What problem does it solve?
Every public service has residents who do not speak the official language well, and they tend to be the people who most need services: new arrivals, disaster survivors, people in crisis. Professional translation of documents is slow and costly, so agencies translate a few key pages and summarise the rest; interpreters are limited and take time to connect, especially for rarer languages; staff fall back on ad hoc translation, including free consumer apps, outside any governance.
Machine translation and speech models now make many languages workable in seconds, but a mistranslated address in an emergency call, a symptom at a clinic or a sentence in an asylum interview can change an outcome. The design question is where AI translation is enough and where a human must check it.
How does it work?
- Translate published content. Web pages, letters and guidance are machine translated with domain glossaries and style settings, then reviewed by a human for high impact texts.
- Translate what the public sends. Documents residents submit in other languages are translated in full, with the original kept alongside for the case file.
- Converse across languages. Chat and voice assistants detect the resident's language and answer in it, grounded in content in the official language.
- Interpret live. At counters, interview windows and on the phone, speech is transcribed, translated and voiced or shown on screen for both sides, with a transcript kept when enabled.
- Escalate to people. Where the law or the stakes require it, a certified interpreter or translator takes over, and staff can call one in at any moment.
- Audience
- Customer facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Web chat, Phone and voice, Kiosk and branch, Internal tools, API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | 20,000 to 891 million | 2 | 1 organization, 1 vendor |
Value drivers: Inclusion and access, Speed and cycle time, Lower cost to serve, Customer experience.
Indicative value
An agency that commissions translation of 20,000 incoming documents a year
USD 250,000 to USD 640,000
Document translation cost avoided per year
How this is calculated
Formula: documents * machineShare * (humanCost - reviewCost). The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Incoming documents translated per year documents, documents per year | 20,000 | 20,000 | The reference agency. Editorial assumption, replace with your own volume. |
| Cost of human translation per document humanCost, USD per document | 30 | 50 | FEMA puts its current cost at approximately USD 40 per document; the range brackets that figure. Source |
| Share of documents where machine translation plus light review is enough machineShare, fraction of documents | 0.5 | 0.8 | Editorial assumption; the rest still need full human translation. |
| Cost of light human review of a machine translation reviewCost, USD per document | 5 | 10 | Editorial assumption. Replace with your own review cost. |
What it leaves out: Covers incoming documents only. It leaves out faster case decisions, interpreter costs on calls and at counters, the value of wider language access and the cost of the translation service.
Who already uses it?
8 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Federal Emergency Management Agency
United States · Government and public sector · 2025
FEMA plans to translate the full text of non English documents that disaster survivors submit with their Individual Assistance applications, instead of relying on a contractor's summary of each document. The agency expects faster case processing and a drop in cost from about USD 40 per document to pennies. Original and translation will both be stored in the survivor's file, as substantiating documents that support assistance determinations.
No outcome disclosed.
U.S. Department of State (Bureau of Consular Affairs)
United States · Government and public sector · 2025
Consular Affairs is piloting Live Consular AI Language Augmentation (LCALA): real time transcription and neural machine translation of the spoken exchange at the visa interview window, with translated audio and on screen text and an optional short time stamped transcript. The visa interview pilot is flagged high impact in the federal inventory; its entry notes that interviews last about three minutes and that misunderstandings can force repeat questions, delays or uneven outcomes. A second pilot, for Overseas Citizens Services and American Citizens Services, supports calls and in person interactions and is described as assistive only, not a replacement for certified interpreters where they are required.
No outcome disclosed.
Internal Revenue Service
United States · Government and public sector · 2020
The IRS Linguistic Policy, Tools and Services team uses a cloud machine translation application on AWS, with the IRS Publication 850 glossary of English and Spanish tax terms, to translate text and files between English and Spanish, Chinese, Korean and Vietnamese and speed up responses to taxpayers. Separately, staff use SYSTRAN neural translation, augmented with a domain dictionary and translation memories, to triage non English documents for relevance to case work and as a starting point for manual translation. Both appear in the federal AI inventory as in operation.
No outcome disclosed.
European Commission
Europe · Government and public sector · 2017
eTranslation is the European Commission's neural machine translation service, launched in 2017, trained on EU professional translation data and offered free to eligible users such as public administrations. It translates text and documents, offers an EU formal or general style, and can translate websites through an API. The Commission says most of its language tools cover all 24 official EU languages and several more, including Arabic, Chinese and Ukrainian, and that eTranslation translated 891 million pages in 2025, up from 19 million in its first year. It now sits alongside related AI tools for summaries and draft replies.
- Interactions handled: 891 million, pages translated in 2025
"In 2025, eTranslation translated 891 million pages."
Claimed by: organization
Delaware County
United States · Government and public sector · 2025
Delaware County (Chief of Communications Anthony Mignogna) uses an automated, real time text to voice translation tool to communicate directly with Spanish speaking callers. The vendor reports that the county and other agencies have significantly reduced call processing time for Spanish calls; the detailed figures sit behind a download form and are not recorded here.
No outcome disclosed.
Madrid Destino
Spain · Government and public sector · 2024
Madrid Destino, the city's municipal tourism office, runs VisitMadridGPT, a virtual assistant that answers visitors in more than 95 languages from the city's official, expert curated tourism site, available when physical offices are closed. The city analyses the questions to find the most requested topics and adjust its website content. According to the story, Madrid attracted 10.6 million visitors in 2023.
No outcome disclosed.
Montgomery County Government
United States · Government and public sector · 2024
Montgomery County first launched Monty to relieve its 311 hotline during the pandemic, with 20 topics, and retired it when demand fell. Monty 2.0, built with Zammo.ai on Azure OpenAI Service and Azure AI Search, answers questions on more than 3,000 topics, with automatic translation into 140 languages, from the county's own knowledge base, and uses the county's geographic data to give address specific answers such as trash pickup days. It went through a seven month beta with a constituent focus group before the full launch in late 2024.
- Interactions handled: at least 20,000, since the beta deployment
"Since its beta deployment, Monty 2.0 has facilitated more than 20,000 constituent conversations, achieving a 50% customer satisfaction rate and reducing unanswered queries from 35%–45% to just 10%–15%."
Claimed by: vendor - Customer satisfaction: 50%, since the beta deployment
"Since its beta deployment, Monty 2.0 has facilitated more than 20,000 constituent conversations, achieving a 50% customer satisfaction rate and reducing unanswered queries from 35%–45% to just 10%–15%."
Claimed by: vendor
Baltimore City 911 (Emergency Communications)
United States · Government and public sector · 2022
Baltimore's emergency communications centre answers about 1.4 million 911 calls a year. Since partnering with Prepared in early 2022, call takers see a live transcript, AI summaries and highlighted key details on every call, which helps with addresses and callers who are hard to understand. Non English calls are transcribed and translated in real time, and for Spanish calls operators can dial in an automated voice translator instead of a third party interpreter. Automated QA now reviews every call, where a contractor previously reviewed roughly 30%. The case study also carries the unattributed line that the tool is "about 98% accurate", without saying what was measured or how, so it is not recorded as an accuracy figure.
- Quality score uplift: 12%, QA scores since automated QA was deployed; the QA method changed at the same time (sampling of about 30% of calls replaced by automated review of all calls), so before and after scores may not be comparable
"Since deploying Automated QA, Baltimore has seen a 12% improvement in QA scores."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Language demand by service and channel, to pick languages
- Domain glossaries and approved translations of key terms
- Rules on which document and conversation types need certified human translation
Systems to integrate
- Content management system and website (translation by API)
- Case management and document stores, to keep originals and translations together
- Telephony, counters and interview rooms for speech translation
- Contact centre and chat channels for multilingual assistants
Complexity: Medium
Machine translation itself is widely available, in some cases free to public administrations; the work is in glossaries, quality review for high impact texts, keeping originals with translations, and integrating speech translation into counters, phones and interview rooms without breaking legal rights to an interpreter.
- 1
Classify content and conversations by stakes
Decide which texts and conversations can use machine translation alone, which need human review, and which require a certified interpreter by law.
- 2
Use a governed service, not phones
Replace ad hoc free apps with an approved service with data protection terms, such as the European Commission's eTranslation for eligible administrations.
- 3
Build glossaries
Load approved translations of programme names and legal terms so the same concept is translated the same way everywhere.
- 4
Keep the original next to the translation
Store both in the case file so a reviewer can check a disputed phrase. FEMA plans to keep the original and the translation together in survivors' files as substantiating documents.
- 5
Pilot live interpretation with staff
Start with short, structured interactions (such as visa windows or Spanish 911 calls), keep transcripts and measure repeat questions and escalations to interpreters.
Guardrails
- Certified human interpretation where the law requires it, and always on request
- Original text or audio retained alongside every translation used in a decision
- Glossaries for programme names and legal terms, maintained by the service
- Data protection terms that keep public data out of model training
- Visible notice to the public that a translation is machine generated
KPIs to instrument
- Quality scores by language on a monthly human reviewed sample
- Time from document receipt to translated file
- Share of conversations escalated to a human interpreter, by language
- Translation cost per document and per call
- Complaints and corrections linked to translation
Human in the loop
Human translators review high impact published texts and any translation used in a decision; staff can call an interpreter at any point in a conversation. Language leads sample machine translations each month by language and correct the glossary.
Common failure modes
- Errors that change a case
- The Guardian reported machine translation errors that affected US asylum applications. Use certified interpreters for interviews that decide status, and keep transcripts.
- Rare languages quietly worse
- Translators quoted by The Guardian in 2023 said AI tools are particularly unreliable for less documented languages, and that major tools did not offer some languages at all. Measure quality by language and route rare languages to people.
- Shadow translation
- Staff paste case data into free consumer apps. Provide an approved tool and block the rest.
- Inconsistent terminology
- The same benefit gets different names on different pages. Maintain glossaries, as the IRS does with its Publication 850 glossary of English and Spanish tax terms.
What are the risks and rules?
EU AI Act
Depends on design
Assistants that talk with residents must tell people they are interacting with AI (Article 50(1)), and AI generated text published to inform the public on matters of public interest must be disclosed unless it has had human review under editorial responsibility (Article 50(4)). Internal translation that neither talks with people nor is published carries no specific obligation. Translation can also sit inside an Annex III process, such as examining asylum, visa or residence permit applications (point 7(c)) or evaluating emergency calls and dispatching emergency services (point 5(d)). Whether the translation component is itself high risk depends on its intended purpose (Article 6(3) exempts systems that only perform a narrow procedural task); either way it should be governed with that high risk process.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). Sets when people must be told they are interacting with AI, when providers must mark generated content, and when deployers must disclose AI generated text published to inform the public. Applies from 2 August 2026.
- Limited English Proficiency (notice on the suspension of lep.gov) (US Department of Justice, Civil Rights Division, North America). The Department of Justice has temporarily suspended lep.gov to implement Executive Order 14224, pending an internal review; its language access materials will be replaced when new guidance is issued. Check the current federal position before relying on older guidance.
- AI translation and language tools (European Commission, Europe). Describes eTranslation and related tools available free to eligible public administrations.
Controls to put in place
- Language access policy stating where machine translation is allowed
- Register entry for translation tools used in decisions
- Data processing agreement that excludes training on public data
- Monthly quality sampling by language
- Staff training on when to call a human interpreter
When it went wrong elsewhere
- Lost in AI translation: growing reliance on language apps jeopardizes some asylum applications. Reporting on US asylum cases harmed by AI translation, such as a city name translated literally in an application; volunteers describe applications denied after mistranslations.
Frequently asked questions
- Is machine translation good enough for public services?
- For information and routine conversations it can be, with glossaries and regular quality checks; the European Commission's eTranslation, free to eligible public administrations, translated 891 million pages in 2025. For decisions, treat it as a draft and keep the original: FEMA plans to translate survivors' documents in full and store the original and the translation together as substantiating documents in the survivor's file, and the State Department's citizen services pilot is assistive only, not a replacement for certified interpreters where they are required.
- Can AI interpret live at a counter or on the phone?
- It is being piloted and used. The State Department is piloting live interpretation at the visa interview window, and Baltimore 911 operators can dial in an automated Spanish voice translator instead of a third party interpreter.
- Which languages can assistants cover?
- Many. Montgomery County's Monty 2.0 answers in 140 languages and Madrid's visitor assistant in more than 95. Quality varies by language, so measure it for the languages your residents speak.
How to cite this page
Blits.ai AI Use Case Library, "AI translation and interpretation for multilingual public services", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/public-service-translation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published