1001 Evidence Brief
🕰️ How long does a smart-city project run before it stops?
Longer than the word failure suggests, and the failures run longest of all. Of the 95 entries here that have finished, the median ran 4 years; the ones recorded as failed ran a median 5.5 years, against 4 for the ones recorded as succeeded. Nothing collapses — it is kept running until somebody stops paying for it.
✔ Every published entry is checked against the public sources it cites before publication.
How long they ran, by how they ended
under 3 years3 to 5 years6 to 9 years10 years or more
| Outcome | Entries | Years it ran | Median | Ran 10 years or more |
|---|---|---|---|---|
| Every finished entry | 95 | 26 under 3 years, 39 3 to 5 years, 24 6 to 9 years, 6 10 years or more | 4 years | 6% |
| succeeded | 41 | 13 under 3 years, 21 3 to 5 years, 5 6 to 9 years, 2 10 years or more | 4 years | 5% |
| scaled back | 28 | 8 under 3 years, 10 3 to 5 years, 8 6 to 9 years, 2 10 years or more | 5 years | 7% |
| failed | 26 | 5 under 3 years, 8 3 to 5 years, 11 6 to 9 years, 2 10 years or more | 5.5 years | 8% |
The bands are unequal on purpose and the smallest one carries the point: only 6% of finished entries ran ten years or more, while 27% were over inside three. A short life is the norm here; a long one is what a project earns by being useful — or by being hard to switch off.
Representative entries
- ShotSpotter Acoustic Gunshot Detection Twelve years, and it did not end because the technology stopped working. The city's Inspector General examined it and the contract was not renewed — the longest-running failure in this record ended on a report, not a fault.
- City Data Exchange Two years. A municipal data marketplace with a named operator, launched, and closed for want of both sellers and buyers — the shape people picture when they hear that a smart-city project failed, and the rarer shape in this record.
- RTA's Autonomous Pod & Shuttle Trials Start year and end year the same. It is in this atlas because a trial that was announced and then never became anything is a documented outcome, and leaving it out would make every other number here look better than it is.
Method
An entry counts here when it states a start year, an end year, and an outcome that is terminal — succeeded, scaled back or failed. That is 95 of the 1216 published entries. Its life is the difference between the two years, so a project that ran from 2012 to 2024 counts as twelve.
Entries still running are excluded, because their length is not yet a fact. Three entries carry an end year while their outcome is still open — a planned completion date rather than a recorded ending — and are excluded for the same reason.
The overall failing share of the atlas is 120 entries, 9.9%. That figure mixes vintages of very different maturity; among the 824 entries that started in 2020.5 or earlier it is 12.0%, and the arithmetic behind that cutoff is on the page about how the record ages.
What this cannot tell you
- Years, not dates. The atlas records the year a system started and the year it stopped, so every length here is accurate to about a year in either direction.
- This is a floor on failure, not a rate. An entry only earns a failing outcome when a named source says so, and a project abandoned quietly rarely has one. The projects that vanished without a press release are not in this atlas to be counted.
- Survivorship runs the other way for old entries: something that died in 2006 seldom left a source behind, so the earliest years of this record are thinner in failures than the world was.
- A median over 95 entries is a small number of projects. The bands beside it are there so the spread is visible rather than hidden behind one figure.
- "Finished" is not the same as "switched off". Some entries here ended because the programme concluded as designed; the outcome column, not the end year, says which.
The slice behind every number
Every figure above is computed at build time from the published data, and the rows it counts are downloadable as one file: how-long-a-smart-city-project-runs.json. Recomputing the numbers is a matter of counting that file. The full datasets are on the open data page, under CC BY 4.0.
Where to read next
How the record ages puts these lengths against the calendar and derives the maturity cutoff from them. What didn't work groups the failing entries by how they went wrong. And the brief on who evaluates smart-city projects asks who, if anyone, was watching while they ran.
All briefs · Who publishes the evidence · How the record ages · The state of the evidence