Evidence · from trial to routine use
How twelve AI programmes spread, or didn’t.
Twelve inspection and asset-operation programmes, from trials to wider use, stops and missing follow-up. For each: what happened when, how far it spread, who confirms it and what is published about its value.
This is a selection to learn from, not an industry sample. It shows the kind of analysis I can do for your own questions.
Four views of twelve programmes
The filter applies to charts 1, 2 and 4. Reviewed 3 November 2026. Select a programme or cell for its sources and limits.
01 / Adoption journeys
What happened, and when.
Each row sets out the dated steps found for one programme. Gaps are shown as unknown: two dated accounts do not prove continuous use in between.
On smaller screens, scroll the diagram horizontally. Every programme opens its evidence record.
Selection and date method
These 12 programmes were selected to examine inspection and asset-operation scaling, with trial-only, routine, wider-use and adverse examples. They are not a cohort recruited at pilot start. No conversion percentage or average time to scale is calculated. “By” marks a source-bound observation, not an inferred start date. FY 2024–25 is a programme range. A publication date, filename or current web access is not an implementation event.
Case strength is shown as context. Strong can support a bounded pilot finding or an adverse outcome; it does not mean scaled or financially successful.
02 / Reach and follow-up
How far did it spread, and who confirms it?
Left to right: how widely the programme is reported to be used. Top to bottom: what later evidence confirms it. Select a programme for its details.
What reach and follow-up establish
“Wider” means reported expansion across a portfolio, network or organisational service, with installation, cumulative inspections and rollout launches identified in each point’s detail. “Routine” concerns the stated workflow, not the whole estate. The stopped/unaccepted row sits under trial because accepted operation was not achieved. Empty cells mean no example from this selection occupies that position.
A later operating account supports another observation of use; it does not prove continuity or benefit. Procurement supports commitment. Contract years are not observed operating years. Numeric footprints with different units are never added, and coverage percentages require matching active-use and eligible-estate definitions. Historical and undated accounts remain visible.
03 / Who uses the same product
Several examples of one product are not always several customers.
Eight products appear in 36 examples, used by 34 different organisations. Many of these accounts come from the suppliers themselves.
Grouping, dependence and source ownership
This is a non-exhaustive selection from the 522 records, not a supplier customer census. Bars count selected case records and distinct named operating adopters within each family. Buildings, contractors, projects, reports and trial phases do not automatically add adopters. Counts across families cannot be added as unique organisations.
Open each family for every adopter, scoped stage, case strength, source basis and nesting notes. Most accounts in several families share supplier ownership; named adopter testimony and procurement corroboration are not independent effectiveness evaluations. No independent replication total or pooled benefit is calculated.
StormHarvester combines five blockage/alerting and two inflow/infiltration adopters; DocNostic excludes SiteLens. VAPAR, FIDO and Buildots cover related tasks with different configurations. These family counts are not repetitions of identical interventions. Thames Water’s FIDO timing is classified here as a four-week trial, matching the scoped source; the broader case record is unchanged.
04 / Evidence of value
What is published about the value?
For each programme: is there evidence of use, quality, benefit, checking effort, full cost and lasting value? Most cells are N: not published. Select a cell for its detail.
N is a disclosure gap, not failure. Adverse findings keep their original scope: Victoria’s audited cost is government venture expenditure, not full technology cost.
Cell definitions and evidential limits
M requires a published comparator, scope and identifiable measurement method; it need not be independent or causal. No positive benefit in this selected review meets that threshold. R includes attributed use, a reported comparison or a documented operating-policy change with incomplete evaluation. E is an estimate or forecast. N means not disclosed in checked material, not failure. A retains the scope of an adverse finding or limitation.
These are different evidence categories, not a score. Do not add R and M, count green cells as proof, or infer the rest of the collection from this selection. Cost disclosure distinguishes a contract or government venture cost from complete implementation and operating cost. Victoria’s audited public venture cost is not full technology TCO. Case strength does not independently validate individual figures. Checking effort, full costs and sustained net value are separate tests.
Your question
Want this for your own work?
I can follow the programmes closest to your assets or workflows in the same way, using the full collection and its sources.