What problem does it solve?
Banks that serve migrant workers, expatriates or tourists meet customers whose language no counter employee speaks. At Shinhan Vietnam Bank the counter staff are mostly local and many customers are Korean, and the need for language support at the counter had been raised repeatedly.
Korean banks have opened branches for foreign residents (Hana Bank in Pyeongtaek, IBK in Ansan). Hana's opens on Sundays for foreign workers who cannot visit on weekdays. A bank branch conversation contains product terms, identity checks and consents, so a wrong translation can change what the customer agrees to.
How does it work?
- Capture speech. A tablet or a device with a transparent display sits between the employee and the customer. Each person speaks in their own language.
- Recognize and translate. Speech recognition, tuned for the customer's language and for the bank's product names and financial terms, produces text that is translated into the other language.
- Show the result. The translation appears on the display for the other person, in one of up to 38 languages in the products described by the banks' vendor.
- Use stored phrases. Frequently used customer service phrases can be sent as a prepared multilingual message with one tap.
- Keep the human in charge. The employee performs the banking steps. A bank should define where a certified interpreter is still required.
- Audience
- Employee facing
- Autonomy
- Assist
- Adoption
- Emerging
- Channels
- Internal tools, Phone and voice, Kiosk and branch
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Inclusion and access, Customer experience, Speed and cycle time.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Shinhan Vietnam Bank
Vietnam · Banking · 2026
Flitto announced that it supplied Chat Translation Enterprise to six main branches of Shinhan Vietnam Bank in the Hanoi and Ho Chi Minh City areas. The system supports up to 38 languages and helps local Vietnamese staff talk with Korean customers at the counter. Financial terms were trained in advance and Vietnamese speech recognition was tuned for its tones. Branch tablets show a customer welcome page. No usage or outcome figure is published.
No outcome disclosed.
IBK Industrial Bank of Korea
South Korea · Banking · 2025
In May 2025 Flitto announced that it supplied Chat Translation Enterprise to IBK Industrial Bank of Korea's Ansan foreign financial center, a branch for foreign customers. Five counters can translate up to 38 languages in real time. The vendor says the system learned the bank's financial terms and product names to improve speech recognition by language. No usage or outcome figure is published.
No outcome disclosed.
Kwangju Bank
South Korea · Banking · 2025
In February 2025 Flitto announced that Kwangju Bank uses its Chat Translation Enterprise for AI based real time translation between counter staff and foreign customers, in 38 languages. Six branches had it installed at the time, and the bank planned eight branches with twelve counters, including a new foreign customer financial center and a mobile branch for foreign customers. Frequently used customer phrases can be sent as multilingual messages by a button in the app, and the speech recognition was tuned for the Jeonnam regional dialect. No usage or outcome figure is published.
No outcome disclosed.
Hana Bank
South Korea · Banking · 2024
Hana Bank opened the Pyeongtaek Foreign Center branch in September 2024, a branch only for foreign residents, with Flitto's AI interpretation product Chat Translation on four counters. Staff and customers each speak their own language and the conversation is translated in real time and shown on a transparent display, in up to 38 languages. The branch opens on Sundays for workers who cannot visit on weekdays. The bank's branch manager gave visitor numbers to the press. No accuracy, time saved or cost figure was published by Hana Bank.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A list of the languages and dialects that foreign customers use at each branch
- A glossary of product names, fees and financial terms with approved translations
- Rules on which conversations need a certified interpreter
Systems to integrate
- Counter devices (tablet, transparent display, headset)
- Speech recognition and translation service
- Optional storage of transcripts under the bank's retention policy
Complexity: Low
The build needs counter tablets or display devices with microphones, a speech translation service with a financial glossary per language, and a procedure for which transactions may rely on machine interpretation. No core banking integration is required for the interpretation itself.
- 1
Pick branches and languages from real demand
Start where foreign customer visits are concentrated, such as a branch near an industrial area, and choose the first languages from visit records rather than from the maximum a product supports.
- 2
Load the bank's terms
Give the vendor the product names, fee terms and form field names so that speech recognition and translation use the same words as the bank's documents.
- 3
Define what interpretation may and may not cover
Allow it for explanation and routine service. Require written confirmation in the customer's language, or a certified interpreter, for consents, credit terms and disputes.
- 4
Train counter staff
Teach staff to speak in short sentences, to ask the customer to repeat back key figures and to switch to a human interpreter when the customer looks unsure.
- 5
Review a sample each month
Have bilingual staff or a partner check a sample of conversations by language and feed corrections into the glossary.
Guardrails
- A human employee performs every banking step and the customer confirms terms on a bank channel
- Certified interpreter available on request and for consents and disputes
- Customer told that the translation is machine generated
- Conversation data kept only as long as the bank's policy allows and not used for vendor model training
KPIs to instrument
- Time to complete common requests such as opening an account, before and after
- Share of counter conversations that need a person to interpret, by language
- Translation quality score on a monthly bilingual sample
- Customer satisfaction of foreign customers by branch
Human in the loop
The counter employee is part of every conversation, checks that the customer has understood key figures and can call a certified interpreter at any time. Bilingual reviewers sample transcripts by language each month.
Common failure modes
- Product terms translated loosely
- A fee or an eligibility term is rendered in a way that changes its meaning. A bank glossary and a repeat back step for key figures reduce the risk.
- Dialect and accent errors
- Speech recognition misses regional or elderly speech. Kwangju Bank's vendor says it tuned recognition for the Jeonnam dialect for elderly customers, which suggests regional speech needs tuning.
- Customers treated as covered when they are not
- A language is supported on paper but with low quality. Measure quality per language and route weak languages to people.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A system that interacts with natural persons must tell them they are dealing with AI (Article 50(1)). Interpretation at a counter does not decide on credit or access to services, so it is not an Annex III use by itself. When the conversation concerns a loan, the credit assessment keeps its own Annex III duties; the interpretation tool does not evaluate creditworthiness, but its errors become a risk for that process.
Rules that apply
Controls to put in place
- Notice to customers that interpretation is machine generated
- Retention and access rules for transcripts and audio
- Language quality sampling with a named owner
- Approved list of conversation types that may use machine interpretation
Frequently asked questions
- Which banks run AI interpretation at the counter?
Flitto, the vendor, announced installations at Hana Bank's Pyeongtaek foreign customer branch (four counters), Kwangju Bank (six branches installed, eight branches and twelve counters planned), IBK Industrial Bank of Korea's Ansan foreign financial center (five counters) and six Shinhan Vietnam Bank branches. Most come from vendor announcements; Hana Bank's also comes from the bank's opening announcement and a reporter's visit. None has published usage results.
- Does it replace a certified interpreter?
No source claims that. It serves routine service conversations, and consents, credit terms and disputes should still go to a certified interpreter or a written confirmation in the customer's language.
- How does it differ from a digital human on a kiosk?
A digital human talks with the customer and answers from the bank's content. Counter interpretation connects a customer and an employee who share no language, and the employee does the banking work.
How to cite this page
Blits.ai AI Use Case Library, "AI live interpretation at bank branch counters", last verified 11 November 2026, https://www.blits.ai/ai-use-cases/branch-counter-ai-interpretation. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 11 November 2026: First published