Bloor Research International

𝟵𝟬% 𝗼𝗳 𝗔𝗜 𝘂𝘀𝗲𝗿𝘀 𝗮𝗿𝗲 𝗰𝗼𝗻𝘃𝗶𝗻𝗰𝗲𝗱 𝗶𝘁'𝘀 𝘀𝗮𝘃𝗶𝗻𝗴 𝘁𝗵𝗲𝗺 𝘁𝗶𝗺𝗲. The data says it's saving them 2.8%. That gap between how productive AI feels and what it actually delivers is one of the most important findings in AI right now. And most people are on the wrong side of it. A study of 25,000 workers by economists at the Universities of Chicago and Copenhagen found the 𝗺𝗲𝗮𝘀𝘂𝗿𝗲𝗱 𝘁𝗶𝗺𝗲 𝘀𝗮𝘃𝗶𝗻𝗴𝘀 𝗮𝘃𝗲𝗿𝗮𝗴𝗲𝗱 𝗷𝘂𝘀𝘁 𝟮.𝟴% 𝗼𝗳 𝘄𝗼𝗿𝗸 𝗵𝗼𝘂𝗿𝘀. A quarter of workers spent more time on the same tasks afterward. Wage impact, even among the heaviest users: near zero. It's not a one-off: → MIT: 95% of corporate generative-AI investments show zero measurable return. → Goldman Sachs: AI's contribution to 2025 US GDP growth has been "basically zero." → Workday: 40% of the time AI saves is lost again to rework. So is AI productivity a myth? At the level most people use it - the chatbot session, the borrowed prompt - largely, yes. Because none of that gain is owned. The time saved evaporates the moment the session ends. Nothing compounds; nothing is left behind. But there's a second, bigger shift the myth conversation misses entirely. The WEF's Future of Jobs Report 2025 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝟮𝟮% 𝗼𝗳 𝗮𝗹𝗹 𝗷𝗼𝗯𝘀 𝗮𝗳𝗳𝗲𝗰𝘁𝗲𝗱 𝗯𝘆 𝟮𝟬𝟯𝟬 𝗮𝗻𝗱 𝟰𝟬% 𝗼𝗳 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗱 𝘀𝗸𝗶𝗹𝗹𝘀 𝗰𝗵𝗮𝗻𝗴𝗶𝗻𝗴. The real question was never "does AI make this task faster?" It's who owns and redeploys their expertise as work restructures underneath them, and who's left holding a skill set with a shrinking shelf life. 𝗧𝗵𝗮𝘁'𝘀 𝘁𝗵𝗲 𝗙𝘂𝘀𝗶𝗼𝗻 𝗘𝗰𝗼𝗻𝗼𝗺𝘆: not renting productivity from a platform session by session, but building a durable, owned asset out of what you actually know, so it compounds instead of resetting to zero every time you close the tab. 𝙏𝙝𝙚 𝙘𝙝𝙖𝙩𝙗𝙤𝙩 𝙛𝙤𝙧𝙜𝙚𝙩𝙨 𝙮𝙤𝙪 𝙩𝙝𝙚 𝙢𝙤𝙢𝙚𝙣𝙩 𝙮𝙤𝙪 𝙡𝙤𝙜 𝙤𝙛𝙛. 𝙔𝙤𝙪𝙧 𝙚𝙭𝙥𝙚𝙧𝙩𝙞𝙨𝙚 𝙨𝙝𝙤𝙪𝙡𝙙𝙣'𝙩. That's the gap Bloor Research International's Fusion Economy was built to close.

𝗧𝗵𝗲 𝗺𝗼𝘀𝘁 𝘃𝗮𝗹𝘂𝗮𝗯𝗹𝗲 𝘁𝗵𝗶𝗻𝗴 𝗶𝗻 𝘆𝗼𝘂𝗿 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗷𝘂𝘀𝘁 𝗵𝗮𝗻𝗱𝗲𝗱 𝗶𝗻 𝗶𝘁𝘀 𝗻𝗼𝘁𝗶𝗰𝗲. And none of your systems know how to keep it. Every organisation eventually loses someone whose knowledge was never written down: the specialist who knew why a client relationship worked, the consultant whose judgement came from twenty years of pattern recognition no manual ever captured. When they leave, that judgement leaves with them. We reach for two tools to stop this. Neither was built for it: → Succession planning decides who takes the role next. It says nothing about what the outgoing person actually knew. (And only 21% of organisations even have a formal plan - SHRM.) → Knowledge management organises documents and FAQs. It can hold a policy. It can't hold the reasoning a 20-year specialist uses to decide when to break that policy. The scale of the gap is the part worth sitting with: Bloor Research International finds that as much as 40 years of specialist knowledge can sit locked inside a single expert's head, with no structure to capture it before they walk out the door. And the appetite to fix it is already there: 98.5% of employees say better knowledge-sharing would make them more productive (Bloomfire). So what actually works?  • Codify the reasoning, not just the output - capture why a decision was made.  • Structure it for reuse, not storage - a doc in a folder isn't captured knowledge.  • Build it before the resignation letter - waiting leaves you weeks to capture decades. The knowledge isn't gone. It's just never been made keepable. 𝙏𝙝𝙖𝙩'𝙨 𝙩𝙝𝙚 𝙜𝙖𝙥 𝘿𝙞𝙜𝙞𝙩𝙖𝙡𝙈𝙚'𝙨 𝙔𝙤𝙪𝙧 𝙆𝙣𝙤𝙬𝙡𝙚𝙙𝙜𝙚 𝘽𝙖𝙨𝙚 𝙬𝙖𝙨 𝙗𝙪𝙞𝙡𝙩 𝙩𝙤 𝙘𝙡𝙤𝙨𝙚.

𝗘𝘃𝗲𝗿𝘆𝗼𝗻𝗲 𝘁𝗵𝗶𝗻𝗸𝘀 𝗝𝗮𝗽𝗮𝗻 𝗮𝗻𝘀𝘄𝗲𝗿𝗲𝗱 𝗶𝘁𝘀 𝗱𝗲𝗺𝗼𝗴𝗿𝗮𝗽𝗵𝗶𝗰 𝗰𝗼𝗹𝗹𝗮𝗽𝘀𝗲 𝘄𝗶𝘁𝗵 𝗿𝗼𝗯𝗼𝘁𝘀. It didn't. And the mistake hiding in that assumption is the one Britain is about to make faster. Japan's real response to a shrinking workforce wasn't technology - it was participation. Female labour force participation pushed to ~76% against an OECD average of 67%. Senior participation was ~25% against 16%. Thirty years of redesigning who works and how, executed at national scale. Only after that ceiling did Japan turn to foreign labour, and only now is it third in line for productivity and AI. By the time AI arrived, it was too late to help cheaply. Not because Japan is behind on tools, but because three decades of accommodating the workforce it could recruit produced a national map of work that AI simply can't read. We call this Labour Debt: the accumulated gap between how work is described and how it's actually done. You cannot automate a process you never properly mapped. Japan doesn't have an AI problem. It has a map problem. Britain sits roughly where Japan was in the early 1990s, with one advantage: the technology arrived before the crunch, not after. That's the whole game, and our research with 500+ HR Directors and CFOs says we're squandering it. Organisations are buying new AI and applying it to job architectures written for a labour market that no longer exists. New tool. Old map. No capacity gain, the exact outcome Japan is living with now. The fix is an order-of-operations fix. Redesign work before you deploy. Map what's actually done, separate the tasks that need real human judgement from the ones that are just habit, then deploy against the redesigned map; not the inherited one. We have been handed the results of a thirty-year experiment for free. Ignoring them would be a remarkable act of national self-sabotage. 𝗙𝘂𝗹𝗹 𝗽𝗶𝗲𝗰𝗲 𝗶𝗻 𝘁𝗵𝗲 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 👇 #FutureOfWork #AI #Workforce #FusionWork #BloorResearch

Every enterprise racing to deploy generative AI assistants and digital twins is solving the same efficiency problem. Almost none of them are solving the same vulnerability. Digital twins have earned their place on the factory floor. They mirror a machine in real time, predict wear, and flag failure before it happens; an indispensable tool wherever physical systems need modelling. Generative AI, meanwhile, has earned its place in the office: drafting, summarising, answering, freeing up hours that used to disappear into first drafts. Two very different technologies, both delivering real value. But look closely at what each one is actually built to protect, and a gap opens up. A digital twin can model a machine's every vibration and still know nothing about the engineer who has spent decades learning to catch the one signal that matters before the sensors do. A generative AI assistant can draft a flawless report and still know nothing about the judgement calls a specific team has made, in a specific market, under specific pressure, for years. Bloor Research's latest analysis puts it plainly: it's synthetic, not sovereign. The knowledge behind the answer was never the organisation's to begin with, and when the person who actually holds that judgement leaves, so does everything neither system was built to keep. That absence carries three concrete risks. Ownership: the expertise increasingly sits inside vendor systems rather than inside the enterprise itself. Value: automation chases efficiency while the context that actually differentiates one team's decisions from another goes unmeasured. Liability: when something goes wrong and the person who understood why is long gone, accountability has nowhere solid to land. None of this argues against digital twins or generative AI; both are doing exactly what they were designed to do. The argument is that neither was designed to do this other thing, and most operating model strategies are being built as though one of them already does. Bloor's analysis points toward a different shape entirely: systems where human and digital capability stay in constant dialogue, where expertise can be audited, owned, and updated rather than quietly rented from a vendor, and where a departing expert leaves behind more than a gap in the org chart. Before the next AI investment gets signed off, the question worth sitting with isn't which tool moves fastest. It's whether you're building a system that protects what only your people know. Read more: https://lnkd.in/gi_XsCqN #GenerativeAI #DigitalTwin #EnterpriseAI #DigitalMe #BloorResearch

Nico DECOCK 𝗹𝗮𝘁𝗲𝘀𝘁 𝗯𝗹𝗼𝗴 opens with a warning employment lawyers might not want to hear: right now, the legal test for whether a business transfer is breaking down, and it's breaking down loudest exactly when outsourcing deals need it to hold. European labour transfer law was built on a simple idea: if the people move, the business moves with them. Count the heads, follow the heads, protect the heads. AI is quietly breaking that assumption. Decock's argument is simple: capability isn't one thing anymore. It splits into layers, and each one can move on its own. There's the contract itself. There's the process written down on paper. There's the judgment a claims handler builds over fifteen years, the kind that never quite makes it into a manual. And there's the trust between people that just comes from working together. Historically, all four moved as one bundle, because the only way to transfer capability was to transfer the people who held it. Now the tacit and relational layers can be lifted out and encoded into a model, while the contractual and codified layers move on paper, and the people don't move at all. Decock points to claims adjudication as the clearest example: a function that employed 200 people five years ago now runs on 15 exception-handlers and an LLM-based adjudication model. The work continues. The economic value continues. The workforce that used to carry it has largely disappeared. Courts are starting to notice. Decock traces a shift in how the CJEU is beginning to frame the identity test that decides whether a transfer has legally occurred - moving from asking whether the people moved to asking whether the models, platforms, and tooling moved. That's not a small technical adjustment. It relocates the entire legal test onto ground the law was never built to examine. The numbers give the shift some urgency. BPO contract values fell 14% in 2025, the lowest point since 2020, and Everest Group estimates that 25-40% of FTEs could be displaced across major BPO markets. That's not a forecast about some future wave of automation. It's a description of what's already showing up in contract renewals. If the law keeps measuring transfers by counting people, it will keep missing the transfers that matter. Read more: https://lnkd.in/gKAMhrJj #IntelligenceArchitecture #FutureOfWork #EmploymentLaw #AI #BPO #BloorResearch

Darren Sack'𝘀 𝗹𝗮𝘁𝗲𝘀𝘁 𝗯𝗹𝗼𝗴 opens with a warning CFOs might not want to hear: right now, EBITDA is lying, and it's lying loudest exactly when you need it to tell the truth. Here's the mechanism. As organisations embed AI, costs aren't disappearing; they're migrating - out of wages, where EBITDA can see them, and into depreciation, interest, leases, and cash provisions, where it can't. The result looks like profitability. It's often just relocation. The numbers make the point better than the theory does. Oracle posted record revenue and a $638bn contracted backlog in FY2026, up 363%, and still burned through roughly $23.7bn in free cash flow. CoreWeave pulled off a 56% adjusted EBITDA margin in the same quarter it reported a net loss. Neither of those is a contradiction. That's what it looks like when the metric everyone's watching stops measuring the thing everyone actually needs to know. Darren traces this back to three accounting mechanics working together: lease accounting rules that reclassify real operating costs as depreciation and interest, capital spending that lands below the line as depreciation rather than above it as wages, and what he calls Ghost Liabilities: accruals, provisions, severance charges, and payables that never touch EBITDA but still have to be paid, in cash, eventually. Labour debt, the fixed payroll cost of roles an operating model no longer needs, is one of ten distinct hidden-debt categories he maps this way, each one a different route for the same trick: rename the cost, and the metric improves even though nothing got cheaper. Darren's own distinction is the one worth sitting with: reform adjusts the past, reset designs the future. Most of what gets called AI-driven efficiency right now is the former, a cleaner-looking number, not a redesigned cost base. The question worth asking isn't whether EBITDA improved this quarter. It's what got renamed to make it look that way. Read more: https://lnkd.in/gEy6wSn7 #CFO #EBITDA #FinancialStrategy #AI #EnterpriseFinance #BloorResearch

September's First Friday Newsletter is here. In this edition: Daniel Howard: 𝗢𝗻 𝗞𝟮𝘃𝗶𝗲𝘄 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗧𝗲𝘀𝘁 𝗗𝗮𝘁𝗮 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁. Test data provisioning has quietly become the bottleneck holding back AI-enabled software development. Daniel looks at K2view's new agentic layer, which builds and delivers compliant test data straight from natural language requirements. Bloor Research International: 𝗧𝗼 𝗕𝘂𝗶𝗹𝗱 𝗼𝗿 𝗡𝗼𝘁 𝘁𝗼 𝗕𝘂𝗶𝗹𝗱 - 𝗔𝘁𝗮𝗰𝗰𝗮𝗺𝗮. For years, only smaller companies with specialised needs seriously considered building their own data management capability in-house. Now some of the largest enterprises with the most sophisticated data functions are asking the same question, and the paper argues the hidden and opportunity costs of building usually outweigh the benefits, especially when speed to value matters more than ever. Paul Bevan: 𝗢𝗻 𝗧𝗵𝗲 𝗖𝗿𝘆𝘀𝘁𝗮𝗹 𝗕𝗮𝗹𝗹 𝗧𝗵𝗮𝘁 𝗡𝗲𝘃𝗲𝗿 𝗪𝗮𝘀. Observability vendors have spent a decade promising to predict outages before they happen. What they've actually built, Paul argues, is anomaly detection with a better UI, and the fix isn't a flashier model; it's the dependency graph nobody wants to build. Cheney Hamilton: 𝗢𝗻 𝗔𝗜 𝗔𝗿𝗿𝗶𝘃𝗲𝗱 𝗧𝗼𝗼 𝗟𝗮𝘁𝗲 𝗳𝗼𝗿 𝗝𝗮𝗽𝗮𝗻. Japan ran out of workers before it ever redesigned how they worked, so AI arrived with nothing left to fix. Britain is heading toward the same demographic wall; UK births just hit a 1977 low, but Cheney's point is that the order hasn't been set yet: AI is here early enough to redesign the work first, if anyone actually does it. Nico DECOCK: 𝗢𝗻 𝗧𝗵𝗲 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗖𝗮𝘀𝗲 𝗬𝗼𝘂𝗿 𝗔𝗜 𝗕𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗖𝗮𝘀𝗲 𝗡𝗲𝘃𝗲𝗿 𝗠𝗲𝗲𝘁𝘀. Every organisation weighs hiring, automation, and outsourcing through separate governance channels, each with its own approval criteria. Nico's argument: the choice among them is governed by nobody, so decisions get made by whichever budget bucket has headroom that quarter, not by comparative economics. Darren Sack: 𝗢𝗻 𝗪𝗵𝗲𝗻 𝗘𝗕𝗜𝗧𝗗𝗔 𝗟𝗶𝗲𝘀. For years, EBITDA has been the headline number investors and boards trust to show how a business is really performing. Now, as AI reshapes operating costs, that trust is starting to look misplaced. Darren points out: the real cost isn't disappearing; it's migrating into depreciation, leases, and interest, where EBITDA was never built to look. https://lnkd.in/g6durDAE  #AIOps #FutureOfWork #EnterpriseAI #DataManagement #BloorResearch

𝗥𝗶𝗰𝗵𝗮𝗿𝗱 𝗦𝗸𝗲𝗹𝗹𝗲𝘁𝘁'𝘀 𝗹𝗮𝘁𝗲𝘀𝘁 𝗯𝗹𝗼𝗴 𝗼𝗽𝗲𝗻𝘀 𝘄𝗶𝘁𝗵 𝗮 𝘀𝘁𝗼𝗿𝘆 𝗳𝗿𝗼𝗺 𝟭𝟯𝟭𝟰: Edinburgh Castle's walls held. Nobody breached them. The attackers didn't need to; they simply climbed the cliff face nobody thought to defend and took the castle. That's roughly what Richard argues is happening to competitive moats in enterprise software right now. For years, vendors have measured their competitive position with one question: can a competitor replicate what I have built, on my terms, inside my category? Call that the classical moat test - switching costs, integration lock-in, and years of accumulated data. It's a real test, and plenty of vendors still pass it. The trouble is that passing it doesn't mean much anymore, because it's the wrong wall being tested. There's a second question hiding underneath the first: is the category itself still the one the market is organising around? That's the category relevance test, and it's the one nobody's climbing over the wall to answer, because the market's already moved past it. What changed underneath all this is three things quietly collapsing at once.  Replication cost: the tools it takes to build a category-defining product have become radically cheaper, so what used to be years of engineering moat is now closer to months. Cycle time: the eighteen months a vendor used to count on before a rival reached parity is no longer a safe assumption. And information asymmetry: knowledge that used to live locked inside a product, hard-won and hard to copy, is now extractable and synthesisable straight out of public documentation. Take HCM as an example. A vendor like Workday can have a genuinely strong moat there - deep integrations, years of switching costs, real defensibility by the classical test. None of that matters much if the market itself is quietly reorganising around Strategic Workforce Planning instead, a category that asks an entirely different set of questions than HCM ever did. The walls hold. The war already moved. The moat was never the wall. Stop measuring how well-defended your category is, and start asking whether it's still the category the market is fighting over. Read more: https://lnkd.in/eKh8wskQ #EnterpriseAI #AIStrategy #EnterpriseArchitecture #BloorResearch

Nobody has declared this war. That doesn't mean it isn't already being fought. The real contest in AI right now isn't about who builds the smartest model. It's about who ends up owning the infrastructure that model runs on, because that's where the value actually settles.  Bloor Research International puts the prize at somewhere between $105 and $117 trillion of global GDP, and the split isn't going to fall to whoever has the most capable tools. It's going to fall to whoever owns the pipes. Right now, three groups are fighting over that split, and only two of them seem to know it. The US has built platform dominance through licensed foundation models and cloud infrastructure that the rest of the world runs on top of. China has taken the opposite path, building its own sovereign stack and exporting it as a parallel ecosystem. Everyone else - the UK, the EU, most enterprises - is a net consumer of infrastructure someone else owns. Saudi Arabia is the exception that proves the rule: a $40 billion sovereign investment through its Public Investment Fund, built specifically so it isn't stuck renting someone else's future. That gap shows up in numbers this series has already covered: a 1.3 million-person structural employment gap in the UK, and 26 to 65% of layoffs concentrated in knowledge-intensive roles. Those aren't separate problems. They're what it looks like when an economy consumes AI infrastructure instead of owning it; the value leaves faster than anyone's tracking it. Bloor's answer works at three levels at once. Nations need to treat this as a deliberate decision about who owns the infrastructure, not a technology procurement exercise. Organisations need what Bloor calls a sovereign intelligence architecture; real ownership over AI capability, data governance, workforce design, and where the value actually gets captured, not just a licence to use someone else's.  And for individuals, that same principle becomes 𝗕𝗹𝗼𝗼𝗿'𝘀 𝗗𝗶𝗴𝗶𝘁𝗮𝗹 𝗠𝗲: portable professional knowledge that stays owned by the person who built it, instead of quietly getting extracted into systems they'll never own a share of. Bloor puts a date on how long this window stays open: 2028. After that, catching up stops being difficult and starts being structurally impossible for most organisations and nations. The contest for the productive digital future is already underway. The question worth asking now isn't whether you're using AI. It's whether you're building any of it, or just renting someone else's. Read more: https://lnkd.in/dmab2rNA #AIStrategy #DigitalSovereignty #EnterpriseArchitecture #FutureOfWork #BloorResearch

Donna Lamden's latest blog starts with the sales meeting every vendor has sat through: the presentation is polished, the innovation is real, the passion for the product is undeniable. And it's rarely what buyers remember. The explanation gets specific fast. Most product teams build around a hierarchy without realising it: a feature is a property of the product, an advantage is what that feature does within its own part of the system. Neither of those is what actually sells. The benefit, the thing buyers actually remember, only exists in the relationship between the part and the whole. You can't demonstrate it by talking about your product in isolation, because it doesn't live inside your product. It lives in how your product changes the buyer's entire system. That's the gap Donna keeps coming back to: vendors solve for the problem their product was built to solve, at the part level. Buyers are living with a problem at the whole level; how everything they already run is wired together, and where the new piece is actually going to sit inside that. Solve the wrong-scoped problem well, and buyers can tell within minutes, even when the demo goes perfectly. Donna points to one vendor who did it differently: instead of opening with features, they asked how the buyer's existing systems were wired together, then explained exactly where their solution would sit inside that chain of inputs, outputs, and outcomes. Everyone else in the room was still pitching parts. That vendor was the only one describing the whole. Before your next customer meeting, ask yourself: what do you actually want buyers to remember when the presentation is over? Read more: https://lnkd.in/g_6Bkr3C #SalesStrategy #B2BSales #CustomerExperience #EnterpriseSales #BloorResearch