Software can flag an empty field in an import document almost immediately. For a business, what matters is whether a complete file clears the first time and whether a correct submission is ever sent back by mistake. Import-document verification was among the applications Wonderful demonstrated to Azerbaijani public bodies on October 29. The digital ministry’s account describes the demonstration but gives no performance results on live applications.
A framework, a demonstration and a public service
The memorandum sets up work on a shared AI platform intended for the Government Cloud. Other demonstrations included checks on residence applications and voice support for myGov users. They are distinct tasks: the response time of a voice assistant says little about the accuracy of document checks.
The cost of an unnecessary return
Consider a correctly completed file that presents the same applicant details in two different formats. A mistaken rejection might take seconds for the agency to issue but cost the importer a working day to resubmit. That is an illustration of the risk, not a reported outcome of the demonstration.
A meaningful first test would select one document type and files of comparable complexity. A specialist and the system would review the files independently, and a second specialist would resolve disagreements. Errors could then be counted before any applicant was affected.
| Measure | What to count |
|---|---|
| Time to completion | From submission to acceptance, including returns; report the median and 90th percentile. |
| False returns | Correct files wrongly flagged as a share of all verified correct files. |
| Missed errors | Incorrect files accepted as a share of all files containing verified errors. |
| Cost per completed case | Cloud, integration, staff review and repeat processing per completed case. |
Assessing savings requires a baseline, a disclosed sample size and the same measures after the trial. A faster first response alongside more repeat submissions might leave the importer no better off. A shorter time to acceptance without an increase in false returns or missed errors would be a stronger result.