Every grant-maker facing an AI-driven surge in applications is, understandably, treating it as a volume problem. The numbers have grown. The quality has fallen. The team is under strain. The solution, logically, is a better process — a tighter triage, a two-stage system, a clearer set of criteria to filter signal from noise before the full assessment begins.

That is a reasonable response. It will probably help.

But it is not sufficient. And if it is the only response, it will produce a more efficient version of the same problem.

Here is why.

The 50p is not the price

There is a question I was asked repeatedly during fifteen years of running a crisis support charity in Bristol.

"Why would we invest time, land, and effort growing food locally when you can go down to Aldi and buy a lettuce for 50p?"

It is a fair question. On the surface, it is a very good one. The global food system has produced something genuinely remarkable: a vegetable grown thousands of miles away, transported across borders and continents, and placed on a supermarket shelf at a price so low it barely registers.

But the 50p is not the price of the lettuce.

It is the portion of the price that appears at the checkout.

The rest — the environmental cost of transportation, the soil degradation from intensive monoculture, the nutritional decline documented in peer-reviewed research, the labour conditions that make the price possible — has been successfully transferred elsewhere. To different budgets. Different people. Different timescales. The system has not reduced the cost of a lettuce. It has redistributed it to places where it is real and significant but invisible to the person making the purchase.

We see the price. We do not see the system.

The same logic runs through the AI application wave. The cheap application is not cheap. It is partially priced.

What the surge is actually revealing

When a grant-maker receives a surge of AI-generated applications — generic, misaligned, high in volume, low in signal — the visible cost is to the assessment team. Staff are overwhelmed. Turnaround times extend. Quality deteriorates. The success rate falls.

But the invisible cost is larger.

The organisations that most need funding — the ones doing genuinely aligned, deeply rooted, relationally grounded work in communities — are the least likely to have invested in AI-assisted application writing. They are often smaller, less well-resourced, led by people whose time is entirely consumed by the work itself. The surge does not just create more noise. It buries signal.

The grant-maker redesigning its triage to manage the volume is solving the problem it can see. The problem it cannot see — the organisations it is now systematically less likely to reach — does not appear in the performance data. It appears, later, in the communities that did not get funded. In the work that was not done. In the people who fell through a gap that tighter assessment criteria quietly widened.

This is cost transfer. The efficiency of the redesigned process has real value. But it is partial. And decisions made on the basis of partial accounting optimise the system in directions that make the hidden costs progressively worse.

The efficiency trap

This pattern — cost transfer masquerading as cost reduction — is one of the defining features of how modern systems fail.

A welfare system that cuts a benefit records a saving. The cost does not disappear. It migrates — to emergency housing, to the NHS, to the criminal justice system — at a multiple of the original saving. Systems stripped to minimum capacity function adequately under stable conditions and fail catastrophically when demand surges. Services optimised for throughput produce short-term recorded outcomes and long-term revolving doors.

In every case, the system appears to be working. In every case, what it is actually doing is moving the bill somewhere less visible.

The Rebuilding Lives model, which I designed and delivered over fifteen years in Bristol, makes this concrete. Sustained, relational, early support costs approximately £4,268 per person. Allowing a crisis to develop to the point of homelessness and responding to it costs in excess of £20,000 — and that figure does not include the compounding downstream costs of deteriorating health, family breakdown, and repeated system contact that typically follow.

The money did not disappear when the early intervention was not funded. It migrated to the most expensive possible destination, at the most damaged possible moment, attached to a life that has had significantly less chance to recover.

Efficiency without honest accounting of the full cost is not efficiency. It is cost transfer.

What a systems-aware response looks like

The grant-makers who will emerge strongest from this disruption are not the ones who design the cleverest triage. They are the ones who use the pause to ask a different question.

Not: how do we process this volume more efficiently?

But: what does this surge tell us about the system we have built — and who are we most likely to be missing?

That is a harder question. It requires sitting with complexity that the standard accountability framework is not designed to accommodate. It requires measurement that follows organisations and communities forward in time, rather than assessing them at the point of application. It requires a willingness to ask whether the organisations that write the most compelling applications are, in fact, the organisations doing the most compelling work.

It also requires something more specific: investing in the preconditions, not just the activities. The relational infrastructure that connects grant-makers to communities — the trusted intermediaries, the ecosystem knowledge, the long-term relationships that allow a programme officer to know what a good application looks like before it arrives — is not an administrative overhead. It is the mechanism that makes good funding decisions possible. Stripping it out in the name of efficiency is cutting the foundations while building the walls.

The organisations facing this challenge right now have a genuine opportunity. The pause — however unwelcome the circumstances that created it — is a moment to look at the architecture rather than just the process. To ask whether the system is optimised for finding the right answer, or only for managing the volume of questions.

The AI application is the 50p lettuce. It appears cheap and abundant. The real question is what it is costing — and who is paying.