Thinking on AI, business
and building what matters.
Practical perspectives from the Ferrous Labs team — for business leaders navigating AI without the patience for enterprise timelines.
Train Your Agent Like a Neural Network (Then Watch It Overfit)
Treating an agent’s configuration as weights took it from 33% to 92% on unseen tasks. Then the shop’s rules changed and it fell to 29%. What the optimiser had actually learned, and how to regularise it.
Read articleIs Your AI Getting Worse? Your Customers Will Know First
When an AI agent gets worse, nothing goes red. Ali Nadian on the three things drifting under every agent, four ways to score free text, and what teams who ship with confidence do differently.
Read articleAI Strategy and Roadmap Consultancy UK: Specialists
Most AI roadmaps are written for process-heavy businesses. If your revenue depends on the judgement of a small number of senior specialists, here is how to tell whether the strategy you have been sold actually fits.
Read articleRAG Evaluation Framework: Metrics, Test Cases and Release Gates
An answer can sound right while citing the wrong source. How to measure retrieval and generation separately, build test cases from real work, and turn the result into a gate you re-run before each release.
Read article8 Real AI Hallucination Examples and What Caused Each
Court filings, a consultancy report, a search engine, an airline chatbot, a launch demo, a reading list, a support agent and a city’s business tool — each traced to the gap that let it through, and what the eight have in common.
Read articleHow to Tell If an AI Is Hallucinating: Signs, Checks and Tools
Six signs a model is making things up, eight checks a team can run on a live feature, what hallucination checkers actually detect — and why catching one instance is not the same as knowing your rate.
Read articleLLM Hallucination Rate: What It Measures and How to Get Yours
Published rates measure a model on a benchmark; your feature has a different number. What the leaderboards can and cannot tell you, and the seven steps to measuring your own — with its unit and denominator attached.
Read articleWhat Is an AI Hallucination? Meaning, Causes and Examples
The reference page: what the term means, why models do it, the documented cases, whether it still happens in 2026, and how to tell if yours is — before the question becomes how often.
Read articleAI Hallucinations in Production: Why They Happen and How to Cut the Rate
Why a language model invents facts, the four places hallucinations actually enter a shipped feature, and what reduces the rate in practice — in order of leverage.
Read articleAI Consultancy for Mid-Market Workflow Readiness
Before you hire an AI consultancy or commission a SaaS build, there is a prior question worth answering honestly: is your workflow actually repeatable enough to productise? Here is how to find out.
Read articleAI Workflow Automation for Technical Services UK
Generic AI workflow tools weren't built for accreditation standards, regulatory constraints, or the judgement-led methods that make expert services valuable. Here is what actually works for regulated technical businesses in the UK.
Read articleValidate a SaaS Idea from an Existing Service
The safest way to de-risk a SaaS build is to run a pricing experiment on your current clients before you write any code. Their response tells you more than any discovery workshop.
Read articleAI Integration Engineering Consultancies Workflow Guide
Most engineering consultancies stall on AI not because they chose the wrong tool, but because they never designed the workflow that connects it to actual client deliverables and team sign-off chains.
Read articleAI for Surveyors UK Recurring Revenue: Workflow First
Most surveying businesses thinking about AI jump straight to 'we should build a product'. Here is why embedding AI into your existing workflows first is smarter, and how it creates recurring revenue without the upfront risk.
Read articleTest Consulting Methodology SaaS Before Building It
Most consultancy founders assume they need a technical co-founder or deep pockets to go SaaS. A concierge MVP lets you prove demand with neither, before a single line of product code is written.
Read articleLegacy Software Is the Ceiling on Your AI Productivity
UK productivity is genuinely rising, and not because of AI. That makes right now the moment to find out whether your systems can keep up with the workers carrying them.
Read articleAI Professional Services Software: The Security Gap
Off-the-shelf AI tools feel like a shortcut, but for professional services businesses handling sensitive client data, the security vulnerabilities baked into third-party LLM platforms may be costing you more than you realise.
Read articleCustom AI Tool Build Workflow Design: Why Builds Fail
Most custom AI tool builds for specialist businesses fail not because the technology is wrong, but because workflow design ignores how senior experts actually make decisions. Here is how to catch that problem before it costs you.
Read articleBuild a SaaS Product Without a Technical Co-Founder
You do not need a technical co-founder to build a SaaS product. You need proof of demand first, and the right build partner second. Here is the structured path specialist businesses actually use.
Read articleAI Customer Experience: What Service Businesses Get Wrong
Chatbots answer the questions clients do not have. Where service-business clients actually get frustrated, and the unglamorous fixes that work.
Read articleHow to Add AI to Existing Software Without a Rebuild
You almost never need to rebuild. The three ways to add intelligence to software you already run, and how to pick the one that fits your architecture.
Read articleHow to Extract Data From Technical Reports With AI
Technical reports are not invoices. Which fields extract reliably, which need a human, and how to test the difference on your own documents in a day.
Read articleCustom Computer Vision UK: What a Build Really Involves
Inspection, defect detection and measurement from images are solved problems, until they meet your equipment. What a real industrial vision build involves.
Read articleWhat Is Bespoke AI Software and When Is It Worth It?
A plain definition, the four things that make software genuinely bespoke, and an honest test for whether your problem justifies building rather than buying.
Read articleAI for Surveyors UK: Where the Real Time Savings Are
Site notes, measurements and report assembly consume more hours than the inspection itself. Where AI helps a surveying practice, and where it must not.
Read articleHow to Turn Your Consultancy Into a Software Product
The method you repeat for every client is the product. The five stages from repeatable service to sellable software, and the stage most businesses skip.
Read articleAI Consultancy for Mid-Market Businesses: What Differs
Mid-market businesses get pitched enterprise process at enterprise prices. The four constraints that genuinely differ, and what they should change about the engagement.
Read articleIntelligent Document Processing UK: Build It or Buy It?
Off-the-shelf extraction has become genuinely good. The document types where it still fails, and the honest test for whether you need to build.
Read articleSignal Processing Machine Learning: Which Approach Wins
Can machine learning replace signal processing? Usually not, and the systems that work best treat them as partners rather than rivals.
Read articlePredictive Maintenance AI: Are You Ready to Build It?
The technology is the easy part. A readiness check on failure history, sensor coverage and whether your maintenance process could act on a warning anyway.
Read articleIndustrial AI: Start With the Data You Already Have
Most industrial AI proposals start with a use case. The ones that work start with an audit of the data already sitting in the historian.
Read articleCustom AI Software Development: How the Process Works
AI projects fail in different places from normal software projects. The five stages of a custom build, and the checkpoint that should end each one.
Read articleBespoke AI Solutions for Business: What Counts as Bespoke
Everything is sold as bespoke. The four tiers of customisation, what each actually costs, and how to work out which one your problem genuinely needs.
Read articleAI Workflow Automation UK: How to Run a First Pilot
A pilot is not a trial licence. How to choose the process, define the number that decides success, and get an answer inside six weeks.
Read articleAI for Engineering Firms UK: A Practical Adoption Path
Most engineering businesses skip straight from experiments to a stalled platform project. The staged path that gets from curiosity to a system people actually use.
Read articleAI for Environmental Consultants: Where It Actually Helps
Site data, monitoring records and report assembly are where the hours go. Where AI genuinely helps an environmental consultancy, and where it should stay away.
Read articleAI Demand Forecasting: How Accurate Is It in Practice?
The accuracy question deserves a straight answer: it depends on your data, and sometimes the simple model wins. What to check before you invest in forecasting AI.
Read articleAI Strategy Consulting: What You Should Actually Get
Too many AI strategies are a deck and an invoice. What a strategy engagement should actually hand over, and the difference between a plan and a proof.
Read articleAI Consultancy UK: How to Choose the Right Partner
Strategy houses, agencies and studios all answer to the same search term. How to tell which kind of AI partner your business actually needs, and the questions that expose the difference.
Read articleAI Agent Development UK: What It Involves and What It Costs
Everyone sells agents; few define them. What separates an agent from a scripted workflow, when you genuinely need one, and what a production build involves.
Read articleWhat Is an AI Transformation Studio and Do You Need One?
The term is new, the model is not: a small engineering team that takes AI from idea to running software. What a studio actually does, and when it is the wrong choice.
Read articleWhy Do AI Pilots Fail? The 6 Reasons Projects Die Before Production
Around 95% of enterprise AI pilots deliver no measurable return. The six failure modes behind that number — and what the successful 5% do differently.
Read articleBest AI Tools for UK Engineering Consultancies in 2026
Report drafting, bid writing, site notes and standards search — the tools worth buying, and the data-to-report layer where bespoke wins.
Read articleBest AI Workflow Automation Tools for UK Businesses (2026 Guide)
Zapier, Make, n8n and Power Automate compared by use case, data residency and judgment-handling — and the line where bespoke beats them all.
Read articleAI Professional Services Software: Real or Illusion?
An AI agent that appears to solve your problem perfectly might be doing exactly that, or it might be gaming its own success metrics. Knowing the difference before you commission custom software could save you a significant amount of money.
Read articleWorkflow Automation ROI: Which Processes to Fix First
Most specialist businesses automate what's easy, not what matters. Here's how to map your workflows by revenue impact before you spend a penny on tooling.
Read articleCustom AI Software UK: Workflow Design Comes First
Custom AI software projects in the UK fail most often not because the model is wrong, but because the integration was designed against a workflow that exists only on paper. Here is why workflow design must come before the build.
Read articleUK AI Compliance Architecture, Built In From Day One
Most specialist businesses treat UK regulatory compliance as a late-stage problem. Builders who wire it into the architecture from day one end up with better software, fewer delays, and clients who actually trust the product.
Read articleWhich Parts of Consulting Delivery Are Actually Repeatable?
A repeatability map separates the steps that run the same way every engagement from the judgement clients pay for. Draw it before you scope any software.
Read articleWhy Generic AI Platforms Fail Specialist Businesses
Generic AI platforms are optimised for the average business — a fiction no specialist actually runs. Here is why bespoke AI software built around your methodology outperforms them, and what to validate before you commission it.
Read articleAI Readiness Assessment for Specialist Businesses UK | What to Test First
Specialist businesses often stall on AI because they treat readiness as a technical question. The more important test is whether your team can adopt and govern AI before you spend anything.
Read articleBespoke AI Software Cost UK: What the Estimate Misses
The build quote looks manageable. The ongoing costs do not. Before you commission bespoke AI software, here is what the proposal almost certainly leaves out.
Read articleHow To Price Bespoke AI Software
Building bespoke AI software on top of a consulting model sounds logical until you run the numbers. Here is the pricing test most UK practices skip entirely.
Read articleBespoke AI Software Needs a Price Test Before Code
Specialist consultancies spend months building bespoke AI software only to find clients won't pay recurring fees. Here's how to test pricing and buyer willingness before writing a line of code.
Read articleProductise Your Consultancy: The Hidden Cost
When consultants budget a move into SaaS, they count the development quote and miss the real cost. Here is how to calculate what you are actually committing to.
Read articleAutomation Threshold: Scale a Consultancy Without Hiring
The concierge model works, but the unit economics don't. The question isn't whether to productise your methodology, but when.
Read articleSaaS Unit Economics for Consultancies
The median software gross margin is 80%. Whether your methodology can reach it depends on how much expert time is left in every engagement.
Read articleWhy Consultancy SaaS Builds Fail and How to Fund Them
Expert-led consultancies keep failing at SaaS for the same reason: they treat a product build like a fixed-price project. There is a better model, and it funds itself.
Read articleProductise Consulting Services with AI Agents
Most specialist consultancies have a repeatable diagnostic shape hidden inside their bespoke work. AI agents can help you productise it, but only if you put the boundary in the right place.
Read articleYou May Already Be Running a Concierge MVP
Your consultancy is probably already running one and calling it the service. The slow part is what makes the product hard to copy.
Read articleEnd-to-End AI Delivery: One Partner for Your AI Lifecycle
Discover how a single trusted partner can guide your organisation from strategy through to live AI systems, with a lean, evidence-driven approach.
Read articleWhat Is the Best AI Opportunity for Your Business Right Now
A practical framework for ranking AI opportunities and picking the right first project for your business, in under a week.
Read articleSeven Business Outcomes Any UK Business Can Predict With AI
Seven practical outcomes any UK business can predict with AI from their historical data, plus the data, tools, and conditions you need to start.
Read articleHow Third-Party Data Is Reshaping UK Business Decision-Making
Most UK businesses ignore the public and third-party data that could sharpen their commercial decisions. A practical playbook for UK leaders.
Read articleWhy UK AI Strategies Fail — and How to Fix Yours
Most UK AI strategies collapse before delivering value. Discover the real failure modes and a method-driven approach that actually works for your business.
Read article7 Types of Bespoke AI Tools for UK Professional Services
Discover 7 types of custom AI software UK businesses are building right now to win better clients, cut costs, and grow without adding headcount.
Read articleIs Your Business Ready for AI? A Lean Approach to Finding Out
Find where you sit on the AI adoption curve and where AI would create the most value. The Lean AI Strategy Diagnostic gives you both answers in five minutes.
Read articleHow Successful UK Businesses Plan Their AI Implementation Strategy
Discover how UK businesses are building effective AI implementation strategies. Learn the proven 4-stage framework that transforms AI adoption from risky to reliable.
Read articleProof of Demand Validation for Expert Businesses
Your track record has already proved the demand. The one question it cannot answer is whether the value survives without you in the room.
Read articleHow Much Does Custom AI Development Cost in the UK?
Custom AI development runs £5k for scoping to £300k+ for production. Why the range is so wide, and why scope is the question rather than cost.
Read articleThe Expertise Paradox: Build the Software That "Replaces" You
Expertise that lives only in people's heads is an asset that depreciates. It leaves at six, it retires, and it cannot be sold.
Read articleWhy 90% of SaaS Ideas Fail - And Why That Number Is Wrong
The 90% figure counts venture-backed startups missing a 10x return, not businesses failing. The ONS puts UK five-year survival at 38.4%.
Read articleFive Signs Your Industry Needs Better Software
Generic software fails specialist businesses for a structural reason: the training data does not exist. Five signs the gap in your industry is real.
Read articleRecurring Revenue for Consultants
UK consulting growth went 23% to 11% to a forecast 3.6%. Recurring revenue changes what you can plan, and a retainer is not the same thing.
Read articleHow to Price Professional Software
Your customer is already paying for the problem your software solves, in hours, rework and errors. That cost is the number you are looking for.
Read articleWhy Most SaaS Advice Doesn't Apply
The SaaS playbook was written when engineering was the hard part. It no longer is, and market access, data and expertise are now the scarce advantages.
Read articleFind out where your business
sits on the AI adoption curve.
A short conversation with Catalyst, our AI guide — or the typed finder. A personalised report. No sales call required unless you want one.