Consider a routine situation in any large organisation. A customer service operation is under strain. There are three credible ways to close the gap: hire twelve more agents, deploy an artificial intelligence (AI) deflection layer that removes 40% of incoming contacts, or extend the existing outsourcing contract by a shift.
Any competent operator could argue for any of the three. They differ in cost, in timeline, in reversibility, and in what happens if demand halves in eighteen months. On paper, this is a single decision with three options.
In practice, it is never one decision. The twelve agents are a headcount request: they travel through HR and the people committee, on the annual planning cycle, justified in full-time equivalents. The AI layer is a technology investment: it travels through IT governance and architecture review, on the project cycle, justified as a business case with a payback period. The contract extension is a commercial variation: it travels through procurement, on the contract cycle, justified as unit cost against a benchmark.
Three routes, three governance bodies, three cadences, three business-case formats, three budget lines. Each route is individually well governed. The choice among them is governed by nobody.
Why technology leaders should care
It is tempting for a chief information officer (CIO) to read this as someone else’s problem, or even as an advantage: the AI business case that never has to compete against the labour alternative is easier to win. It is neither.
First, the asymmetry cuts both ways. In the quarter where the technology budget has no headroom and the establishment does, the organisation hires for a problem automation would have solved better. In the quarter where the reverse holds, it automates a problem twelve people would have solved better. Which option wins is determined by budget headroom and executive initiative, not by comparative economics. Both outcomes are failures; they are just invisible as failures, because no record exists of the comparison that was never made.
Second, the approved-in-a-vacuum AI portfolio accumulates. Every automation investment your governance approved was individually justified by a competent body applying a sound test. The estate they add up to, sitting alongside the workforce and the supplier contracts, was designed by no one. Enterprises are discovering this now, as they try to reconcile workforce plans, automation roadmaps and sourcing strategies produced through processes that may intersect, but were never designed to make the composition decision together.
Third, and most practically: the exit costs are never on the same page. Twelve employees, a licensed platform and a contract extension carry radically different reversibility profiles. Any one governance body sees one of them. No body sees all three, so the enterprise cannot see its own optionality.
The gap has a shape
Enterprise architects will recognise part of this territory. Capability-based planning, as documented in TOGAF (The Open Group Architecture Framework), treats capabilities as the unit of planning and delivery, and gives the discipline a vocabulary for describing the capability landscape and informing solution choices. What enterprise architecture (EA) has never been given is the mandate. An EA function can model a capability with real precision and inform the choice among delivery options, and then discover it holds no enterprise decision right to conclude that the answer is twelve people rather than a platform. The method exists. The decision right does not.
That is the accurate way to state the problem. It is not that nobody in the enterprise can compare hiring, automating and contracting. Transformation offices do it temporarily. Investment committees do it for whatever was routed to them as capital, which the headcount request never is. The comparison happens through temporary, partial or instrument-specific governance, never through a standing mandate. The fragmentation is not an accident of maturity. The three routes were built separately, decades apart, for good reasons, and no one has since been given the job of putting them back together.
This decision domain has a name: capability composition. The decision about which mix of capacity sources, employed people, machines, contracts and ecosystem partners, a capability should be built from. Naming it matters, because a decision without a name is very difficult to assign.
What to do while the governance catches up
- Run the audit question. Take the last significant capability gap your organisation closed. Ask who chose between hiring, automating and contracting, and whether the options were ever costed on the same page. If the answer is that nobody chose, you have located the gap in your own governance, and you have done it in one meeting.
- Put reversibility in the AI business case. If your automation investments are approved without the labour and sourcing alternatives beside them, you can at least stop approving them without exit costs. A payback period is not a reversibility profile.
- Ask where the comparison would live. Not who should own it, because that question is genuinely hard, and the obvious candidates each own one of the instruments, which is precisely the problem. But an enterprise that can answer where the three options would meet, on what cadence, in what format, is ahead of the large majority that cannot.
This blog will return to the governance of capability composition, across human, machine, contractual and ecosystem capacity, as that research develops. For now, the one-meeting audit above is the place to start.
The uncomfortable discovery may not be that your organisation chose the wrong answer. It may be that nobody can show where the choice between the answers was ever made.
Print Article for this blog Download here
