The argument
- In the measurable parts of the services market, pricing is coming apart from headcount. The reading is contested and we set out the rival explanations below.
- That makes capability, not people, the unit the enterprise buys, builds and loses. It now originates from people, models, licences, tooling and configurations of all four.
- The asset question applies symmetrically. Human and machine capability both create value, both create liability, and neither is reliably one or the other.
- The consequence is structural rather than rhetorical. Composed capability has no single owner in the current executive model, and no function can claim it by inheritance.
Start with the number, not the mantra
ISG reports that BPO annual contract value fell 14% in 2025, the lowest since 2020, while underlying activity held broadly steady. ISG attributes this to automation compressing task-centric, seat-priced work. Everest Group estimates automation could displace 25 to 40% of FTEs in major BPO markets.
What the number does not prove
Annual contract value measures what clients pay, not where capability sits. A 14% fall is consistent with our reading. It is also consistent with several others, and an analyst house should say so before building on it.
Competitive pricing pressure in a crowded provider market would produce the same line on the same chart. So would shorter contract terms, smaller deals, or renegotiation of existing work without any change in how it is done. None of these require capability to have moved anywhere.
What lifts the automation reading above the alternatives is that it is not the only indicator pointing that way. Everest models material FTE displacement. HFS reports labour arbitrage has stopped differentiating providers, which is a statement about what clients will pay for. Three houses, three methods, one direction.
That is corroboration, not proof. The pricing signal is consistent with capability decoupling and has not isolated it.
What the old claim was for, and what it assumed
“People are our greatest asset” did real work. It moved HR out of administration and established people as a source of value rather than a cost. Dave Ulrich carried that further, from talent to organisational capability to stakeholder value, and his framing of talent advantage as a product of AI and human ingenuity continues it.
The progression was never wrong. It carried an assumption that only becomes visible now: that capability and people were the same object, so improving one improved the other. That held while people were the only source of productive capability available.
It was unstable even then. Excellent people inside duplicated structures or obsolete work destroy value reliably. The economically relevant unit was never the person alone; it was what the enterprise could produce through them. Ulrich and Lake were pushing at that distinction in 1990.
Where the tool stops being a tool
The usual framing is that AI helps people work. Human performs, technology enables. That holds for most deployments today.
It stops holding at a specific boundary: when a system interprets information, generates options, decides within bounds, executes and coordinates with other systems to produce output the enterprise sells. At that point it is not enabling capability. It is supplying it.
To see what moves when headcount falls, capability has to be decomposed rather than treated as one object. Our transfer research separates four layers.
- Contractual. Roles, terms, accrued rights. Moves by law. Everyone manages this layer.
- Codified. Process documentation, models, data, tooling, IP. Moves by contract and licence, and increasingly needs no person attached.
- Tacit. Judgement, sequencing, workarounds, what normal looks like on a bad day. Moves only if the individual transfers and stays.
- Relational. Customer trust, regulator familiarity, institutional credibility. Attaches to people and rebuilds slowly or not at all.
Automation moves the centre of gravity from layers 3 and 4 into layer 2. It does not eliminate the human layers, which remain hardest to replace, but it shrinks the headcount they require while the codified layer carries more of the output. That is why people and workforce are ceasing to be synonyms. The workforce becomes a composition: employees, contractors, models, agents, machines and configurations of all of them.
What would have to be true for the old model to hold
Worth stating as a test rather than a conclusion. If people remain the correct atomic unit, we should observe the following. Human beings remain the primary productive actors. AI augments rather than supplies. Decision authority stays structurally human. Workforce economics continue to correlate with headcount and labour cost. Job-based structures remain adequate descriptions of how capability is organised.
Where those hold, the traditional model survives and should be left alone. Where they stop, it needs replacing.
That makes this empirical rather than philosophical. The conditions are breaking unevenly: ISG finds industry-specific BPO the least AI-enabled category, with 58% still in manual or labour-led models and only 6% at AI-first outcomes. The direction is clear. The distribution is not.
The asset question is symmetrical
Human capability creates value and also becomes liability: skills obsolesce, capacity exceeds demand, the economics of the work shift underneath it.
Machine capability has the same duality and is not the benign half of the pair: technical debt, vendor dependency, governance exposure, concentrated operational risk, supervision costs. ISG finds 56% of enterprises say they lack the internal skills to govern AI-enabled BPO effectively, a liability sitting on the asset side of most business cases.
Which makes it an architecture problem, not an HR one
Replace part of a human workflow with a model and you have changed cost, capacity, authority, accountability, risk, dependency, decision rights and procurement exposure. Finance owns the economics, technology the architecture, risk the exposure, procurement the dependency, HR the human consequences.
On our reading, nobody owns the composition.
That is not a turf observation, and it should not be settled by whichever function argues hardest. HR cannot claim it by having historically managed employees, technology cannot claim it because AI is technology, finance cannot claim it because capability has a cost. Each brings a partial model to a whole problem. Neither candidate owner currently reports the readiness that owning this would require. WTW finds 65% of HR professionals feel unprepared for the M&A volume they expect; the ISG 56% governance gap says something comparable from the technology side. Neither statistic is about composed capability directly, which is itself the point. The problem is not yet measured from either side, and unmeasured problems do not get owned.
The question underneath
The progression from people to talent to organisational capability to stakeholder value is not wrong. It is incomplete, because it assumes capability resides in people.
Our confidence in that reading is moderate. The evidence is consistent across three independent houses and the mechanism is intelligible, but it rests on market-level indicators rather than on measurement inside enterprises, and the distribution is uneven enough that the old model still describes most of the market accurately today.
Where capability resides partly outside people, advantage stops being about having the best workforce or the best AI. It becomes a question of how the two are composed into something that produces value at acceptable risk.
The better question is which capabilities the enterprise requires, where each should reside, and what configuration produces value at acceptable risk. It has no function to route it to, which is why it belongs at board level rather than inside one.
The gap this leaves is specific. Nobody publishes contract-level data pairing headcount against output volume across a delivery-model change, which is the measurement that would separate capability decoupling from ordinary pricing pressure. Nor is there transition-specific attrition data, or any measure of service degradation attributable to knowledge loss rather than process change. Bloor is opening a research programme to close those gaps with primary data. If you have redesigned a service around AI-enabled delivery since 2023, particularly one where headcount fell and output did not, I would like to compare notes.
Sources
ISG Index Insider, February 2026, and 2026 BPO Market Lens. Everest Group, November 2024. HFS Research, June 2026. WTW M&A readiness survey, reported June 2025.
Ulrich, D. (2025), “Talent Advantage = AI (Artificial Intelligence) × HI (Human Ingenuity): A Formula for Business and HR Leaders,” LinkedIn, 28 October 2025. Ulrich and Lake, Organizational Capability: Competing from the Inside Out (1990). Nyberg, Kehoe, Ulrich and Wright (eds.), The Age of HR: Delivering Stakeholder Value Through Strategic Organizational Capability (2026), Center for Executive Succession.
Analysis of operating-model design. A companion research note, Capability Continuity in Cross-Border Transfers, sets out the four-layer model and its legal mechanics in full.
Print Article for this blog Download here