The hours were never the product

What the paper argues

A consultancy was never bought for the hours. It was bought to carry risk a client could not carry alone, and for pattern knowledge built across many clients. The billable hour was a unit of pricing, a proxy for those things. Over a year, Accenture has lost more than half its market value and Capgemini is down around 40 per cent, yet revenue at the same firms has in several cases still grown. As with offshore, fixed-price and managed services before it, AI reprices the unit and leaves the job intact: it compresses the production work that filled the hours, but it does not carry accountability for what gets shipped.

A more capable model raises the floor of plausible-looking output, and plausible-looking output is precisely what an organisation without governance cannot challenge. The competitive question is now whether a firm can govern AI-assisted delivery at the point where decisions actually get made, rather than document that it intends to. A governance programme is not a precondition completed before developers begin. It is instrumented around work already happening, and none of the six steps the paper sets out requires stopping or buying anything.

Key points

• Make one decision reconstructable: for a single change that shipped with AI assistance last week, answer from existing records what was asked of the model, what it produced, and who decided it was fit to ship (Provenance Integrity).
• Name who is answerable for what shipped, not who is most skilled with the tool. If the answer is “the model” or “the team, collectively”, that is a supervisory gap (Supervisory Clarity).
• Move one constraint to where the decision is made. A scope that sits outside the runtime context does not constrain (Scope Definition).
• Pick the highest-volume pipeline and ask what review still means there. A signature is not engagement (Review Substance).
• Try to ship something that should violate a control you believe is in force. If it ships anyway, the control is documented, not operating (Constraint Fidelity).
• Govern first where exposure is highest, not where capability is most exciting (Capability Calibration).