The Supply Chain Inside the Supply ChainDeep Dive, Chapter 3

English mental health care is best read as a supply chain rather than a service. A referral enters at the front end and is triaged against community capacity. If escalation is required, a bed or intensive pathway is matched against inpatient capacity. If a diagnosis lands, a medication is dispensed against pharmacy and wholesaler stock. Each stage has its own regulator, its own funding envelope and its own political deadline.

What none of the three stages has is a real-time matching layer. Triage happens against a local, static picture of capacity. Bed placement happens over the phone. Medication availability is resolved pharmacy by pharmacy, day by day, area by area. In supply-chain language, England has warehouses and hauliers but no order-management system — and it has been substituting human effort for that missing system for a decade.

That distinction matters commercially. A capacity problem is solved with capital and workforce, both of which are slow, politically contested and already committed. A matching problem is solved with software, data standards and integration — which is fast, cheap relative to the value released, and unclaimed. The three exhibits that follow each quantify a matching problem that has been consistently mis-diagnosed as a capacity problem.

England's mental health waiting list runs to roughly 1.7 million people. Of those, an estimated 735,000 hold an open referral for ADHD assessment specifically — a single diagnostic category accounting for close to half the entire backlog. That concentration is the tell: this is not diffuse demand across dozens of pathways, it is one dominant queue that a routing layer could address directly.

There is no single national triage or routing layer matching referral urgency to available community capacity. Individual Trusts build their own local models — Hertfordshire's CAPA model is one widely cited example — largely from scratch, each solving the same matching problem independently and each carrying the full build and maintenance cost alone.

The distribution, not the average, is where the damage sits. Median wait from referral to second contact for the quarter ending May 2026 is 37 days. The 90th-percentile wait on the same measure is 169 days: one in ten patients waits more than four times longer than the median. A median of 37 days sounds almost manageable. The 90th percentile is the number that describes what happens when there is no system deciding who gets seen next — only whichever local process a given Trust happened to build.

In 2016 the government set a national ambition to eliminate 'inappropriate' out-of-area placements — patients sent away from their local area for an inpatient bed purely because none was available locally — by March 2021. That deadline passed unmet. The target has since been reset to March 2027 under the NHS Medium Term Planning Framework, and £75 million of capital has been committed against it.

In the meantime, the absence of a real-time, system-wide bed-matching layer means capacity gets allocated by phone calls and relationships. The predictable consequence is concentration: a small number of large private providers are positioned to absorb the overflow, because they are the reliable answer when a Trust's local search fails.

The total annual cost of adult out-of-area placements ran to £164 million in 2023/24. Over half of that flows to two named private companies — Priory at £44.4 million and Cygnet at £42.8 million. One Trust's individual out-of-area placements previously cost £428,000 each; a single quality-improvement programme cut that Trust's placements from 110 to 1, demonstrating that the problem is solvable capacity-matching, not an unsolvable resourcing gap.

That last data point is the investment case in one line. If a single Trust-level improvement programme can take placements from 110 to 1, the binding constraint was never bed supply. It was visibility of bed supply.

NHS Right to Choose let patients get assessed by any qualified provider, cutting typical ADHD assessment waits from three-to-five years down to a few months in some areas. That is a genuine infrastructure win at the referral stage — a routing reform that worked because it created optionality where there had been a single queue.

But it also released a large, newly-diagnosed patient population straight into a medication supply chain with no real-time stock-visibility layer. Elvanse, Concerta XL, Medikinet XL and Ritalin have all faced persistent, worsening shortages since 2022. Pharmacists currently resolve this by phoning around and checking wholesaler stock by hand, day by day, area by area.

The queue did not disappear. It moved downstream — into a medication supply chain that has no equivalent infrastructure fix. And the same 735,000 open ADHD referrals from Exhibit A are the population about to hit this exact bottleneck, at scale, the moment their assessments clear.

Every other stage of this supply chain has at least attempted a system-level fix. Medication availability is still being solved one phone call at a time.

None of these three problems is a clinical problem, and none of them is purely a funding problem — the government has committed £75 million in capital for Exhibit B alone. They are matching problems: real-time demand on one side, real-time supply on the other, and nothing built to connect them at national scale.

Stage 1, referral and triage routing, is fragmented and rebuilt Trust by Trust, with 1.7 million on waiting lists and 735,000 on ADHD referrals alone. Its urgency driver is structural: mental health demand is projected to grow 4.4% a year, faster than any other part of the NHS. Stage 2, bed capacity and placement matching, has already missed one deadline and now carries a hard second one for March 2027, at £164 million a year with over half concentrated in two private providers. Stage 3, medication availability and verification, remains manual and phone-based, with 735,000 soon-to-be-diagnosed patients entering the same bottleneck.

Compute infrastructure gets built because it is capital-intensive and fundable in one large cheque. Chatbots get built because they are fast and cheap. A real-time matching layer across referral, capacity and medication in NHS mental health is neither — it is unglamorous, it crosses organisational boundaries, and no single Trust has the mandate or budget to build it alone. That is precisely the description of a vertical AI infrastructure opportunity nobody has claimed yet.

Value in a matching layer accrues to whoever holds the canonical, real-time picture of supply. In practice that means the party that integrates first across enough Trusts to make its view the default answer. Network effects are strong but geographically bounded: an ICB-level footprint is defensible, and a multi-ICB footprint becomes the de facto national layer without ever being procured as one.

The commercial model most likely to work is not per-seat software. It is a share of avoided cost — out-of-area placement spend avoided, locum and administrative hours avoided, breach penalties avoided — because the buyer already measures those numbers and reports them publicly.

Three things would break the thesis. First, NHS England builds the layer centrally and mandates it, collapsing the addressable market to integration services. Second, information-governance friction across Trust boundaries proves slower than the procurement cycle, so no vendor reaches critical density before budgets reset. Third, the March 2027 deadline is quietly moved a second time, removing the forcing function that turns a good idea into a funded programme. Of the three, the second is the most likely and the most survivable — it delays rather than eliminates.