Evidence from real operational work.
Evidence from controlled AI workflows and established operational deployments. Compare the task, the result and the basis for each claim.
Start with a bounded workflowSIRV AI · Customer sprint evidence
Two workflows tested against operational records.
The demonstrations addressed two different operational tasks. Findings, customer confirmation and remaining validation are distinguished below.
Unauthorised-access investigation
Access-control events were reconciled to surface cases for investigation. CCTV examination and final verification remained with a human reviewer.
Case-identification findings were demonstrated during the sprint. The time comparisons are estimates, not independently verified production benchmarks. The 20-minute estimate includes human CCTV review and verification.
Operational reporting and management assurance
The demonstration produced operational activity analysis and assessed the completeness of individual records against specified, provisional reporting-quality criteria.
The important distinction: beyond compiling a report, the workflow highlighted records with missing evidence and explained deficiencies for management review.
To prepare an example management report, followed by further time spent analysing the information. This is not a measured AI time saving.
- Activity reportingOperator activity, performance and category breakdowns
- Trends and exceptionsWeekly patterns and exceptions requiring attention
- Proactive assuranceVisibility of assurance activity undertaken
- Record-quality assessmentMissing supporting evidence and explained deficiencies against provisional criteria
Illustrative assessment sequence, not a reproduced customer record or an established operational feedback process.
The customer confirmed that the demonstration met the original requirements.
Reporting-quality criteria were provisional. Final criteria, management feedback practices and any deployment arrangements require agreement. Operational time savings and improvements in reporting quality have not yet been independently measured.
Results from established deployments
These examples demonstrate the operational foundations on which controlled AI can build. They span conventional reporting, real-time oversight and AI-supported intelligence, and are not all SIRV AI outcomes.
UKRI
DLR
Elizabeth line
UCB
Compare the operational work
Compare the operational problem, what changed and the evidence available. Sprint demonstrations and established deployments are distinguished. Outcomes are not directly interchangeable.
| Organisation | Operational problem | What was demonstrated or changed | Evidence | Relevant workflow |
|---|---|---|---|---|
| Anonymised AI sprint | Access-control records requiring investigation and CCTV verification | Reconciled event data, surfaced additional valid cases and structured human review | 20% more valid cases identified Processing: 6h to 20m estimated. Elapsed time: 30d to 20m estimated. Not verified production savings. | Operational intelligence and investigation |
| Anonymised AI sprint | Manual management reporting, activity analysis and incomplete operational records | Demonstrated activity breakdowns, weekly trends, exceptions, proactive assurance and identification of missing evidence against provisional criteria | Customer confirmed the demonstration met original requirements. Manual preparation baseline: 3+ hours, excluding further analysis. No independently measured AI time saving or quality improvement. | Operational intelligence; actions and follow-up |
| UKRI / STFC | High-volume operational records across two sites | Shared system of record, live visibility and progression towards governed AI | 50% reduction in reporting time | Operational intelligence; actions and learning |
| DLR | Manual and fragmented proof of activity | Real-time reporting, historical audit trail and cross-functional oversight | 1.5m+ location scans; 500,000 sheets of paper removed | Intake and routing; operational intelligence |
| Manchester Arndale | Fragmented security data and incomplete risk picture | Structured reporting, visualisation and AI workflow development | 50% faster reporting; 28% reduction in violent assaults | Operational intelligence; document evaluation |
| Elizabeth line | Fragmented internal and external threat information | Combined operational picture and AI-supported intelligence feed | Approximately 45 days faster reporting and earlier awareness of disruption | Operational intelligence and investigation |
| UCB | Fragmented travel, weather and local risk information | One operational view with AI-supported monitoring of external feeds | Estimated 5 to 10 minutes saved per employee each week | Operational intelligence |
| Aon | Manual reporting and compliance administration | Purpose-built workflow applications and standardised process | 50% reduction in report drafting time | Document evaluation; actions and follow-up |
Explore the individual case studies
Elizabeth line
Combining internal and external data to improve threat awareness, operational reporting and resource deployment.
Evidence: ~45-day reduction in reporting time and earlier operational awareness.
Read the case studyUKRI / STFC
Creating a dependable operational record across two sites, then using that foundation for governed AI workflows.
Evidence: 50% reduction in reporting time.
Read the case studyDLR
Replacing manual proof-of-presence processes with real-time reporting, searchable history and operational dashboards.
Evidence: 1.5m+ scans and 500,000 sheets of paper removed.
Read the case studyManchester Arndale
Moving from fragmented records toward structured reporting, visualisation and defined AI-supported workflows.
Evidence: 50% faster reporting and improved situational awareness.
Read the case studyUCB
Bringing travel, weather and local risk information into one operational view, with AI filtering external feeds for relevance.
Evidence: estimated 5–10 minutes saved per employee each week.
Read the case studyAon
Digitising defined internal workflows where off-the-shelf and internal options did not meet the operational requirement.
Evidence: 50% reduction in report drafting time.
Read the case studyWhy the earlier reporting cases still matter to SIRV AI
Operational AI is only as useful as the sources, workflow and review process around it. Several SIRV customers began by improving the operational record before adding intelligence or AI-supported work.
- Reliable source data: incidents, activity and decisions are captured consistently.
- Defined workflow: the operational task and expected output are explicit.
- Evidence and traceability: findings can be linked back to records and sources.
- Human judgement: consequential decisions remain with the responsible person.
“SIRV helped us move beyond basic reporting into a system that actively supports decision-making.”
Les O’Gorman, Director of Facilities, UCB
Have one operational workflow worth proving?
Start with a defined task, agreed sources, measurable outputs and clear review points. Test the workflow on real work before deciding how far to deploy it.
Run a SIRV AI trialNo generic AI programme required.
