Every chat becomes a scored lead or a support ticket: an AI chat agent for a Spanish study-abroad agency
A Spanish study-abroad agency enrols international students onto Spanish language courses and supports them with accommodation, payments and visas. We built a website AI chat agent that answers prospects' questions around the clock, scores every prospect from 1 to 10, and sends current students' issues to the team as support tickets.
Snapshot
- Client type
- Study-abroad enrolment agency for Spanish language courses
- Country
- Spain, serving international students
- Industry
- Education
- System
- AI Chat Agent (client-facing)
- Use cases
- Customer support · lead qualification · lead capture and routing
- Stack
- n8n · Gemini File Search · Airtable
- Timeline
- Delivered
- Headline numbers
- 1–10 quality score on every prospect · 3 escalation labels · team reply by email within 24–48 hours
What did the client need?
Accurate answers around the clock, a way to spot serious prospects, and support requests kept out of the sales inbox.What did we build?
An n8n chat agent on the website, answering from a managed knowledge base and writing leads and tickets to Airtable.How does it run day to day?
Visitors chat on the site; leads and tickets land in two Airtable tables the team works through on a dashboard.
What did the client need?
Prospective students ask detailed questions whose answers depend on their circumstances. What they pay upfront, when they can start and which documents they need all change with their nationality, length of stay, course and accommodation. A non-EU student staying more than 90 days, for example, has to pay the full course fee upfront to get the acceptance letter their visa requires.
Answering by email was slow, and the inbox mixed new enquiries with requests from students who had already booked. The agency needed to answer prospects accurately at any hour, know which enquiries were serious, and stop current students' issues getting lost among sales conversations.
What did we build?
| Phase | What the agent does |
|---|---|
| 1. Greeting and contact capture | Asks for the visitor's full name and email before answering anything |
| 2. Classification | Asks whether they're a current student or looking to enrol |
| 3a. Current student | Collects the issue, logs a support ticket and tells the student the team will email within 24–48 hours |
| 3b. Prospective student | Answers from the knowledge base and gathers course type, start date, age, nationality and phone number in conversation, not as a form |
| 4. Information | Gives precise answers on courses, dates, prices, accommodation, payment terms, credits and visas, then recommends a course and start date |
| 5. Close | Shares the registration link and explains the booking steps, or logs the details of anyone not ready yet |
Three ways out of any hard question. The agent never guesses. Current students' problems, questions the knowledge base doesn't answer, and requests for exceptions to policy each become an Airtable record with one of three labels, and the student is told the team will reply by email.
Lead scoring. Every prospect gets a quality score from 1 to 10, based on signals such as a specific start date, detailed questions and raising visas or accommodation unprompted. The score, a summary and a recommended follow-up go into the lead record in a fixed format.
How does it run day to day?
Website widget ─► n8n AI agent ─► Gemini File Search (knowledge base)
└► Airtable: leads table · support tickets table ─► team dashboard- Managed retrieval. The knowledge base is one document in Gemini File Storage, queried by the agent at runtime, so there's no vector database to host. Updating it means replacing a file.
- Money terms up front. The agent explains deposits, booking fees and how visa refusals affect refunds before a student asks.
- Visa answers by nationality. It asks for nationality first, then explains the route that applies.
- Guardrails in the prompt. Answers come only from the knowledge base, it never discusses other students, applies an age requirement, and it's clear it speaks for an authorised agency, not the institution itself.
- English only. The agent runs in English.
What were the results?
We don't have lead volumes or conversion figures to publish yet. What the agent does on every conversation is measurable by design: it captures a name and email before answering, scores every prospect from 1 to 10, and routes every unanswerable question or current-student issue to the team with one of three labels.
- 1–10Quality score on every prospectBuild spec
- Escalation labels for the team's triageBuild spec
- 24–48 hTeam reply promised to escalated chatsAgency's service promise
What did we learn, and what are the limits?
- Separate support from sales at the start. One question, "current student or new?", keeps support requests out of the lead pipeline and stops the agent selling to people who've already paid.
- Give the agent a clear way out of every hard question. The three labels give the team an instant triage view, and show which questions the knowledge base should cover next.
- Clean up a policy-heavy knowledge base before launch. Payments, refunds and visas are where a wrong answer costs a student money, so the draft was corrected and checked during the build.
- Spell out the financial traps. How payment, visa and refund rules interact is where students get caught out, so the agent explains them proactively.