Industrial Research Unit — Straits Institute for Applied AI
Applied AI statistics
Every count the Institute publishes about real-world AI deployment, in one place, updated on each weekly sweep. Free to cite with attribution and a link.
These are counts of applied AI — artificial intelligence put to work inside a job somebody already holds, in a named organisation, on a fixed date, with a published source. Nothing enters this record that fails those four conditions, which is why every number here is smaller than the figures usually quoted. The method sets out what is excluded and why.
For the analysis of what these numbers show — measurement, technique, build posture, direction of use and regulation timing — see the research note on applied AI.
Headline figures
Every figure below carries its own citation. Copy the one you need.
5,322 applied AI systems verified worldwide.
Across 249 countries, 17 industries and 311 occupational roles · swept weekly
11% of verified deployments carry a measured outcome.
594 entries filed as evidence of impact
66% are a system going live with no published result of any kind.
3,510 entries filed as deployment only
46% of applied AI systems use generative or language models.
Applied generative AI · 2,450 entries; the remaining 54% is vision, prediction and scoring
24% use computer vision.
Second-largest technique category, and the largest in several countries
22% of all verified deployments are in government and the public sector.
A fact about disclosure as much as deployment — see the limits below
21% are pointed at a person who did not choose to interact with them.
Enforcement and monitoring · facial recognition, risk scoring, behaviour flagging, biometric verification
41% are citizen-facing.
Systems the public chooses to use: assistants, chatbots, booking and enquiry services
32% of public-sector systems were built in-house.
39% are a global platform deployed largely as sold
Deployment precedes the first governing rule by a median of 22 months.
Measured where both dates are documented; some countries regulated first
328 rules governing AI use are logged in the record.
Filed as regulation, dated and sourced like every other entry
By kind of evidence
Each entry is filed as exactly one kind. Deployment means a system in service. Evidence of adoption attaches a usage figure. Evidence of impact attaches a measured change in time, cost, error rate or throughput. Regulation is a rule governing the practice.
Entries by kind of evidence
Table 1
| Kind of evidence | Entries | Share |
|---|---|---|
| Deployment | 3,510 | 66% |
| Evidence of adoption | 890 | 17% |
| Evidence of impact | 594 | 11% |
| Regulation | 328 | 6% |
By technique
Filed by the technique performing the task described in the source. Systems combining techniques are filed by the one the source names as primary.
Applied AI systems by technique
Table 2
| Technique | Entries | Share |
|---|---|---|
| Generative & language models | 2,450 | 46% |
| Computer vision | 1,290 | 24% |
| Prediction, scoring & risk | 880 | 17% |
| Speech & translation | 420 | 8% |
| Other ML & optimisation | 282 | 5% |
By industry
Government leads by a wide margin because public bodies publish tenders, answer legislatures and are covered by local press. A private deployment of equivalent scale is usually invisible until a vendor is permitted to name the client. This distribution is one of disclosed activity, not of activity.
Entries by industry
Table 3
| Industry | Entries | Share |
|---|---|---|
| Government & public sector | 1,180 | 22% |
| Healthcare & life sciences | 640 | 12% |
| Finance & insurance | 585 | 11% |
| Education & training | 430 | 8% |
| Transport & logistics | 360 | 7% |
| Retail & consumer | 330 | 6% |
| Legal & professional | 300 | 6% |
| Manufacturing & energy | 295 | 6% |
| Media & telecom | 230 | 4% |
| Agriculture & food | 175 | 3% |
| Technology & software | 165 | 3% |
| Business operations | 150 | 3% |
| HR & people | 140 | 3% |
| Marketing & creative | 120 | 2% |
| Sales & commercial | 95 | 2% |
| Communications | 70 | 1% |
| Strategy & leadership | 57 | 1% |
By region
Regional totals track press density, language coverage and disclosure norms as much as deployment volume. Sub-Saharan Africa is under-counted for the reasons set out under limits, not because deployment there is rare.
Entries by region
Table 4
| Region | Entries | Share |
|---|---|---|
| Asia-Pacific | 1,520 | 29% |
| Europe | 1,410 | 26% |
| North America | 830 | 16% |
| Latin America & Caribbean | 690 | 13% |
| Middle East & North Africa | 560 | 11% |
| Sub-Saharan Africa | 312 | 6% |
Country-level cuts for every tracked country:
By occupational role
The record holds 311 distinct roles. In no entry does a system take the role; it takes a slice of it. Radiologist entries concern triage order and first-read flagging, not diagnosis. Judge entries concern draft preparation, not verdicts. A single entry may name more than one role, so these totals exceed the entry count.
Most frequently touched roles
Table 5
| Role | Entries | |
|---|---|---|
| Civil servant | 214 | |
| Police officer | 168 | |
| Customer service agent | 151 | |
| Judge | 129 | |
| Tax officer | 118 | |
| Radiologist | 104 | |
| Teacher | 97 | |
| Citizen-services officer | 92 | |
| Nurse | 86 | |
| Customs officer | 74 | |
| Bank service agent | 71 | |
| Municipal officer | 63 |
By build posture
Who wrote the deployed system. Built in-house means the deploying organisation wrote it. Local vendor covers domestic suppliers and national models. Global platform covers Copilot, Gemini, Claude, ChatGPT, Agentforce and similar, deployed largely as sold. Where an organisation self-hosts or fine-tunes a global model, it counts as in-house only if the application layer was written by the organisation itself.
Build posture by region, share of entries
Table 6
By direction of use
Who the system faces. Citizen-facing systems are ones the public chooses to use. Internal administration is staff-only. Enforcement and monitoring systems have a subject rather than a user: the person they act on did not opt in. Mixed-use entries are filed by the function the source names first.
Direction of use by region, share of entries
Table 7
Regulation timing
Because the record dates both deployments and rules, the interval between a country’s first logged public-sector deployment and its first logged governing rule can be measured directly. Positive values mean the deployment came first.
Months between first deployment and first governing rule
Table 8
+38
+31
+27
+24
+19
+17
+14
+11
-3
-5
-8
-11
| Country | Months | |
|---|---|---|
| United States | +38 | deployment first |
| India | +31 | deployment first |
| Philippines | +27 | deployment first |
| Brazil | +24 | deployment first |
| United Arab Emirates | +19 | deployment first |
| South Korea | +17 | deployment first |
| Japan | +14 | deployment first |
| Singapore | +11 | deployment first |
| Chile | -3 | rule first |
| Canada | -5 | rule first |
| Finland | -8 | rule first |
| Netherlands | -11 | rule first |
Method, sources and limits
What is swept
National and local press, government and regulator publications, procurement notices, parliamentary records and official gazettes, in the working languages of each region. The sweep runs weekly and the record is amended continuously between sweeps when corrections arrive.
The admission test
A candidate enters the record only if four conditions hold at once.
- 1
A named organisationNot a sector, not an unnamed enterprise. If the body cannot be named, the claim cannot be checked.
- 2
A real jobSomebody currently holds the role the system touches. AI applied to work nobody was doing is a product, not an application.
- 3
A date“Will deploy”, “is piloting” and “has deployed” are three different claims and are filed differently.
- 4
A published sourceSomething a third party can open. The link is stored on the entry and shown with it.
What is excluded
- Vendor case studies. The organisation is named by the party that sold the system.
- Adoption surveys. A count of answers to a question, not of deployments.
- Strategy and roadmap announcements. Evidence about a budget cycle, not a workplace. Logged separately, never as deployments.
- Model capability results. A benchmark describes a laboratory.
- Funding rounds and partnerships. Capital and memoranda frequently precede nothing at all.
- Unattributed pilots. Fails the first condition whatever the trial is worth.
How entries are filed
Each admitted entry carries a kind, an industry, a country, one or more occupational roles, a date and a source URL. Three further tags are applied where the source supports them: technique, build posture and direction of use. Entries where a tag cannot be assigned from the source are left untagged rather than guessed, and are excluded from that tag’s percentages.
Limits
- Disclosure bias. The record measures what is published. Public bodies publish; private firms mostly do not. Comparisons across industries are comparisons of disclosure as much as of deployment.
- Language and press density. Countries with a dense local digital press are over-represented. Where deployments are reported in print, in minority languages, or not at all, they are under-counted. This is the main driver of the Sub-Saharan Africa total.
- Self-reported performance. Where an entry carries a result, that result is almost always the deploying body’s own. The record marks these rather than cleaning them. Inclusion implies no independent evaluation.
- Announcement inflation. Some governments announce the same system more than once. Duplicates are merged to the earliest dated source, but coverage in several languages occasionally survives de-duplication.
Corrections
Corrections are accepted from anyone and logged against the entry with the date and the nature of the change. If an entry misstates a deployment you are responsible for, write to the research unit and it will be amended. The correction log is public.
Reuse, data and citation
Every figure on this page is free to use, commercially or otherwise, with attribution and a link. Each chart carries a button that copies its underlying numbers as tab-separated values, and each headline figure copies as a complete sentence with its citation attached.
Attribution line
Download
Every verified entry with kind, industry, country, role, date and source URL. Download
By country, profession, industry or date range, usually within two working days. Contact the research unit.
What these numbers show, in five findings. Applied AI: the systems are running
Citing this page
Contact
Industrial Research Unit, Straits Institute for Applied AI — research@straitsai.institute. Journalist enquiries are answered the same working day.