Agentic procurement triage for an NHS trust
Routed and matched 73% of purchase requests end to end, escalating only genuine exceptions.
73%
Problem, approach, and the outcome
The client is an NHS trust with a central procurement team serving clinical and operational departments across a large hospital estate. Public-sector governance means every purchasing decision must be accountable and auditable.
The team was small relative to the volume of requests it handled, and any time clinical staff spent chasing orders was time taken away from patient care, a cost that goes well beyond the procurement budget.
The procurement team drowned in a flood of low-value, routine requests that buried the ones actually needing judgement. Most requests were straightforward, but they still had to pass through the same stretched team, creating a bottleneck.
Turnaround stretched to days, and clinical teams ended up chasing orders instead of treating patients, a poor use of scarce clinical time and a source of real frustration across the trust.
NHS governance meant nothing could be auto-approved without a clear, reviewable rationale, so speed could not come at the cost of accountability. Any automation had to be as defensible as a human decision, with its reasoning on the record.
We built an agent that classifies each incoming request, matches it against the catalogue and the relevant purchasing framework, and completes the routine ones end to end. The high-volume, low-complexity work (which is most of it) flows through without human touch.
Anything ambiguous or off-framework goes to a buyer with the agent's reasoning and evidence attached, so the human starts from an informed position rather than a blank form. Every decision, automated or escalated, carries a rationale that stands up to audit.
We phased it in by spend category, starting where the catalogue coverage was cleanest and the risk was lowest, then widening as confidence grew. That let the trust see the accountability trail working on safe categories before extending it.
- 73% of requests routed and matched end to end
- Every decision carries a reviewable rationale
- Average request turnaround down from 3 days to same-day
- Buyers refocused on high-value and off-framework spend
More AI & Machine Learning case studies

Autonomous reconciliation for a payments processor
Resolved 89% of reconciliation breaks without a human, every action logged for audit.
Read the full case study
Self-healing incident remediation for a manufacturer's OT platform
Cut median incident remediation from 41 to 7 minutes across a manufacturer's OT platform, with a human approval gate on every action.
Read the full case study
Underwriting copilot for a speciality insurer
Drafted first-pass quotes from submission packs 3.1x faster, every clause cited back to the wording it came from.
Read the full case studyGet a senior architect on the call, first time, every time.
No SDR gauntlet. 30 minutes with an engineer who can scope the problem, name the risks, and give you an honest feasibility call.
