Multilingual AI lead qualification calls: an energy subsidy build
A phone-based screening system that asks homeowners a fixed set of eligibility questions in several languages and routes the outcome.
The problem
An energy organisation needed to qualify homeowners for subsidy programmes over the phone. Eligibility had to be screened the same way every time before qualified prospects were passed to the next stage.
Challenges
Screening the same way every time
Eligibility for a subsidy depends on specific criteria about the property and the household. The system had to ask the same structured questions on every call, so two homeowners in the same situation reach the same outcome regardless of who or when they phoned.
Holding up in more than one language
A script that works in English can break down in another language through phrasing, pacing or misunderstood questions. The build had to keep the conversation reliable across languages without maintaining a separate set of eligibility rules for each one.
Turning answers into a clear outcome
A conversation produces free-form answers, but the business needs a decision: pass the lead on or conclude the call. The system had to read what the caller said, judge whether they qualify and route the result accordingly.
How the system works
Qualification script
The voice system asks a structured series of eligibility questions about the homeowner's property and circumstances, matched to the subsidy programme's requirements.
Cross-lingual build and testing
The core logic and conversational instructions were written in English. We then checked them in live multilingual conversations to confirm the dialogue held up in other languages.
Outcome routing
The system reads the caller's answers to decide whether they qualify. Qualified leads move forward, and the call is concluded for those who are not eligible.
Design decisions
A structured qualification script
The questions follow the subsidy programme's specific criteria in a fixed order. We preferred a structured script to open-ended chat because consistency matters more than variety when the outcome decides whether a household is passed to the next stage.
Logic written once, tested in live conversation
The core logic and prompts were written in English, then debugged through live conversations in another language. That kept a single source of eligibility logic while still checking, with real speech, that the dialogue behaves properly in each language.
Language-neutral prompt design
The qualification logic is written so it does not depend on any one language. Adding a language becomes a matter of validating the conversation, not rewriting the questions, and a small team can operate the system without a platform crew.
Result in production
A production voice agent that screens programme eligibility through conversation and works reliably across multiple languages.
Homeowners are screened for programme eligibility in a live phone conversation, without a person asking the questions on every call.
Eligibility is assessed consistently, so qualified prospects are passed on and others are concluded in the same way each time.
The same screening works reliably in more than one language.
What we would do differently
We would invest earlier in a proper eligibility rule engine rather than encoding the rules in the voice system's prompt. It makes audits easier and lets non-technical staff update criteria without a deployment.
Who this pattern fits
Any team facing similar volume, compliance or coordination demands could use this pattern. The technology stack can operate independently where needed and takes little operational effort to run.
Tools used
ElevenLabs
Speech synthesis for the voice agent, chosen for production-grade latency and audio quality.
Twilio
Telephony for inbound and outbound calls, SMS and call recording.
Multilingual prompt design
The language-neutral qualification logic that drives the questions in every language.
Related
Frequently asked questions
How does the system decide whether a homeowner qualifies?
It asks a structured set of eligibility questions about the property and circumstances, matched to the programme's criteria. From the answers it decides whether the caller qualifies, then passes qualified leads onward and concludes the call for those who are not eligible.
How was the system tested in another language?
The core logic and prompts were written in English, then debugged and validated through live conversations in a different language. Testing with real speech shows whether the dialogue holds up in practice, not only on paper.
Which languages can the system speak?
Any language ElevenLabs supports at production quality. The qualification framework is language-neutral and draws on one structured questionnaire, so adding a language does not mean rewriting the eligibility logic.