Health Economics Metrics

Topic 57 of 76 · Software engineering and digital delivery

Cloud Unit Economics (FinOps)

Cloud unit economics translate raw cloud spend into cost per unit of output — per customer, per transaction, per case resolved, per token. It is the FinOps capability that turns "our AWS bill is £400k/month" into "serving one patient costs £0.83."

Why it matters

Total spend numbers can't answer the questions that matter: is the product getting more or less efficient? Does growth improve or destroy margin? What should we charge? Unit costs answer all three. For digital health specifically, "cost per case resolved" is a health-service unit cost — directly comparable to the National Cost Collection figures a commissioner uses for every other service, which makes it the natural language for pricing digital pathways against traditional ones.

The math

Unit cost = total allocated cost (incl. shared/platform costs) / units delivered

Two families:
  resource-efficiency units: cost/GB stored, cost/vCPU-hour, cost/token,
                             cost/build-minute
  business units:            cost/customer, cost/transaction, cost/consultation,
                             cost/case-resolved

Marginal vs average discipline applies (marginal-vs-average-cost.md):
committed/reserved spend makes marginal unit cost ≈ 0 until the next
commitment step — price expansion decisions at marginal, efficiency
trends at average.

Worked example

A digital triage service: cloud spend £62,000/month (compute £30k, data £18k, shared platform allocation £14k), handling 380,000 triage episodes/month:

Average cost per episode = 62,000 / 380,000 ≈ £0.163

Commissioner comparison: telephone triage ≈ £8–12/call, GP consultation ≈ £42
→ digital episode runs at ~2% of the cheapest human alternative — the
channel-shift economics of gds-service-metrics.md, from the cost side.

Trend check: last year £0.21/episode at 240k episodes → improving scale
economics (fixed platform costs amortizing), worth a headline in the QBR.

Software engineering connection

Unit economics is where engineering choices become finance-legible: an architecture that halves cost-per-episode is a pricing advantage; one that scales super-linearly is a time bomb visible only in this metric. Practices that transfer from health costing: publish the allocation rules (shared costs distorted per-unit figures until PLICS standardized patient-level costing — your platform-cost allocation needs the same rigor); pick units the buyer thinks in (commissioners buy episodes, not vCPUs); and feed unit costs into every ICER and budget impact model as the authoritative cost denominator. For AI features, the unit is the token — see inference unit economics.

Pitfalls

  • Ignoring shared costs: unit costs excluding platform/security/on-call allocations understate by 30–50% and collapse on audit.
  • Vanity denominators: "cost per API call" flatters; "cost per completed patient episode" informs.
  • Average-cost pricing of marginal decisions: charging teams average unit cost for usage that is marginally free drives waste-avoidance theater (see national tariff for the NHS version of this incentive bug).

Sources