The Rise of the Synthetic Organisation Part 3

As aChief Transformation Officer I have worked with sat through a quarterly review, looking at a slide that said her AI programme was on track. The pilots were running. Adoption was climbing. The vendor scorecards were green. The board was reassured.

Not one thing in her company had fundamentally changed.

Not the operating model. Not the work. Not the structure through which decisions get made. The programme was not failing. It was being absorbed.

Part I is named the immune system. Part II named the moves to overcome it. Part III is about the work of redesigning the operating model while it is still running -- slowly, deliberately, against the resistance of every system the organisation built to defend itself.

It is the hardest work in this cycle. It is also the only work that turns an AI investment into anything durable. Almost no one is doing it well.

McKinsey's 2025 State of AI survey tells the story of absorption in three numbers. Eighty-eight per cent of organisations now use AI in at least one function. Only a third have scaled it across the enterprise. Just thirty-nine per cent report any EBIT impact at the enterprise level.

The pilots run. The dashboards refresh. The slide pack circulates. The operating model does not move.

Boston Consulting Group studied more than two hundred at-scale AI transformations and found the value distributes ten-twenty-seventy: algorithms, technology, and then everything to do with people, processes and change. The technology is one-tenth of the answer. The workforce is seven-tenths. Yet capital across global enterprises flows the other way -- into platforms, not into the workforce architecture required to use them.

This is the single largest misallocation of capital in the current AI cycle.

Part I introduced Klarna as the working example of a synthetic organisation. Eighteen months later, the story has a second chapter. Customer satisfaction fell. Klarna brought human agents back to handle the work AI could not. The correction is the lesson, not the failure. Klarna did not get into trouble because the AI was bad. They got into trouble because they treated the workforce question as a cost question, and discovered the hard way that it was a capability question.

The technology was never the part that needed to change. The workforce was. The workforce still is.

Build the Right Business Case

The business case for AI is being built wrong almost everywhere it's built.

It is being framed as a cost story—headcount avoided, hours saved, processes automated. CFOs have heard that story for twenty years. They have learned to discount it. They have watched a decade of digital transformation programs promise efficiency and, mostly, deliver more software licenses and a slightly larger IT function.

A durable AI business case has four layers.

The first is the cost layer. What the program saves. Real numbers, but one-time events. They are the ticket to the meeting, not the reason the meeting matters.

The second is the capability layer. What the workforce will be able to do in twelve months that competitors cannot. McKinsey's data is unambiguous: among AI high performers -- those capturing more than five per cent of EBIT from AI -- the strongest single predictor of value is fundamental workflow redesign, not tool adoption. Yet only twenty-one per cent of organisations using generative AI have redesigned any workflows at all. Almost eighty per cent are layering AI on top of existing processes and waiting for the value to appear.

It does not.

The third is the strategic layer. What the organisation will become in three to five years. This is the question Part II asked the telcos to answer when voice calls became free: what business are you actually in? The cost case never asks. The strategic case answers and commits capital accordingly.

The fourth is the human layer -- the layer most AI business cases skip entirely. People need a role in the organisation, and the role needs to be legible to them. The technology is a tool. The workforce is the agent. Pretending otherwise produces a workforce that quietly disengages from a future it cannot see itself inside.

The CFO reads the cost layer. The COO reads the capability layer. The CEO and the board read the strategic and human layers. The Chief People Officer reads all four. A case written for one of these audiences and not the others gets funded, and quietly stalls.

Map the Use Cases at the Functional Level

This is where most AI business cases break down.

The cases that get funded describe AI at the industry level—AI in financial services, AI in healthcare, AI in retail. The cases that actually move the operating model describe AI at the function level. These are not the same exercise. Treating them as the same is the most common reason serious capital gets approved against thin reasoning.

The functional map names three things for each function within the enterprise. The specific use cases that matter. The operating model shift that each one forces. And the redesign required to actually capture the value rather than layer the AI on top.

The third is the one that decides whether the program delivers.

The reflex of every function under cost pressure is to bolt AI onto the existing workflow, preserve the existing roles, and claim productivity. This is the most expensive mistake in the cycle, and it is being made at scale right now. The value is not in adding AI to the work. The value is in getting out of the way of the AI -- redesigning the work so the agent can do what it does best, and redeploying the humans to where they compound rather than where they supervise.

A workforce kept in its current shape and asked to oversee tools has absorbed the cost of the technology without unlocking the upside.

Four functions illustrate the discipline.

Workforce and People. The cadence changes from quarterly to continuous. The output changes from a forecast to a live model. The skill mix shifts from analyst-heavy to model- and judgment-heavy. The redesign required is to collapse the gap between planning and execution—to let the line manager interact with the model directly rather than receive a deck two weeks late from a central team. Layered on top of the existing function, the AI produces faster forecasts that arrive too late. Redesigned around it, the function disappears as a separate motion and reappears as an always-on capability inside the business.

HR Operations. The shared-services model that most large organisations have run for fifteen years was built on a specific economic logic -- concentrate transactional work in one place, drive down cost per ticket, and escalate complex cases to specialists. AI breaks that logic. The centre of gravity moves from the ticket queue to the exception. The redesign is to redeploy the people who used to handle volume into roles that handle complexity. Policy interpretation. Case advocacy. Employee experience design. The work where human judgment is the actual product.

Talent Acquisition. Picture a TA leader running a function of forty recruiters across three regions. She has six months of data on an AI sourcing agent her team has been using inside the existing process. The agent finds candidates faster, but conversion has not improved, and recruiter satisfaction is falling. She does not buy more licences. She rebuilds the function. Sourcing, screening, and scheduling go to the agent. The recruiters -- fewer of them now, but more senior -- are redeployed against two motions the old workflow could not afford: candidate advocacy through the offer stage, and hiring-manager coaching at the front of the funnel. Time to hire halves. Quality of hire moves for the first time in three years. The cost line goes down because the headcount went down. The capability line increases because the work was redesigned rather than automated. This is the move almost no TA function is making.

Operations. The use cases here are vast—demand forecasting, predictive maintenance, quality inspection, scheduling, and exception management. The failure mode is consistent. AI gets deployed against the existing process steps and produces marginal improvements at each one. The redesign that captures value is the one that asks which process steps should no longer exist. A maintenance function with predictive AI does not require the same preventive maintenance cadence. A planning function with live demand signals does not need the same S&OP rhythm.

The same logic applies to Finance, where private equity sponsors have already moved ahead of the public market -- eighty-five per cent of PE buyers now factor AI-enabled finance capabilities into company valuations, and finance functions that have not been redesigned are being penalised at exit.

Without the functional map, the cost layer is the only layer anyone can actually trace. That is exactly why programs get absorbed.

With it, the CFO sees cost, the COO sees capability, the CEO sees strategy, and the workforce sees - concretely, by function, with names against the work - what the human layer means for them.

The 5B Workforce Architecture

The 5B model -- Build, Buy, Borrow, Bot, Bundle -- was originally a way of thinking about workforce supply. Applied to AI, it becomes a workforce architecture for the synthetic organisation. Each lever does different work, costs different money, carries different risk, and matures on a different clock.

Build is the only capability that compounds inside the organisation. Internal practitioners who understand both the domain and the technology deeply enough to redesign work, not just operate tools. Without them, every AI investment must be mediated by an external partner, and the capability never accrues. It is rented from someone else. BCG's 2025 AI at Work survey shows the cost of deferral already compounding: less than a third of organisations have upskilled even a quarter of their workforce. [5] Build is also the slowest lever -- eighteen to thirty-six months to mature -- which is precisely why it is almost always deferred.

Buy is the lever boards reach for first. Hire the Chief AI Officer. Hire the head of data science. Show the market you are serious. It is also the lever most likely to disappoint. Stanford's 2025 AI Index notes the irony: the companies leading in AI investment are the ones finding it hardest to attract and retain top AI talent. [6] Individuals hired into organisations not structurally ready for them tend to leave within eighteen months -- after concluding that the role they accepted and the role they were given are not the same role. The immune system absorbed them, gave them a title, and changed nothing else.

Borrow -- consulting partners, vendor professional services, specialist firms -- is the right lever for accelerating early moves and holding the line while internal Build matures. Used well, it puts experienced practitioners alongside internal candidates from day one, so capability transfers as the work is being done. The discipline is to use Borrow as scaffolding, with internal Build deliberately constructed underneath it while the scaffolding is still up. The best partners encourage exactly that -- they know an engagement that ends without leaving capability behind is one neither side should be proud of. The risk to manage is not the partner. It is the client-side instinct to defer Build because Borrow is working, and to discover three years later that the capability was always rented.

Bot is the agent layer. The work that should be done by AI is repetitive, rules-based, high-volume, or simply better suited to a machine. The standard mistake is to treat Bot as an automation of existing tasks. That gets you cost savings on processes you should not be doing in the first place -- a faster way to do the wrong work. Klarna's course-correction makes the point. The AI assistant did not fail at the work. It succeeded at work the company should have redesigned before automating.

Bundle is the SaaS layer -- the platforms that come with workflow, data models, and AI capability already built in. Talent intelligence. Workforce analytics. Agentic workflow. This is no longer optional infrastructure. It is the substrate the synthetic organisation runs on. Trying to transform without it is like trying to run a modern finance function without an ERP. Technically possible. Practically suicidal.

Each of the five is a different bet, on a different timeline, with a different risk profile. Treating them as interchangeable is the most expensive mistake in this space

Build the People Who Build It

Inside every successful AI transformation is a small group of people who can do something most of the workforce cannot.

They can stand between a business problem and a model and make the right call. They understand the domain deeply enough to know what good looks like. They understand the technology deeply enough to know what it can and cannot be trusted to do. And they have the standing to make decisions that change how work gets done, rather than recommendations that get politely tabled.

These are Part II's Black Belts, now doing the operational work.

Picture one of them. She has 12 years in supply chain operations and 2 years of serious engagement with large language models -- not as a hobbyist, but as someone who has built, tested, and discarded a dozen agent workflows within her own function. A regional planner comes to her with a question the business has been asking for months: Can we use AI to compress our forecast cycle from three weeks to one?

She does not answer yes. She does not answer no.

She takes the planner through three questions. Which part of the three weeks is the AI actually good at. Which part depends on tacit judgement the model cannot replicate. And what happens if the forecast is wrong faster? By the end of the conversation, the question has been rewritten. The answer is not a tool. It is a redesigned workflow that uses AI for two of the three weeks and keeps a senior human in the loop for the third—because that is where errors become expensive.

Nobody outside the organisation could have given that answer. She did not give it either. She built the conditions under which the planner gave it himself.

These people are rare. They are also the single most important capability the organisation can develop, because every other lever in the 5B model depends on them. Without them, Buy hires arrive in a vacuum. Borrow partners cannot be effectively absorbed. Bot deployments automate the wrong work, beautifully. Bundle platforms get bought but not adopted.

BCG puts a number on it. Organisations that combine workflow redesign with structured upskilling produce roughly four times the three-year shareholder return of AI laggards.

This is a multi-year programme, not a training course. Choose the right people -- domain depth and intellectual range matter more than current technical skill. Give them real problems, not sandboxes. Pair them with external expertise during the learning phase. And protect them from being absorbed back into business-as-usual the moment they become useful.

Organisations that get this right end up with a capability competitors cannot replicate. Organisations that do not are permanently dependent on whoever they bought their last platform from.

There is no third option.

The Absorption Problem Is Now the Market

The most important development in this AI cycle is not a model release.

It is the dawning realisation among the platform vendors that the constraint on their growth is no longer the technology. It is the customer's ability to absorb the technology and redesign the operating model around it. The vendors have looked at the absorption problem, recognised it as the central battleground of the next five years, and started building the ecosystem to solve it.

ServiceNow is the clearest case. In December 2025, HFS Research published a study finding that more than two-thirds of ServiceNow implementations stall after the first wave of use cases -- and crucially, not because the platform is failing but because the service delivery model around it has not adapted to an AI-first operating model. [8] That is the absorption problem, as identified by an industry analyst, for a specific platform category.

ServiceNow's response has been to lean into operating-model partnerships explicitly. In February 2026, Kearney launched BPR.ai on the ServiceNow platform -- a business-process-reengineering practice whose pitch is "fundamental reimagination of how work is structured" rather than tool deployment. [9] TCS signed a multi-year partnership the same month, specifically targeting hire-to-retire HR, finance, supply chain, procurement and employee services -- the functional map, delivered by a systems integrator on a vendor platform. [10] In April 2026, DXC extended its seventeen-year ServiceNow relationship with a Customer Zero deal, backed by eighteen hundred ServiceNow consultants.

Databricks has built the same machine at a larger scale -- a partner network of more than 500 certified consulting firms representing roughly 100,000 practitioners worldwide. The featured roster reads like the global consulting industry's table of contents. Accenture. Capgemini. Deloitte. EY. Infosys. PwC. TCS. Wipro.

And Anthropic has taken the same position in two complementary moves. In October 2025, the alliance with Deloitte became Anthropic's largest enterprise deployment - Claude was rolled out to all 470,000 Deloitte professionals across 150 countries, with 15,000 certified practitioners and a joint Centre of Excellence. [13] Then, in May 2026, Anthropic co-founded a standalone enterprise AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs—with $1.5 billion in committed capital —targeting mid-sized companies that large systems integrators do not economically serve at scale. [14]

Anthropic's framing of the move was careful. Enterprise demand for Claude is significantly outpacing any single delivery model. Translated: protect the Big Four channel for the large-enterprise work where the systems integrators lead, and stand up a complementary vehicle for the mid-market where they do not.

The pattern across all three vendors is identical. They have understood that the absorption problem -- the gap between technical capability and organisational ability to use it -- is the binding constraint on their own growth. They are not trying to solve it alone. They are building consulting ecosystems, sometimes co-investing in them, because the operating-model redesign work cannot be sold as software. It has to be delivered by people who can sit inside the client organisation, redesign the work, and transfer the capability.

That is consulting work. It always has been.

Two consequences follow.

The Bundle and Borrow decisions are no longer separate. Choosing a platform is choosing an ecosystem of delivery partners, and the partners are now part of the substrate.

And the platform vendors are spending billions on the assumption that their customers will not be able to do this redesign work themselves. They may well be right for the next three years. But the organisations that prove them wrong over the next five -- the ones that build genuine internal capability under the scaffolding -- will be the ones whose AI investment compounds rather than dissipates.

Build is still the answer. The scaffolding is now better, more abundant, and more sophisticated than it has ever been.

The question is whether you build under it while it is up.

Sequence the Moves

The order of the five moves matters more than the moves themselves.

The sequence that works is roughly: Bundle, Borrow, Build, Buy, Bot. Platforms first, because everything else depends on the substrate. External expertise second, to deploy them well. Internal practitioners are developed in parallel, so capability transfers occur as the work is being done. Selective hires into specific gaps. And Bot deployed last, against work that has been deliberately redesigned rather than lifted and shifted into an agent.

The sequence that fails equally repeatedly is the inverse.

Bot first, because it looks like a quick win. Buy second, because the bots are not delivering, and someone needs to be accountable. Borrow third, because the hires cannot fix it alone. Bundle fourth, because the underlying platform problem becomes undeniable. And Build never -- because by the time the organisation realises internal capability is the actual answer, the budget is gone and the appetite is gone with it.

The programme is declared a partial success. A pause is recommended. The absorption is complete.

Instrument the Programme So the Board Can See It

Every AI transformation that stalls stalls at the same moment. Eighteen months in, when the cost case has been delivered, and the capability case has not yet been compounded. The board looks at the numbers, does not see the compounding, and asks the question every serious transformation eventually faces. Is this still working?

A capability dashboard looks different from a cost dashboard.

It tracks how many internal practitioners have moved from training into production work. How many workflows have been redesigned rather than automated? How many agents are running under human oversight with measured handoff quality? And one number the cost dashboard never carries -- the share of AI value the organisation could continue to generate if every external partner left tomorrow.

That number is the cleanest available measure of whether the capability is compounding inside the organisation or only while the invoice is being paid.

A board that can see the capability accruing will keep funding the programme through the deferral years. A board that can only see cost will kill the programme at month eighteen and call it a lesson learned.

Which board you are building for is a design decision, made early, by whoever writes the reporting pack.

The Three Horizons

The next ninety days. One business unit. One workflow. One honest redesign. Not a pilot. A redesign. Pick a function where the work is legible, the leadership is willing, and the metrics are clear. Put a borrowed expert alongside a nominated internal candidate who will become a Black Belt. Write the four-layer business case before any tool is procured, with the functional map attached so the work is concrete on the page. Decide on the substrate platform. Do not revisit the decision when the first procurement objection arrives.

By day ninety, the redesign has delivered one thing the old workflow could not -- and one internal practitioner has moved from learner to leader.

The next twelve months. Three to five workflows redesigned, each on the same substrate. A small cadre of Black Belts -- five to fifteen people, depending on organisation size -- operating inside the business rather than as a centre of excellence. A capability dashboard running alongside the cost dashboard, visible to the board. When the first hires are made against gaps, the Build pipeline cannot close fast enough.

The next three years. The operating model has been redesigned, quietly, one function at a time, while the business kept running. The workforce looks different—not smaller, necessarily, but differently shaped, with work redistributed among humans and agents in ways the old org chart could not have produced.

The organisation can answer the question Part II asked the telcos. What business are you actually in, when cognitive work is free?

The answer is not a slogan. It is visible in how decisions are made, how capital is allocated, and how the work itself is designed.

The Choice in Front of You

The technology is no longer the hard part.

The models are good enough. The platforms are mature enough. The integration patterns are well known. The consulting ecosystem to deliver them has never been deeper.

What is hard -- and what almost nobody does well -- is redesigning the operating model while it is still running. That work is slow, expensive, and produces no quarterly headline. It is exactly the kind of work every institutional reflex in a large organisation is designed to defer.

It requires the leadership ingredients Komori named in Part II -- heart, intuition, courage, balance, decisiveness -- applied to a domain most boards still treat as an HR problem.

It is not an HR problem. It is the redesign of the operating model while it is still running, against the active resistance of every system it created to defend itself. That is a CEO act.

Part II ended on the question the telcos had to answer when voice calls became free: what business are you actually in? The Part III version is sharper.

When cognitive work becomes free, what workforce are you actually building?

Not the org chart you have. Not the headcount plan you submitted. The workforce that will exist in three years, by design or by default, depending on what you decide -- or fail to decide -- in the next twelve months.

The scaffolding is up. The question is whether you build under it.

The synthetic organisation is not the one with the most AI. It is the one whose workforce was deliberately designed to make the AI matter.