UK AI Investment Hits £20.6m Average as ROI Forecast Reaches 39%

Why the benchmark matters for London boards

Boardrooms across the capital are moving past the question of whether to adopt AI and are now asking how much is enough. The SAP figure represents a cross-sector average, meaning regulated industries such as financial services and healthcare may sit above it while retail or professional services could sit below. Knowing the average allows a CTO to frame a budget request as a strategic decision rather than a speculative bet.

It also changes the conversation with procurement and finance. When the expected return is quantified at 39 percent over a defined horizon, the internal rate of return calculation becomes easier to defend. That clarity helps avoid the stall that happens when AI spend sits in an innovation pot without a clear owner or payback timeline.

Where the spend actually goes

The £20.6 million average covers infrastructure, model licensing, data preparation, governance tooling and the talent required to operate it all. A typical London enterprise might allocate a third to cloud compute and GPU access, a third to data engineering and platform build, and the remainder to specialist headcount and compliance work. The split varies by maturity: earlier-stage programmes spend more on foundations, while mature teams shift budget toward use-case development and optimisation.

Understanding this composition prevents the common error of approving a large model licence without budgeting for the data pipelines that make it useful. It also highlights why governance is not optional — regulators and insurers increasingly expect documented model risk management as a condition of cover.

What to do this week

1. Request the current AI spend breakdown from finance and map it against the infrastructure, data, talent and governance categories.
2. Identify the top three use cases with defined success metrics and a path to production within six months.
3. Schedule a 30-minute session with the CFO to walk through the 39 percent ROI benchmark and agree a shared reporting cadence.
4. Audit existing vendor contracts for auto-renewal clauses on GPU capacity or model APIs that could inflate costs without approval.
5. Assign a single owner for AI governance reporting so the next board pack includes a compliance status update.

Measuring progress against the 2028 horizon

CTOs should treat the 39 percent figure as a directional target, not a guarantee. The path to that return depends on moving workloads from proof of concept into operational workflows where revenue uplift or cost reduction is tracked monthly. That means instrumenting every agent and automation with logging that feeds a dashboard owned by finance, not just engineering.

Quarterly reviews should compare actual contribution against the forecast trajectory. If a workstream is flat after two quarters, the default action is to pause and re-scope rather than extend funding. This discipline separates the organisations that hit the benchmark from those that add to the average spend without the average return.

The average investment figure is a signal, not a target. The organisations that reach the projected return are the ones that treat AI as a portfolio of measured bets rather than a single transformation programme.