business on a part-time or retained basis — typically 1–3 days per week — providing strategic leadership and operational input without a full-time employment commitment.

Fractional leadership makes most sense during specific phases: building a function before it justifies a full hire, navigating a transition, or accessing expertise the business needs periodically but not continuously. The failure mode is using fractional leadership as a budget compromise when what the business actually needs is a committed full-time leader.

AI systems that produce new content — text, images, audio, code, video — by learning patterns from training data and generating new outputs that match those patterns. Large language models (LLMs) are the most widely deployed category.

Generative AI is powerful and genuinely useful for a wide range of business tasks. It is also probabilistic — meaning it can produce confident-sounding incorrect outputs (hallucinations). Any business deployment needs to design for this: where are human review checkpoints? What happens when the output is wrong? The answer should be part of the architecture, not an afterthought.

When an AI model generates output that is factually incorrect, fabricated, or unsupported by its inputs — often with high apparent confidence. A structural characteristic of probabilistic language models, not a bug that can be fully eliminated.

Hallucination is not a defect in any one model — it is a property of how large language models work. The practical implication for business deployments: treat all AI-generated factual claims as requiring verification in high-stakes contexts. Design workflows where a wrong output is visible and correctable before it causes harm.

A specific, measurable metric that reflects performance against a strategic objective. KPIs answer: are we moving in the right direction on the things that matter most?

The most common KPI failure is tracking metrics that are easy to measure rather than metrics that reflect real progress. Revenue growth is a KPI. Website sessions is a metric that may or may not be a useful proxy. A good KPI is directly connected to a strategic outcome, has a clear baseline, and has a target that represents meaningful progress — not just improvement.

A type of AI model trained on large quantities of text data to understand and generate human language. The foundation for tools like ChatGPT, Claude, and Gemini, as well as many enterprise AI features.

LLMs are the most commercially deployed AI technology of this era. Their strength is natural language: summarising, drafting, classifying, extracting, answering questions across a broad range of topics. Their weaknesses are consistent: they can be confidently wrong, they struggle with precise numerical reasoning, and their knowledge has a cutoff date. Business deployments need to design around these properties, not assume they don't apply.

A member of a company's board who is not part of the executive management team. Provides independent oversight, challenge, and strategic guidance. Has legal fiduciary responsibilities as a director of the company.

NEDs are most valuable when they bring genuine independence — the willingness to ask the uncomfortable question and the authority to have it taken seriously. The failure mode is the rubber-stamp NED: present at meetings, supportive of management, providing neither challenge nor insight. A NED who never disagrees is not providing independent oversight.

A goal-setting framework in which an objective describes where you are going and key results describe, specifically and measurably, how you will know when you've arrived. Widely used in technology companies to align teams around outcomes rather than activities.

OKRs are frequently implemented as a compliance exercise rather than a strategic tool. The tell: key results describe activities ("launch the feature", "complete the audit") rather than outcomes ("X% of users adopt the feature within 30 days", "no high-severity findings in audit"). If your key results would still be green even if nothing changed in the business, they need to be rewritten.

The transition of an AI (or any technology) project from a controlled, small-scale test environment to full operational deployment. One of the most commonly failed transitions in enterprise technology.

Pilots succeed in conditions that rarely generalise: self-selected enthusiastic users, clean scoped data, close expert oversight. The gap between pilot success and production readiness is where most AI projects stall or fail. The transition needs its own project plan, its own scope, and — critically — its own assessment of whether the conditions that made the pilot work actually exist at scale.

The degree to which a product satisfies strong demand in a specific market — indicated by organic growth, high retention, and users who would be genuinely disappointed if the product were unavailable. Not simply "people are using it."

Product-market fit is genuinely binary in its consequences: without it, growth is a grind that rarely unlocks at scale; with it, many distribution and retention problems become more tractable. The honest test is the "very disappointed" measure: what share of users would be very disappointed if the product went away? Below 40%, you likely don't have it.

A plan that communicates where a product is going, and why, over a given time horizon. A roadmap is a statement of strategic direction — it is not a commitment to deliver specific features on specific dates.

Most product roadmaps are fiction: a list of features sequenced by internal negotiation rather than customer insight, presented as a delivery plan with false precision. A good roadmap communicates the problem being solved, the hypothesis being tested, and the sequencing logic — not a Gantt chart of features that nobody believes in.

The use of AI tools by employees without organisational awareness, approval, or governance — analogous to shadow IT. Common in businesses where official AI tools are absent, slow to procure, or not fit for purpose.

Shadow AI is almost certainly happening in your organisation right now. Employees are using consumer AI tools to do their jobs faster. This creates real risks: confidential data in external systems, outputs without quality control, and no institutional learning. The solution is not prohibition — it rarely works and drives the behaviour underground. The solution is a defined framework: clear guidance on approved tools, acceptable data handling, and what AI-generated outputs require review.

The accumulated cost of shortcuts, deferred decisions, and workarounds in a software system. Like financial debt, it accrues interest — small compromises compound over time into significant constraints on the ability to change or extend the system.

Technical debt is not always a mistake. Sometimes incurring it is the right call — getting to market faster, validating a hypothesis before investing in the right solution. The problem is undisclosed and unmanaged technical debt: shortcuts taken without acknowledgement, never scheduled for remediation, quietly accumulating until the system becomes expensive to maintain and risky to change.

The combination of software systems, tools, and platforms a business uses to operate — from core operational systems (ERP, POS, CRM) through to data infrastructure, integration layers, and customer-facing applications.

The tech stack a business inherits often constrains what it can do next more than it enables it. Legacy systems create integration complexity, data silos, and change costs that are frequently underestimated. Any significant digital or AI programme needs an honest view of the stack it is building on — and a realistic assessment of what "good enough" looks like versus what needs to change.

The use of software to automate a sequence of tasks that would otherwise require manual human action — routing, approvals, notifications, data transfers between systems.

Workflow automation is one of the highest-ROI technology investments available to most businesses, and one of the most underused. The barrier is typically not cost — tools like Zapier, Make, and enterprise iPaaS platforms are broadly accessible — but the absence of documented, stable processes to automate. You cannot automate a process that isn't yet well-defined.