# Answers in two languages, mid-call.

> We gave operators approved answers in both languages during the call, owned and maintained by their own team.

- Source: https://synconai.com/case-studies/bilingual-ai-knowledge-for-contact-centres
- Publisher: SynconAI (https://synconai.com)
- Type: Case study
- Evidence: Delivered work
- Industry: Communications, Media & Technology
- Products: Custom AI, Vector search, Large language models, Web application
- Service: Custom AI Development (https://synconai.com/custom-ai-development)
- Reading time: 1 minutes

## In short

We gave operators approved answers in both languages during the call, owned and maintained by their own team.

## How we help

We gave operators approved answers in both languages during the call, owned and maintained by their own team.

## Before and after

| Area | Before | After |
| --- | --- | --- |
| Finding answers | Searching while the caller waits | Approved answers surfaced during the call |
| Languages | Quality depended on who answered | The same answer in both languages |
| Content | Hard to keep current | Maintained by the team itself |

## Results

- Shorter, calmer calls
- Consistent answers whoever picks up
- Gaps in content are visible and fixed

## KPIs we track

- Time saved per task
- Accuracy after review
- Adoption by the team

## How it works

Custom AI fills the gaps packaged features do not cover: speech-to-text, document and form extraction, retrieval over your own knowledge, and model pipelines that run outside the CRM and write results back to the record.

Production AI is mostly engineering around the model: data handling, evaluation, human review, monitoring and cost control. That is what separates a demo from a system people rely on every day.

## Our approach

1. **Define the task and the bar.** Agree what the AI must do and how accuracy will be measured.
2. **Prepare the data.** Collect representative examples, terminology and approved content.
3. **Build with review.** Design the pipeline with a person approving outputs where it matters.
4. **Evaluate and harden.** Test against real cases, tune for local language and edge cases, and add monitoring.
5. **Run and improve.** Operate it in production and improve accuracy from real usage.

## What to get right

- Keep a person in the loop where decisions carry risk.
- Measure accuracy on your own data, not a benchmark.
- Write results back to the system people already use.
- Plan for cost, monitoring and model updates from day one.

## Questions

### When should we build custom AI instead of using Agentforce?

Use Agentforce for work inside Salesforce. Build custom AI for specialised tasks such as speech, documents or heavy research, and connect the results to Salesforce.

### Can AI work in languages other than English?

Yes. We test and tune for the language and local terminology your users actually use.

### Can the AI run on our own infrastructure?

Yes. Self-hosted options keep data and cost under your control, with results written back to Salesforce.

---

SynconAI. https://synconai.com/case-studies/bilingual-ai-knowledge-for-contact-centres