AI Use Cases and Applications Across Major Industries

AI is ‘everywhere’ according to every headline you read. The problem with that phrase is that it obscures the reality of how unevenly it actually is deployed. In a number of sectors it is genuinely transforming the way work is done. In a whole host of others it’s just quietly useful at the fringes. In a few it’s nothing more than a slide in a pitch deck. To simply say that ‘AI is revolutionising everything’ actually makes it harder to see where it’s having any real impact.

Here’s the real insight that cuts through the noise and it’s the very essence of this guide. Get rid of all the industry jargon and what you’re left with is AI doing the same six things in every case. It predicts. It personalises. It automates. It sees. It understands language. It optimises. A fraud model in a bank and a tumour detector in a hospital are fundamentally the same type of machine, just applied to different problems. Once you see that pattern then ‘AI for X’ ceases to be a mystery and becomes utterly obvious.

So this is a practical map rather than a hype tour. We’ll lay out the patterns that keep repeating, go through how they manifest in the various industries with real world use cases and be absolutely clear about where AI is delivering and where it’s just been overhyped. We build AI in most of these sectors so this is the view from the workshop rather than the trend report.

The patterns that repeat: what AI actually does

Before the industry tour, learn the six moves. At its core, as IBM frames it, AI is about getting machines to perform tasks that normally take human intelligence — but in practice today that resolves into a short, concrete list of things it’s genuinely good at. Nearly every real-world AI use case, in any sector, is one of these six applied to a specific domain. Hold them in mind and the rest of this guide reads like variations on a theme.

• Prediction and forecasting. Learning from historical data to anticipate what comes next — demand, risk, equipment failure, customer churn. The quiet workhorse of business AI, built on machine learning.
• Personalization and recommendations. Tailoring what each person sees or is offered — the product, the content, the next best action. A recommendation system is the engine behind this, and it shows up anywhere there’s a customer to serve.
• Automation of repetitive work. Handing off the high-volume, rule-ish tasks — sorting, extracting, routing, processing — so people spend their time on the parts that need judgment. It’s the same move whether it’s a bank processing loan paperwork or a hospital coding medical records.
• Computer vision. Reading and interpreting images and video: spotting a defect, reading a document, recognizing a face, understanding a scene. The same capability spots a tumor in a scan, a crack in a turbine blade, and a weed in a field.
• Language understanding and generation. Working with human language — answering questions, summarizing, translating, drafting. The capability behind every chatbot and voice assistant, and behind every tool that reads a contract, summarizes a report, or drafts a reply.
• Optimization. Finding the best option among an impossible number of choices — routing a fleet, scheduling a factory, allocating a budget, pricing in real time. It’s the quiet math that moves a delivery network, a production line, and a power grid.

That’s the whole toolkit, more or less. Every use case below is one of these six, pointed at an industry’s particular problem. If you remember nothing else, remember the six — they’re the fastest way to tell a real AI opportunity from a buzzword.

It’s also worth noticing how often the real systems combine several of them. A modern customer-service setup, for instance, uses language understanding to grasp the question, prediction to anticipate what the customer needs, and automation to resolve it — three patterns in one workflow. The most valuable applications tend to stack the moves rather than rely on a single one. That’s a useful tell, too: a genuinely transformative use case usually braids two or three of the six together, while a thin “we added AI” feature often turns out to be none of them wearing the label.

AI use cases across major industries

Now the tour. For each industry, the point isn’t an exhaustive list — it’s to show which of the six patterns are doing the heavy lifting, and where the value genuinely sits.

Healthcare

One of the most profound and high value applications. Computer vision can look at medical images and flag up tumours and abnormalities that a fatigued human eye might well have missed. Predictive models can identify patients at risk of decompensation long before their vitals actually crash. Language models can be used to structure the sheer volume of clinical notes which is such a massive drain on the time of a clinician. AI can also be used to accelerate drug discovery by taking the millions of molecules that are in the pipeline and whittling them down to a few that are actually viable. A significant amount of the near term, lower risk value is in the fairly unglamorous operational layer (scheduling, billing, prior auth etc) which is the very thing that drives so much of the clinician burnout. That’s the area where a mistake is just an inconvenience rather than a disaster. It’s worth noting the pattern in the industry here: the high profile diagnostic AI is real, but it’s very slow to roll out due to the regulatory environment. The back office automation is the one that is delivering value today in a quiet way, saving money and reducing stress. The pattern is very much one of assistance – the AI flags and suggests, the clinician makes the decision. That’s the right model and that’s the way our healthcare work is built.

Finance and banking

Finance has always been a data driven business (long before ‘AI’ was a thing) so it’s a very natural fit. The prediction models are used to score credit risk and to catch fraud in real time by identifying transactions that are out of the norm for a particular customer. The optimisation models are used for algorithmic trading and portfolio optimisation. The language models are used to answer customer queries and to parse filings and news for signals. Fraud detection alone is the primary driver of most banks’ spend on this stuff as the cost of a single fraud ring that goes undetected is massively greater than the cost of the model that catches it. Robo-advisors and automated compliance monitoring are further examples of the value that can be delivered in the background. AI agents in finance are increasingly being used to bring all of this together, albeit with the caveat that the highly regulated and high trust environment of finance means that there will always be a human in the loop on anything that is actually consequential.

Retail and e-commerce

Retail is the home turf of personalisation. Recommendation engines are a massive driver of what people discover and buy online. Predictive models are used to forecast demand and prevent stock either going stale or running out. Language powered assistants are used to guide customers and deal with support. Vision is also making inroads (visual search, cashierless stores etc). The common thread is the same one that’s always been the goal of retail: get the right product to the right person at the right time. On the back end of the shop too, AI is used to fight fraud and chargebacks, to price dynamically and to balance stock across stores and warehouses (the unglamorous bits of the business that are essential to protecting the margin).

Manufacturing

Manufacturing’s prime use case is predictive maintenance. Sensors on the machines feed into the models which will then flag a part that’s about to fail before it actually stops the line, turning what would otherwise be a very expensive unplanned stop into a planned fix. Computer vision is used for product inspection and it’s far more efficient and consistent than human quality assurance, able to spot that hairline crack that a person will just glance past on the thousandth unit of the shift. Optimization is used to work out the best production schedule and the best level of parts to have in the system. AI guided robotics are used for the physical work. In an industry where a line stop is costing you a fortune by the hour the returns are as tangible and measurable as you’re going to get. That’s why it was such an early and committed adopter.

Supply chain and logistics

Very few areas are more suited to AI really, it’s an end to end optimisation and prediction exercise. You’ve got models forecasting demand across the thousands of SKUs, planning the most efficient routes and then dynamically rerouting as circumstances change, optimising warehouse layouts and picking, and flagging any disruption before it snowballs. The pandemic has all of us in the supply chain industry acutely aware of just how fragile our supply chains can be and AI is a major part of the solution in terms of making them more resilient (running scenario simulations, rapidly finding alternative suppliers, holding just the right amount of stock rather than too much or too little). Our work on AI agents for the supply chain is very much in this space.

Marketing and sales

Marketing is all about personalisation and language. It’s about segmentation, generating and testing content at scale, predicting which leads are worth pursuing and then personalising the message to the individual. Sales teams use AI to score and prioritise their prospects, to draft their outreach and to work out the right moment to follow up. The danger here is that you end up with volume rather than value as AI makes it so easy to churn out generic content that no one wants. The teams that win use it to refine their targeting and to free up the sales team to do the human bits of selling rather than to just flood inboxes. When used well it can score and route leads so that the sales people are spending their time on the prospects that are most likely to close and can surface the next best action from a CRM that nobody has time to actually read.

Customer service

The most obvious AI that people encounter. Language models power the chatbots and assistants that instantly answer the trivial queries, route the more complex ones to the appropriate human and read the sentiment of the message to ensure an angry customer is prioritised. Done well it means faster answers and the human agents are freed up to deal with the problems that actually need a human. Done badly it’s just a frustrating loop that will never let you talk to a human and that’s a design failure, not a technology failure. The ideal setup is one where you have a chatbot that’s easy to use and a simple path to a human, with AI working in the background to support that human by pulling up the full history, suggesting answers and summarising the case so the customer never has to re-explain themselves.

Media and entertainment

Media is all about personalisation and increasingly about generation. The recommendation engines that determine what you watch, what you read and what you listen to are central to the business. AI is used in production (editing, VFX, dubbing, localisation) and is fundamentally changing the way content is made, with all the attendant creative and ethical issues. The personalisation that keeps the audience engaged is the commercial engine. The generation side is the fast moving, contentious frontier. AI is also the engine behind the ad targeting that funds so much of the business and the content moderation that allows platforms to function at scale (something no human team could possibly do).

Insurance

Insurance is fundamentally a prediction and automation business. AI accelerates the claims process by being able to read and extract data from documents and photos, it’s used in underwriting to assess risk and it can identify the patterns that indicate fraud. Computer vision can be used to assess the extent of vehicle or property damage from images, turning a days long claim into a matter of minutes. As in finance it’s a regulated industry where the AI is assisting in the decision and not making the final decision unsupervised. Telematics and usage data allow insurers to price on actual behaviour rather than on broad demographic groups and AI chat is used to take the first notice of loss so the claim is in motion as soon as it’s reported.

Legal

Legal work is drowning in documents, which is exactly what language AI is for. It accelerates discovery by finding the relevant files in a mountain of them, reviews contracts and flags risky clauses, and speeds legal research. It won’t replace a lawyer’s judgment, and the well-publicized cases of AI inventing fake citations are a standing reminder to verify its output. But as an assistant that compresses weeks of document review into days, it’s genuinely useful. Smaller firms especially gain leverage they couldn’t otherwise afford — research and review capacity that used to require a roomful of associates.

Education

The promise of education is personalisation on a scale that teachers simply couldn’t achieve on their own. You can have lessons and quizzes that adapt to the individual student, immediate feedback and tutoring assistants available at any time. LLMs can write the content and answer the mundane questions, freeing the teacher to do the actual teaching. The caveat of course is that education is a very relational thing and the aim is to support the teacher and the learner, not to remove the relationship altogether. There is also a very real amount of admin that can be taken off the teacher’s plate (grading, scheduling, the myriad of routine questions) which is a massive amount of time that is freed up.

Transportation and mobility

Transportation is a prediction and optimisation story end to end. Route optimisation finds the best route for a delivery fleet and then re-routes it in real time as traffic changes, saving both fuel and time on a scale that no human dispatcher could possibly achieve. Predictive maintenance looks at the sensors on the vehicles, the trains and the aircraft to spot a failing component before it actually fails and leaves a truck stranded or a plane grounded. Computer vision is at the heart of driver assistance and of the long and gradual journey to autonomy (it’s reading the road, spotting hazards, holding a lane). Transit agencies use demand forecasting to match their service to the actual way people travel. The real take away from the whole self driving thing is that the assistive, optimising applications are already delivering real value today, whereas full autonomy remains confined to very limited mapped, geofenced areas in a few cities. The real money is in the unglamorous bit.

Energy and utilities

Energy is fundamentally an optimisation problem: balancing supply and demand in real time. That’s an area where AI is ideally suited. It’s used to forecast demand across the grid and balance the intermittency of renewables against the more predictable supply of other sources to smooth out the inevitable fluctuations that make clean energy so difficult to manage. Predictive maintenance is another area where it’s being used to great effect, allowing problems with turbines, transformers and pipelines to be identified before they lead to outages. In oil and gas it’s used to optimise drilling and production and to model reservoirs from seismic data. Vision and sensor analysis are used to monitor infrastructure over thousands of miles of pipeline, which would be utterly impossible to do by any human inspection team. As the grid becomes more complex and more renewable the optimisation problem becomes harder and harder to solve and hence more valuable to solve well. That’s why utilities have moved from pilots to actual deployment much faster than their reputation for caution would have you believe.

Agriculture and farming

Agriculture is actually one of the most promising areas for AI – the margins are so narrow and the variables (weather, soil, pests, water) are precisely the sort of thing that prediction and vision are so good at. Precision farming uses computer vision from drones and tractors to identify disease, weeds and stressed plants on an individual plant basis and then applies water, fertiliser and pesticide only where it’s required rather than spraying the whole field. Prediction models can forecast yields and the optimum times to plant and harvest. For smallholders in the developing world, mobile based advisory tools are bringing the capability of industrial farms into the ordinary phone. The benefit is not just in the cost – it’s in using less water and less chemical input to achieve the same or better yield – it’s one of the very few areas where the efficient and the sustainable are one and the same.

Real estate

Real estate is using AI in the areas of valuation, matching and the paperwork in between. Prediction models can assess a property’s value from its location, features and market data far more quickly than a manual valuation and can flag up any prices that seem out of line. Matching engines are used to match buyers and renters with the properties that are actually suitable for them, eliminating the endless scroll. Language models are handling the initial wave of queries and the bulk of the document work (listings, leases, disclosures etc) and computer vision can take a walk-through video and turn it into a floor plan or automatically tag the features of a listing. It doesn’t in any way replace the local knowledge of an agent or their ability to negotiate, but it does compress the inherently slow and administrative nature of the business.

Human resources and recruiting

HR were early adopters of AI for screening and scheduling and it’s in that area that some of the most pertinent cautionary tales are to be found. The AI is used to screen and rank CVs, to schedule interviews and to answer the myriad of routine queries that fill an HR inbox. It’s also used in the form of predictive models to flag potential flight risks before an employee actually resigns. The downside of course is bias – a model trained on a company’s historical decisions will learn and reproduce whatever bias was present in that history and there have been numerous well-publicised examples of this as well as actual legal exposure. HR is the most obvious example of the rule that applies in all areas – AI inherits the bias of its data – and in such a high stakes environment involving human decisions it absolutely demands careful oversight rather than un-supervised automation. It can and should be used to assist and accelerate, not to make the final decision without any human intervention.

Hospitality and travel

Hospitality and travel are fundamentally about personalisation and language. Recommendation engines are all over the place, suggesting hotels, restaurants and experiences based on a traveller’s profile and history. Chatbots and assistants are used to take the call, deal with the booking, handle check-in, deal with itinerary changes and answer the endless stream of “what time is checkout?” queries 24/7 in all the languages of the world. Predictive analytics are used to set dynamic pricing for rooms and flights and to forecast demand so that operators can staff and stock appropriately for the week ahead. Real time translation is of course a huge help in removing the friction of travel. It’s a service business at its heart so the winning proposition is to free up your staff from the repetitive tasks so that they can deliver the real human hospitality that no model can ever replicate (that’s the bit that guests actually remember).

Construction

Construction is a late but rapidly growing adopter as their problems (safety, delays, cost overruns) are all very expensive and increasingly data rich. Computer vision is being used to monitor sites for safety breaches and to track progress against plan via photos and drone footage. Prediction models are used to forecast delays and budget overruns before they become catastrophic, learning from the patterns of previous projects. Optimisation is used to sequence the thousands of interdependent tasks of a build and the crews and plant that do them. Generative design is used to explore structural options that a human may not have thought to consider. The constraint of course is that construction data has historically been very messy and project specific so the ones who are seeing the value are the ones who have been serious about capturing clean data from the outset.

Telecommunications

Telecoms have massive, complex networks which are the perfect environment for prediction and optimisation. AI is used to forecast network load and adjust capacity in real time, predict equipment failure before it causes a service outage and detect the anomalies that are the early warning signs of outages or fraud. On the customer side it’s used to power the chatbots and the churn prediction models that operators use to retain their customers in such a cut throat market. As networks become denser and more software defined the scale of the optimisation problem becomes such that automation isn’t just a nice to have, it’s absolutely essential. There’s simply too much going on too quickly for humans to be able to tune it all by hand.

Government and public sector

The government have been very cautious and that’s entirely appropriate given the stakes and the level of scrutiny. The use cases are very real though: the sheer volume of documents and applications that public sector bodies are forced to process, the answering of citizen queries via AI assistants, the detection of fraud in benefits and tax and the optimisation of everything from traffic lights to the allocation of resources. The caution is absolutely justified as in the public sector you’re dealing with people who have no choice but to interact with the system. The bar for fairness, transparency and accountability is therefore orders of magnitude higher than it is in a commercial context. That’s the world where “human in the loop” and “explainable” go from being nice to have to being absolutely essential.

Gaming

Gaming has been using AI for longer than almost anyone else (if you count NPC behaviour) and the modern applications go far beyond that. ML is used to tune the difficulty to keep you in your flow, to personalise the content and the recommendations and to power the anti-cheat systems that keep the competitive games fair. Generative AI is now being used to create assets, environments and dialogue and is set to deliver on the promise of much more dynamic and responsive worlds. Gaming is also a proving ground in the sense that the techniques you develop there (e.g. reinforcement learning agents mastering environments of extraordinary complexity) are very likely to go on to find application in other industrial settings later.

The list could keep going further still — but the lesson is already clear. The industries differ; the six moves don’t. Wherever there’s a costly problem and usable data, at least one of them tends to fit.

The generative AI layer across every industry

There’s a newer shift worth pulling out on its own, because it cuts across every sector at once. The generative AI wave — the large language models and image, audio, and video generators that broke into the mainstream — added a genuinely new capability on top of the six patterns: machines that don’t just analyze and predict, but produce. Text, code, images, designs, synthetic data, drafts of almost anything. It’s the fastest-moving part of the whole field, and it’s reshaping work in every industry we’ve covered — often arriving in a sector through a side door, as employees quietly start using general-purpose tools before the organization has any formal strategy for them.

The pattern of impact is very much the same. In marketing it’s about drafting and testing content. In software it’s about writing and reviewing code. In media it’s about assisting in production. In law and finance it’s about summarising and drafting documents. In customer service it’s about powering the actual chatbots. In healthcare and research it’s about helping to synthesise the literature and structure the notes. The common thread is that generative AI is absolutely superb at producing a first draft of knowledge work very quickly. That’s precisely how the careful teams are using it: as a tireless junior that will draft, suggest and accelerate. A human is still in complete control and ownership of the final output.

The caveats of honesty go with it everywhere. Generative models produce the plausible rather than the true (they can just as easily state something false as something true, that’s the phenomenon of hallucination). They are the product of the biases in their training data and they raise all sorts of issues around IP, authenticity and what should be disclosed. The same rule applies in every industry: generative AI is a massive enabler for the people who actually have their hands on it and a real liability for anyone who takes its confident output as the finished article. It expands what AI can do, it doesn’t remove the need for judgement.

The benefits that cut across industries

Different sectors, but the payoffs rhyme. Whatever the industry, the value of AI tends to land in the same handful of places.

• Efficiency and cost. Automating high-volume work and optimizing operations cuts cost and frees people for higher-value tasks. The most common reason AI projects get funded.
• Better decisions from data. Prediction turns the data a business already has into foresight — what will sell, what will break, who will churn — so decisions rest on more than instinct.
• Scale and availability. AI handles volume no team could, around the clock, without tiring — from support queries to document review to monitoring.
• Personalization. Treating each customer as an individual rather than a segment, at a scale that was never possible by hand.
• New capabilities. Some things simply weren’t possible before — real-time translation, instant image analysis, drug candidates found in silico. Not efficiency gains, but genuinely new abilities.
• Freeing people for higher-value work. When AI absorbs the repetitive load, people get to spend their time on judgment, creativity, and relationships — the parts of the job machines are worst at and humans are best at. Handled well, that’s the most durable benefit of all.

One caveat belongs with all of these: the payoffs are potential, not automatic. The same project can deliver every one of them or none, depending on execution — which is why two companies in the same industry, adopting the same kind of AI, so often end up with wildly different results. The benefit never really lives in the technology; it lives in the fit between the technology and a real problem, built well. Keep that in mind reading any list of AI’s advantages, including this one.

The data behind the enthusiasm is real, if you read it carefully. Analyses from McKinsey put enormous numbers on AI’s economic potential, and the annual Stanford AI Index tracks adoption climbing year over year. Both are worth reading with a clear head — the potential is large and genuine, and the gap between potential and realized value is where most of the hard, unglamorous work actually lives.

The honest limits: hype versus reality

Here’s the bit that the breathless coverage completely overlooks. The impact of AI across the various sectors is very real, just extremely uneven. Knowing where the line is is what will differentiate the clever adopter from the one who’s just got burned.

Start with the uneven part. Some industries — finance, retail, logistics, media — have been genuinely reshaped, because they’re data-rich and full of prediction and personalization problems. Others have barely been touched in practice, whatever the headlines claim, because their problems are physical, relational, or short on usable data. “AI is transforming every industry” is true as a trajectory and misleading as a description of today.

The use cases that actually work share a profile, and it’s worth memorizing. There’s a clear, costly problem — not “we should use AI” but “this specific thing costs us a fortune.” There’s enough good data to learn from. The scope is narrow rather than sweeping. And there’s a human in the loop wherever the stakes are high. Use cases missing those traits are where budgets go to die.

It’s also worth being honest about the gap between a demo and a deployment. AI demos spectacularly well — a slick proof of concept can make almost anything look solved. Getting that same capability to work reliably, at scale, on messy real-world data, integrated into the systems people actually use, is a different and much harder job. A lot of the disillusionment around enterprise AI comes from mistaking the first for the second. The demo is the easy 20%; the last 80% is where the real effort lives, and budgeting as if the demo were the finish line is how projects stall halfway.

Cost and return absolutely need to be considered. AI projects have real ongoing costs (compute, data work, specialist talent, maintenance) and a very large number of them simply don’t recoup those costs. They tend not to because they were trying to achieve some capability rather than solve a problem that had a quantifiable cost attached to it. The ones that do pay off are the ones where you could have put a dollar value on the problem before a single line of code was written. ‘Interesting’ is not a business case. ‘This specific failure mode costs us millions a year’ is.

The real barriers, notably, are rarely the algorithms. They’re data quality (most organizations’ data is messier than they think), integration with existing systems, cost, regulation, and the plain human work of change management. NIST and other bodies are building out the frameworks for managing AI’s risks — bias, reliability, transparency — precisely because deploying this responsibly is harder than deploying it at all. The model is usually the easy part; everything around it is the project.

The jobs question is a fair one. AI is fundamentally altering the nature of work in every sector it touches and to pretend it doesn’t is to mislead. The immediate reality in virtually every sector is one of augmentation rather than outright replacement. AI is taking on the drudge work so that humans can concentrate on judgement, on relationships and on the non-routine. The longer term picture is genuinely unknown and bodies like the WEF are monitoring the shift in the balance of jobs created and lost. Anyone claiming to know the net effect is just guessing. The realistic planning assumption is that it’s going to be a process of change and adaptation rather than a sudden clean sweep. That’s exactly what the OECD’s research into AI across economies confirms.

How to find and build your AI use case

If you’re reading this to work out where AI fits your own organization, here’s a sector-agnostic framework that keeps the effort grounded.

1. Start with a costly, data-rich problem — not with “AI.” Find the thing that’s genuinely expensive — downtime, churn, fraud, slow processes, missed demand — and start there. The problem leads; the technology follows. “We need an AI strategy” is how money gets wasted.
2. Check which of the six patterns it fits. Is it a prediction problem? A personalization one? Automation, vision, language, optimization? If your problem maps cleanly onto one of the six, you likely have a real AI opportunity. If it maps onto none, be suspicious.
3. Make sure the data exists and is usable. AI learns from data, and no model survives bad data. Before anyone builds anything, confirm you have enough clean, relevant data — because fixing the data is usually the bigger job than building the model.
4. Pilot narrow and measure against a real baseline. Pick one use case, run it against how you do things today and measure honestly. A small proven win gives you the right to scale. A big roll out that skipped that step is very rarely going to survive contact with reality.
5. Keep humans in the loop where the stakes are high. In anything that’s consequential (health, money, safety, legal) design for human oversight from the outset. It’s not a lack of ambition at all; it’s the way you build a responsible and sustainable system.
6. Plan for integration and change not just the model. The difficult bit is almost always not the algorithm itself but the actual integration into the existing systems and getting people to use it. Budget for that or the most brilliant model in the world is going to sit there gathering dust.

One more thing the framework will not tell you but you’ll learn through experience: you will iterate. The first version of an AI feature is almost certainly not the one that will end up being the version that users actually use. Models drift as the world changes, the data pipelines break and what users actually do with the feature will surprise you. The teams that win treat the launch as the beginning of the work rather than the end of it. It’s about ongoing monitoring, retraining and refinement rather than just shipping once and walking away. AI is less a thing you install and more a capability that you have to keep tending.

None of this is industry-specific, which is the point: the discipline travels even when the use cases don’t. Working out which pattern fits your problem — or whether it’s an AI problem at all — is exactly what a short AI consulting conversation is for, before you commit a budget to finding out the hard way.

What separates the companies winning with AI

Across all these industries, the organizations getting real value from AI tend to look alike — and it has surprisingly little to do with which model they picked. A few shared traits show up again and again.

They start with the problem not the technology. The winners don’t really have an AI strategy so much as they have a list of expensive problems they’re working through (some of which are perfectly suited to AI). They start small and prove it. A narrow pilot with a demonstrable result is what gets you to the next one rather than a moonshot which will burn a year and a budget before you’ve even got a result. They take their data very seriously indeed having learned early that the model is the easy bit and the data is the project. The teams that have clean, accessible and well-governed data are the ones that move the fastest.

They deliberately keep the human in the loop. Rather than going down the route of full automation and then having to work out how to get trust or compliance, they treat AI as a product not a project. You ship it, you measure it, you keep improving it. They have realistic expectations. They’ve got their head around the fact that AI is both powerful and limited. They’re not the evangelists who think it’s going to do everything, they’re not the sceptics who think it’s all just hype. That middle ground is, in the end, the one where the actual value gets captured. It’s a very unglamorous reality.

If you ask me what the one thing we’ve learned from the entire cross-industry exercise it’s that AI rewards the dull virtues: well defined problems, good data, a very narrow scope, human judgement, honest measurement. The firms that have those things are the ones that win in every industry. The ones that are just chasing the word ‘AI’ are the ones that mostly don’t.

Bottom line

AI isn’t one big revolution sweeping through every industry at the same time. It’s those same six capabilities (prediction, personalisation, automation, vision, language, optimisation) delivering real, tangible value in the right circumstances i.e. where there is a clear problem and good data to learn from. It’s utterly underwhelming in the circumstances where it’s just slapped on for the sake of the narrative. The industries that are winning with it aren’t the ones going all in on “AI”. They’re the ones that are actually solving a specific problem which is a costly one.

That’s the lens you need to apply to your own business. Don’t ask “how do we use AI?” Ask “what’s our most expensive, data-rich problem and does it map to prediction, personalisation, automation, vision, language or optimisation?” If it does then you’ve got something to build. If it doesn’t then you’ve just saved yourself a very expensive detour. That one reframing is worth a hell of a lot more than any amount of hype as it takes a rather vague aspiration and turns it into a decision you can actually test.

It’s worth stepping back and looking at the trajectory. The map we have of what’s been transformed and what’s remained essentially untouched is just a snapshot, not a final judgement. The frontier is always moving. Generative AI is moving the knowledge work industries along at pace and sectors which were thought to be immune a few years ago are quietly discovering their data and their use cases. The sensible thing to do isn’t to say that AI will or won’t be relevant to your industry. The sensible thing to do is to accept that it will, in an uneven fashion and on its own timetable. The organisations that have developed the habits we’ve talked about (clear problems, clean data, honest measurement) will be the ones ready to move when their time comes.

If you’re trying to find where AI genuinely fits your industry — or you want a straight answer on whether an idea is worth building — that’s the kind of conversation our team likes. You can get in touch, and if the honest answer is that you don’t need AI for this yet, we’ll tell you. When there’s a real use case in it, our AI development team takes it from there.