AI's Hidden Bill: Why Half of UK Consulting Clients Are Pulling Back on Deployments

The bill has landed. After three years of confident AI announcements from boardrooms across the UK, the Q2 2026 edition of KPMG's Global AI Pulse, drawn from 2,145 senior leaders across 20 countries, has found that 42% of corporate leaders regard the costs of operating AI as 'largely invisible'. The consequences are no longer theoretical.

The Usage-Billing Shock

Earlier in 2026, providers including Anthropic and OpenAI shifted core services away from flat-rate subscriptions toward usage-based billing. For many consulting clients and their advisers, this was the moment the maths stopped working. Firms that had budgeted annually for AI research tools found themselves burning through that allocation in weeks. KPMG found that a third of senior corporate leaders identified limited understanding of AI costs as a direct barrier to deploying AI agents, and that nearly half of organisations had rephased deployments when costs outweighed expected value.

This is not a niche edge case. It is the dominant story in professional services right now.

The Utilisation Problem Nobody Wants to Name

The SPI 2026 Benchmark data reported by Consultancy.uk gives consulting practices an uncomfortable backdrop: billable utilisation in professional services is sitting at an all-time low of 66.4%, with client net-promoter scores sliding 12% in a single year and EBITDA compressed to around 9.8% compared to a high above 15% in 2023. Practices that sold AI transformation engagements on a promise of ROI now face clients who cannot demonstrate that return.

The numbers from KPMG show why. Organisations with strong cost visibility are four times more likely to report established ROI than those without (25% versus 6%). Yet only 24% of leaders say their CEO carries ultimate accountability for AI-driven business outcomes. When accountability is distributed across a C-suite, it effectively belongs to no one, and measurement frameworks are the first casualty.

PwC's own Global AI Jobs Barometer, published this month, captures the productivity paradox neatly: productivity growth is 40% higher at companies most exposed to AI versus least exposed, but the gains are concentrated in a minority. The firms not capturing that uplift are not failing on technology. They are failing on governance, sequencing, and measurement discipline.

What the Formation Data Reveals

AIBD analysis of Companies House data shows 2,032 new SIC 70.22 (management consulting) incorporations in Q3 2026, down 84.1% on the prior period, a formation cohort collapse suggesting the easy-money wave of AI consultancy startups may be cresting. UK IPO trademark data points the same direction: Class 35 filings (business services) reached just 2,405 in Q3 2026, down 77.8% versus the prior period. Source: AIBD analysis of Companies House and IPO data, as of July 2026.

The structural gap underneath both numbers is striking. Fully 98.2% of active SIC 70.22 companies hold no Class 35 trademark. That is not a compliance curiosity. It reflects how thin the brand investment is in a cohort where most entrants were pitching AI advisory services built on commodity tools, not proprietary methodology.

Boutique consultancies built on pass-through API access are the most exposed. When the underlying cost base shifts with usage-based billing, margin assumptions built on flat-rate subscriptions evaporate. The Consultancy.uk AI agent talent piece published yesterday noted that AI Engineer freelancer communities grew 229% in the past year on the Malt platform. Human expertise is being pulled back in precisely because the agentic systems that were supposed to replace it have not yet delivered at scale.

The Accountability Trap

There is a pattern here that consulting practices need to confront before their clients do. The MCA's member survey found 77% of UK consulting firms have integrated AI into their systems or enabled employees to use AI models, with 68% increasing automation. The Management Consultancies Association figure on integration is not the same as a figure on measured value delivered to clients.

KPMG's data shows organisations where the CEO is explicitly accountable for AI decisions are significantly more likely to realise meaningful business value (57% versus 21%) and nearly four times more likely to report established ROI. The lesson for consultancy practices is identical: accountability has to sit at the top and be attached to financial outcomes, not activity metrics.

The ICO's March 2026 report on automated decision-making in recruitment, drawing on evidence from over 30 UK employers, found a related failure mode: tools described as decision support were, in practice, making substantive decisions, with human review amounting to rubber-stamping. The same dynamic is playing out in consulting delivery. AI is being described as augmenting the professional, but often it is replacing the diligence step that the client thought they were buying.

The Divergence Ahead

So. The firms that capture the productivity gains PwC documents are those using AI to amplify human performance and enter new markets, not simply to cut headcount and compress timelines. The firms that will struggle are those that treated AI as a billing mechanism rather than a delivery architecture.

The usage-billing shock of 2026 is not a reason to retreat from AI investment. It is a reason to finally build the measurement frameworks that should have been there from the start. Clients who cannot see where costs accumulate will not sign the next statement of work. That is the only ROI metric that matters right now.