Picture a finance director I’ll call Helen. £6m turnover services firm, twelve months into a Copilot rollout that touches sales, operations, and admin. Her latest hours-saved figure is 600 staff hours in the quarter, validated by a proper time-study. The number is real. She looks at the gross margin line for the same quarter and finds it unchanged. The partners are now arguing about whether the AI is working at all. Helen does not have the language for what is actually happening.
What is happening is value leakage. Hours-saved is a productivity metric. Margin is a financial metric. The two diverge through a structural pattern that has been studied for decades, and the gap is rarely a failure of the technology.
What does the productivity paradox mean for AI?
The McKinsey, Goldman Sachs, and BCG body of work on the productivity paradox identifies a consistent finding: measured improvements in labour productivity do not automatically convert to bottom-line improvement. The hours that AI saves do not necessarily show up as margin expansion or cost reduction. The gap exists because the productivity gain is one thing and the financial outcome is another, with several pathways between them.
This applies to AI specifically because AI productivity gains are real and measurable, but the conversion to margin requires deliberate management of where the freed-up time goes. Without that management, the time is absorbed before it ever reaches the bottom line.
Goldman Sachs has reframed this point precisely: technology creates productivity potential; management strategy determines whether that potential becomes financial reality. If hours-saved does not expand margin, the AI is doing its job. The firm has not done theirs on reallocation.
What are the five pathways value leaks through?
The first pathway is client price reduction. If an AI tool allows the firm to produce work 30 percent faster, clients sometimes negotiate a 20 percent reduction in fees, capturing a substantial portion of the productivity benefit. Costs go down 30 percent, revenue goes down 20 percent, net margin expansion is 10 percent. This is rational client behaviour, not a defect.
The second is work expansion. The freed-up time gets used on more complex work, work that was previously skipped, or work that internal stakeholders newly prioritise. Output volume rises rather than costs falling. The team feels busier, not lighter.
The third is idle capacity. The freed-up time results in spare capacity the firm does not have new work to fill. Professionals work fewer billable hours while keeping the same salary. Margin neither expands nor contracts; the time just disappears.
The fourth is process redesign overhead. Implementation requires workflow changes, training time, and management overhead. These costs consume part of the theoretical time savings, particularly in the first six to nine months.
The fifth is quality management and rework. AI output that requires validation, review, or occasional correction absorbs part of the speed gain. The hours that remain are smaller than the headline figure suggests.
Where does the saved time actually go in services firms?
The empirical pattern from post-implementation analyses of services-business technology adoption is consistent. Approximately 40 to 50 percent of freed-up time results in cost reduction or margin expansion. Approximately 30 to 40 percent goes to work expansion or service improvement. Approximately 10 to 20 percent is absorbed into slack or transition costs. These percentages move depending on industry, technology, and how aggressively the firm manages reallocation.
Regulated professional services see this most acutely. A law firm, accountancy, or clinical practice cannot simply reduce staff when hours are freed. Equity arrangements, retention concerns, and capacity buffers for client demand all push back against translating time into headcount cost reduction. The freed time more typically shows up in higher utilisation, more advisory work, or quieter weeks for senior staff.
This is why a £6m firm can run a clean time-study that says 600 hours saved last quarter and find no margin movement at the same time. The hours moved. They just moved into work expansion and slack rather than into the cost line.
A firm that aggressively manages reallocation can push the cost-reduction proportion to 60 or 70 percent. A firm that lets it happen passively typically sees the reverse, 20 or 30 percent reaching margin.
What is the pre-decision question that prevents most leakage?
The single question that prevents most leakage is asked before the AI is bought, not after it is deployed. Where will the freed-up time go? The answer needs to be specific. Named clients, named services, named work that the freed capacity will fund. Without that specificity, the leakage is structural rather than corrective.
If the firm has identified specific new initiatives, new clients, or new services that the freed capacity will support, that should be articulated at proposal stage with a financial projection. The reallocation strategy is part of the business case, not a problem to be solved afterwards. The CFO can read the case and see how the productivity gain converts to revenue or margin.
If the firm has not pre-planned the deployment of freed time, it should not expect to see margin expansion. It should expect the freed-up time to be absorbed gradually into slack or work expansion, which means the ROI shows up in service quality or client retention rather than the cost line. That is still a real benefit; it is just a different benefit, and it should be modelled as such.
The discipline is unglamorous and unfamiliar, which is why most SMEs do not do it. The firms that do it consistently see substantially better conversion of productivity gain into margin than firms that do not.
Should you model both scenarios in the proposal?
The honest version of an AI proposal models both reallocation outcomes side by side. Scenario A: if the freed-up capacity goes to new clients or services, ROI is X. Scenario B: if it gets absorbed into work expansion or quality improvement, ROI is Y, where Y is typically lower or deferred.
The CFO sees the range, not a single optimistic number. The board can then decide which scenario the firm is committing to and what reallocation discipline is required to land closer to A than B. The conversation moves from “will this work?” to “which of these outcomes do we want, and what does it take to get there?”
That is a different proposal-stage conversation, and it is a better one. It also flushes out the firms that have not done the reallocation thinking. If the consultant cannot describe how the freed-up time will turn into financial outcome, neither version of the proposal is grounded.
If you are looking at a quarter where the time savings are real but the margin has not moved, the leakage is the diagnostic. The fix is a deliberate decision about where the freed-up time goes, made before the next deployment rather than after. If you’d like to talk through what the reallocation strategy looks like for your firm, book a conversation.
