About Blog Business First AI Founder Freedom Book a conversation

Plain-English AI

Every vendor uses the same words. Few owners get a clean explanation of what they actually mean.

This section is a working glossary of the AI terms that show up in vendor pitches, news headlines, your stack setup, your risk conversations, and your bill. Each post answers four questions in plain English. What is it? Why does it matter for your business? Where will you actually meet it? When to ask vs ignore? Alongside the term explainers there are decision guides for the choices that follow, like SaaS vs self-hosted, general-purpose vs vertical, and human-in-the-loop vs full automation. Written for business owners by someone who builds with these tools every day.

A practice manager at a desk holding a printed proposal in one hand with an open laptop alongside

Zero-shot vs few-shot learning: when AI works on tiny data

A plain-English decision guide for UK SME owners. Start with zero-shot, escalate to few-shot if needed, and only consider fine-tuning if specific volume thresholds are crossed.

Two colleagues at an office table looking at a laptop dashboard with a printed quote beside them

What is AutoML? Why it matters for your business

AutoML automates the technical engineering of machine learning so a UK SME can build a churn or lead-scoring model in weeks for platform fees, not a £50,000 data science hire. A plain-English guide to what it does, where it pays back, and where it is genuinely the wrong tool.

A person at a desk reviewing two printed quotations side by side with a calculator and a notebook

What is edge AI? Why running AI locally matters for your business

A plain-English guide to edge AI. The architecture decision that changes who holds your client data, how fast your system responds, and whether the cloud bill keeps growing every quarter.

A person at a desk in a small office reviewing two printed vendor quotes alongside a laptop open to a search interface

What is a vector database? The infrastructure under modern AI search

A vector database stores meaning, not keywords, and powers AI search, RAG, and recommendation systems. For most UK SMEs the question isn't whether you need one. It's whether you need a dedicated one.

A woman at a meeting room table tapping a printed report with a pen, asking a question to a vendor presenter across from her

What is overfitting vs underfitting? The procurement gate before deployment

Overfit and underfit models fail in opposite directions, and the SME risk surface tilts heavily towards overfitting. Ten procurement questions that surface the failure mode before sign-off.

Two people at a desk reviewing a printed invoice next to a laptop showing a pricing page

AI inference vs training: why your AI bill behaves the way it does

Training is the upfront, capital-style spend that vendors absorb. Inference is the recurring, per-query cost that lands on your invoice. Confusing the two is the root of most SME AI budget overruns in 2026.

Two people at a small office meeting room table reading a printed procurement questionnaire, one holding a pen, the other pointing at a clause on the page

What is responsible AI? The procurement question that decides 2026 contracts

A plain-English guide to responsible AI for UK SME owners, covering the five international pillars, the regulatory baseline in 2026, and a five-step starter that puts a defensible answer in place inside three to six months.

Three colleagues at a meeting room table reviewing a printed regulation document with sticky tabs

What is the EU AI Act? Why it matters for UK businesses

A plain-English guide to the EU AI Act for UK SME owners. Why UK incorporation provides no shield, the four risk tiers, the provider versus deployer split, the 2026 deadlines, and a five-step starter compliance plan.

A woman at a desk reviewing printed vendor proposals next to an open laptop showing a CRM dashboard

What is reinforcement learning (and RLHF)? Why it matters for your business

A plain-English guide to reinforcement learning for business owners, why RLHF makes ChatGPT useful, and where RL already runs inside the software you already pay for.

Insurance broker managing director at her desk reviewing a quote document with claims paperwork stacked beside her

What is transfer learning? A plain-English guide for owners

Transfer learning is how a 30-staff UK firm gets a custom AI capability without a data science team. Here is what it costs, when to use it, and when to pick something simpler.

Two people in a meeting room reviewing a printed customer report and a laptop together

Supervised vs unsupervised learning: which one your data supports

The supervised versus unsupervised choice looks technical, but it is really a labelling-budget question. A practical decision guide for UK SMEs in 2026.

A managing director at her desk comparing three printed vendor quotes next to a laptop showing an inbox of supplier invoices

AI vs automation: which one your business actually needs

A plain-English buyer's guide for UK SMEs choosing between rule-based automation and AI. Different tools, different problem shapes, very different bills.

A person at a desk reviewing four printed vendor proposals laid out next to a laptop and a calculator

What is a transformer in AI? Why modern AI works the way it does

A plain-English guide to the transformer, the neural network architecture inside every commercial AI tool you will be pitched in 2026.

A practice manager on a phone at her desk with a laptop and notebook open in front of her

What is conversational AI? Why it matters for your business

A plain-English guide to conversational AI for owners, what it actually is, where it works, where it fails, and the questions to ask before you buy.

A production manager checking quality data on a tablet beside a working conveyor line in a small UK precision-engineering factory

What is computer vision? Why it matters for your business

A plain-English guide to computer vision for business owners, what it does, where you'll meet it, and the question that decides whether a project works.

A person at a desk comparing printed vendor proposals next to an open laptop

What is NLP (natural language processing)? Why it matters for your business

A plain-English guide to NLP for owner-managed businesses, what it is, where you already meet it, and why the newest tool is rarely the right one.

A woman at her desk reviewing a printed vendor shortlist with a laptop open beside her.

What is a neural network? Plain English for owners

A plain-English guide to neural networks for UK SME owners. The mental model, where you'll meet them in 2026 software, and the right procurement questions to ask vendors.

A person at a desk comparing two printed proposals with a notebook and pen, factory floor visible behind a glass partition

What is deep learning? Why it matters for your business

A plain-English guide to deep learning. The technology behind ChatGPT, image recognition and voice transcription, and the procurement question to ask before a vendor charges you to build one from scratch.

Two people at a meeting room table comparing printed proposals, one holding a pen and pointing at a price column.

Machine learning vs AI: which one do you actually need?

Four vendors. Four prices for the same problem. Here's how to read what each one is actually selling.

A managing director and a colleague reviewing three vendor proposals across a meeting-room table.

What is machine learning? A plain-English guide for business owners

Machine learning is software that learns patterns from your historical data to predict, classify, or score new data. Here is what that means for a UK SME, where it pays back, and where to ignore the pitch.

A woman at a desk reviewing a printed list while working at a laptop in a small office

What is generative AI? Why it matters for your business

A plain-English guide to generative AI for owner-managed businesses, what it is, where it pays back, and where it does not belong.

Two people at a desk reviewing two printed vendor quotes side by side with laptops open

RAG vs fine-tuning: which one your firm actually needs

A plain-English decision guide for UK SME owners. Default to RAG, add fine-tuning later, and reframe vendor pitches that lead with the wrong question.

A person at a desk reviewing two pricing pages on a laptop with a notebook and pen beside them

No-code vs custom AI build: which one your firm actually needs

A plain-English decision guide for UK SME owners. Default to no-code, plan for the migration moment, and architect for hybrid from day one.

Two colleagues talking across a meeting-room table with a laptop and a printed report between them

Chatbots vs workflow automation: which one your firm needs

A plain-English decision guide for UK SMEs picking their first serious AI deployment. Workflow automation first, chatbots second, with rare exceptions.

Two professionals in a meeting room comparing a hiring brief with a consultancy proposal on the table between them.

In-House AI Hire vs Outsourced: Which Your Firm Needs

The AI hiring decision turns on one question. Is AI a project or a continuing capability? Here is how to tell, and what each path costs in the UK 2026 market.

Three people in a meeting room reviewing a printed vendor quote and a laptop together

Build vs buy AI: which one your firm actually needs

Most boards default to build when the vendor already solves 75 to 85% of the problem at a fraction of the cost. A practical decision guide for UK SMEs in 2026.

A person at a meeting room table leaning forward with a pen, listening to a vendor demo on a laptop screen, a printed brochure beside them

What is model collapse? Why training data hygiene matters

Model collapse is the failure mode where AI trained on AI degrades across generations. Why 'trained on the latest open web' is a yellow flag, not a feature.

Two people at a desk looking at a laptop showing a referral traffic dashboard, with a printed report between them

What is GEO? When optimising for AI citations actually matters for your firm

A plain-English guide to generative engine optimisation. What it is, when it crosses from research interest to budget priority, and the in-house moves that pay back across both SEO and AI search.

Two people at a meeting room table reading a printed document, one pointing at a paragraph, a laptop open between them

What is digital sovereignty? When it actually changes your procurement

A plain-English guide to digital sovereignty for UK and European SME owners, and the three procurement contexts in 2026 where it stops being a slogan and starts deciding whether you win the deal.

Two people at a meeting-room table reviewing a printed forecast with a laptop open to one side

What are scaling laws? Why your AI bill behaves the way it does

A plain-English guide to scaling laws, the empirical curves that decide what frontier capability costs, why your AI invoice keeps moving, and how to match model tier to task without overpaying.

Two people leaning over a desk reading printed specification sheets with two laptops open in front of them

What is open-source vs closed-source AI? The procurement choice in front of you

A plain-English procurement frame for the open-vs-closed AI choice. Five practical factors, real 2026 numbers, and the deployment-mode call many owner-managed services firms end up making.

A person seated at a meeting-room table reviewing a printed proposal with a pen in hand and a laptop closed beside them

What is AI alignment? Vendor research vs vendor marketing

A plain-English guide to AI alignment. What the research field actually contains, what vendor brochures usually mean by it, and the procurement questions that turn the brochure claim into a contract clause.

A person at a meeting-room table reading a printed proposal with a pen in hand and a laptop closed beside them

What is a frontier model? When to pay frontier prices and when not to

A plain-English guide to frontier models, the small set of largest AI systems at the cutting edge, and the procurement question worth asking before you pay frontier prices.

A person at a meeting-room table speaking to two others with a laptop and notebook between them

What is AGI? Why the term should not enter your procurement decisions

A plain-English guide to AGI. What the labs actually mean, why the definitions disagree, and why the term should sit in your investor briefing rather than your buying decisions.

A person at a desk reviewing a long chat transcript on a laptop screen, notebook beside them in a small office

What is conversation history? Why your AI bill scales quadratically

AI models are stateless. Every multi-turn chat re-sends the whole transcript with each new question, and the cost curve bends quadratically. The plain-English version of why your bot's bill triples by month two.

A person at a desk reviewing a paper invoice next to a laptop showing extracted line items in a small office

What is structured output (JSON mode)? The death of regex-and-pray

Structured output forces an AI model to return data in a schema you define. Schema validity goes from 70-85% to 100%, and the regex-repair layer disappears.

A person at a desk reviewing a printed invoice next to an open laptop in a small office

What is chain-of-thought reasoning? When the premium earns its keep

Chain-of-thought asks the model to show its working before answering. The accuracy lift is real on hard problems, the cost is real on every other call. The plain-English version of when to flip reasoning on and when to leave it off.

A person at a desk in a small office reviewing a search interface on a laptop next to an open bound contract

What is an embedding (and a vector database)? The lock-in nobody named

Embeddings turn text into numbers that capture meaning. Vector databases store and search them. The catch nobody mentions at procurement: switching providers later means re-embedding the entire corpus.

Two colleagues at a shared monitor reviewing a list together, one pointing at the screen and the other writing on a printed sheet

What is function calling? The primitive under agentic AI

Function calling is how an AI model asks your business systems to do something rather than just talk about it. A plain-English guide to what it is, where it earns its keep, and the governance questions to settle before you switch it on.

A person sitting at a kitchen table comparing a printed page of notes with the screen of an open laptop

What is a system prompt vs a user prompt? Why the distinction matters

A system prompt is the standing instruction your AI sees before every conversation. A user prompt is what your team types in. The plain-English distinction, why it controls cost, governance and IP, and what to ask your vendors.

Two colleagues at a desk reviewing a long list of authorised apps on a laptop screen, one pointing at the screen, the other writing notes

What is OAuth? Why it matters for your business

A plain-English guide to OAuth, the standard behind every 'sign in with Google' button, and the leaving-employee problem that quietly accumulates 400 unrevoked grants in a 30-staff firm.

A small operations team standing around a laptop in an office, one person pointing at the screen

What is a webhook? Why it matters for your business

A webhook is a vendor pushing an event to your URL the moment it happens. The plain-English version of why it beats polling, where you will meet it, and the four security gotchas to ask vendors about.

Two colleagues at a shared desk reviewing a laptop screen with printed notes alongside

What is MCP (Model Context Protocol)? Why it matters for your business

MCP is the open standard that lets AI tools connect to your business systems through one shared interface. A plain-English guide to where it earns its place and where waiting still makes sense.

Two people at a kitchen table reviewing a laptop screen with paperwork spread between them

What is an API key? Why it matters for your business

A plain-English guide to API keys, the long secret strings behind every vendor integration, and the four governance gaps that turn a leaked key into a five-figure invoice.

Two colleagues at a desk looking at a documentation page on a laptop, one pointing at the screen, the other taking notes

What is an API? Why it matters for your business

Every vendor in 2026 says they have an API. A plain-English guide to what that actually means, and the four questions an SME owner should ask before treating the claim as a procurement decision.

Two colleagues at an office table reviewing a printed contract, one pointing at a clause

What is vendor lock-in (in AI)? Why it matters for your business

AI vendor lock-in is quieter than cloud lock-in and harder to see coming. A plain-English guide to the five vectors that quietly raise your switching cost, and how to price them before you sign.

Two people at a meeting room table reviewing a printed proposal alongside a laptop showing a spreadsheet

What is total cost of ownership (TCO) for AI? Why it matters for your business

A plain-English guide to AI total cost of ownership, the eight categories that turn a £5k licence quote into a £30k year-one bill, and when to actually build a formal TCO model.

Two people at a kitchen table reviewing a printed invoice next to a laptop

What is hybrid AI pricing? Why it matters for your business

Hybrid AI pricing splits the bill into a predictable per-seat fee and a variable overage rate. The variable axis is where the surprises live. A plain-English guide for owners.

Two people at a desk comparing AI pricing sheets between a laptop and a printed page

What is per-seat vs usage-based AI pricing? Why it matters for your business

Per-seat pricing rewards distributed light usage, usage-based pricing rewards concentrated heavy usage. A plain-English guide to picking the model your firm actually needs.

Two people at a desk reviewing an API dashboard on a laptop with a printed error log beside them

What are rate limits? Why they matter for your AI bill

A plain-English guide to AI rate limits, the invisible ceiling that turns a quiet pilot into a Friday-afternoon outage, and the procurement questions that head it off.

Two people at a desk reviewing a printed vendor quote next to a laptop

What is inference cost (vs training cost)? Why it matters for your business

Training cost is the vendor's problem, not yours. Inference cost is the only AI cost that ever lands on your bill. A plain-English guide to what you actually pay for, where it shows up, and when it is worth managing.

Two people at an office desk reviewing printed pages and a laptop together

What is prompt caching? Why it matters for your business

Prompt caching is the largest single cost lever available to UK SMEs running production AI today. A plain-English guide to what gets cached, when it pays back, and the workload shapes where it does not.

A person at a desk reviewing printed contract pages next to an open laptop in a small office

What is a context window? Why it matters for your business

Context window is the model's working memory in tokens. Vendors advertise 1M, but effective context is closer to 60-70% of that. The plain-English version of why bigger is not always better.

Two people at a desk reviewing a billing dashboard on a laptop together

What are input vs output tokens? Why it matters for your business

Output tokens cost four to six times more than input tokens. A plain-English guide to what that means for your AI bill, and which lever actually moves it.

A person at a desk reviewing an invoice on a laptop screen with a printed spreadsheet beside it

What is a token? Why it matters for your AI bill

A plain-English guide to AI tokens, the billable unit hidden inside every vendor pricing page, and how a £20 trial becomes a £170 monthly bill without anyone noticing.

A person at a desk reading a printed document with an open laptop next to them

What is AI encryption (at rest, in transit, in use)? Why it matters for your business

Encryption at rest and in transit are now table stakes. For an AI service, the third state, encryption in use, is where the gap sits and where vendor claims quietly stop.

A person at a desk reading a printed document with a pen in hand and a laptop open beside them

What is synthetic data? Why it matters for your business

Synthetic data is artificially generated data that mimics real data. The vendor claim 'GDPR-clean because it's synthetic' is the line that does not survive a competent ICO query.

A person at a desk reading a printed document with a pen in hand, a laptop open beside them

What is human-in-the-loop oversight? A plain-English guide

The phrase is on every AI vendor's compliance pack and now in your insurer's renewal questionnaire. Here is what it means, what counts as real, and what the regulators look for behind the claim.

Two people across a kitchen table reading a printed letter together, one explaining, the other listening

What is bias in AI? Why it matters for your business

A plain-English guide to bias in AI for owners. Why "bias-free" is a marketing phrase, what to ask the vendor, and where to route it under UK law.

A person at a desk reading printed pages next to an open laptop, pen in hand

What is an AI audit trail? Why it matters for your business

An AI audit trail is the evidence that any decision your AI tool was involved in can be reconstructed and defended. Most owners think their vendor is logging it. What they are usually getting is metadata, not the conversation.

Two people in conversation across a desk, one pointing at a printed report, the other listening

What is explainable AI (XAI)? Why it matters for your business

A plain-English guide to explainable AI for owners running AI in regulated decisions, and the gap between a vendor's dashboard and what a regulator will actually accept.

A person at a kitchen table reading a printed questionnaire next to an open laptop and a coffee cup

What is model poisoning? Why it matters for your business

Model poisoning is the corruption of the data, weights, or documents an AI tool learns from. The plain-English version of the question that has started showing up in your supplier and insurance paperwork.

A person at a desk reading a contract document with a pen in hand, marking margin notes

What is data residency (for AI)? Why it matters for your business

A plain-English guide to data residency for AI. Why "EU residency" is rarely one promise, the five layers a vendor can split between jurisdictions, and the questions to send before you sign.

A person at a desk comparing two printed pages of text next to an open laptop in a small office

What is model drift? Why it matters for your business

Model drift is the slow degradation of an AI tool that worked six months ago and quietly stopped working as well. In 2026 there are two flavours owner-led businesses need to tell apart.

A person sitting at a kitchen table reading their inbox on a laptop, mug in front of them

What is prompt injection? Why it matters for your business

Prompt injection is the AI security flaw that cannot be patched, only contained. A plain-English guide to what it is, where you will meet it in 2026, and what to ask your vendors.

A person at a desk reading a laptop screen with a slight frown, phone face-up beside the keyboard

What is an AI hallucination? Why it matters for your business

AI hallucinations are confident, fluent answers that are simply not true. The practical question for an owner is not how to eliminate them, but where in the business they are acceptable and where they are not.

A person in an office pulling a paper folder from a shelf next to an open laptop

What is RAG (Retrieval-Augmented Generation)? Why it matters for your business

RAG is the architecture that lets an AI tool answer questions using your documents instead of inventing answers from training data. The plain-English version of why every vendor is using the word.

A person at a desk reviewing hardware specifications on a laptop screen with a printed sheet beside them

What is model quantization? Why it matters for your business

A plain-English guide to model quantization. The compression technique that decides whether you can run an AI model on your own hardware, and the question your IT vendor is hoping you do not ask.

A person at a desk writing on a notepad with a laptop open beside them

What is prompt engineering? Why it matters for your business

Prompt engineering is not a job title or a magic trick. It is the practical discipline of writing instructions to an AI tool so it produces what you actually need, reliably and at predictable cost.

A person holding a phone showing a photo of a paper document, with a laptop open in front of them

What is multimodal AI? Why it matters for your business

A plain-English guide to multimodal AI. What it actually means when a vendor says it, where the genuine value is, and where the marketing exceeds the engineering.

Two people at a kitchen table reviewing printed pages and a laptop screen together

What is fine-tuning (an AI model)? Why it matters for your business

Fine-tuning is what vendors mean when they say 'we'll customise the model on your data'. A plain-English guide to what it is, when it pays back, and when prompt engineering or RAG would be the cheaper answer.

Two people side by side at a screen, one pointing at something the other is checking

What is an AI agent (and agentic AI)? Why it matters for your business

AI agents are the most-pitched, most-oversold AI capability of 2026. A plain-English guide to what an agent actually is, where the technology genuinely fits, and the governance questions to settle before you switch one on.

A person at a desk reading a printed proposal with a laptop and notebook in front of them

What is a foundation model? Why it matters for your business

A plain-English guide to foundation models. The category that sits underneath every AI product you will be pitched in 2026, and the question worth asking before you commit to one.

A person at a desk with PDFs visible, working on a laptop with a chat-style interface open

What is a custom GPT? Why it matters for your business

A plain-English guide to custom GPTs and the equivalents from Claude and Google. What they are, what they are not, and what to ask before paying a vendor £500 a month to build you one.

A person at a desk reading a suggestion in the side of an open document on screen

What is a copilot (in AI)? Why it matters for your business

A plain-English guide to AI copilots for business owners. What sits behind the word, why it stopped meaning anything specific, and what to ask back.

A person at a desk reviewing a printed checklist with a laptop and notebook in front of them

Should you be using AI for this at all? A qualification framework

A plain-English qualification framework for UK SME owners. Six gates that decide whether a specific task is a real AI candidate, or one that should stay manual.

A person at a desk reviewing a pricing comparison on a laptop with a printed spreadsheet beside them

SaaS AI vs self-hosted: which one your business actually needs

A plain-English progression guide for UK SME owners. Start with SaaS AI, migrate to self-hosted only when one of three thresholds is crossed.

A person at a desk with a stack of paper documents on one side and an open laptop on the other

Prompt engineering vs fine-tuning: which one your business actually needs

A plain-English progression guide for UK SME owners. Start with prompt engineering and retrieval. Move to fine-tuning only when specific thresholds are crossed.

A person at a desk reviewing a recommendation on a laptop with a pen in hand

Human-in-the-loop vs full automation: which one your business actually needs

A plain-English risk-routing guide for UK SME owners. The right oversight model depends on the per-task risk profile, not a single firm-wide policy.

A person at a desk reviewing two software interfaces on a laptop with a notebook beside them

General-purpose AI vs vertical specialist AI: which one your business actually needs

A plain-English decision guide for UK SME owners. The general-purpose AI platforms and the sector-specialist tools both have a real case. The choice depends on workflow density and compliance load.

Two people in conversation across a table, one explaining something on a laptop screen

What is an LLM (Large Language Model)? Why it matters for your business

A plain-English guide to LLMs for business owners, what they are, why every vendor mentions them, and the questions to ask back.

html

If any of this sounds familiar, let's talk.

The next step is a conversation. No pitch, no pressure. Just an honest discussion about where you are and whether I can help.

Book a conversation