Every enquiry called in under 2 minutes: a speed-to-lead AI voice agent for a Spanish sports-education agency
A Spanish sports-education agency sells football camps and programmes to parents of young players around the world. We built an outbound AI voice agent that calls every new camp enquiry within 2 minutes, confirms the details, flags better-fit programmes and hands a scored lead, with a next-step label, to a human consultant.
- Form submitted
- Filters
- Agent calls
- Odoo updated
Snapshot
- Client type
- Sports-education agency: football camps and programmes
- Country
- Spain, serving families worldwide
- Industry
- Education & coaching
- System
- AI Voice Agent (outbound)
- Use cases
- Speed-to-lead · lead qualification
- Stack
- VAPI · Twilio · ElevenLabs · n8n · Airtable · Odoo (migrated from HubSpot mid-project)
- Timeline
- First project of a monthly retainer, from February 2026
- Headline numbers
- < 2 min from form to call · 1–10 lead score · 6 next-step labels · calls kept under 8 minutes
What did the client need?
Qualify enquiries properly, cut the noise for consultants, and steer suitable players towards the agency's own programmes.What did we build?
An outbound agent that never sells: it verifies, checks fit, raises budget only when needed and hands off to a person.How does it run day to day?
Every form that passes the filters gets a call within 2 minutes; misses are retried once, then handed to sales.
What did the client need?
The agency's sales team of sports consultants struggled to qualify leads properly in the CRM, and found it hard to interest families in programmes they hadn't asked about, such as the agency's own high-performance camps. Consultants spent hours on noise: players contacting them directly instead of parents, budgets too low for the programme and mismatched programme choices.
The goal was to steer suitable players towards the agency's own programmes without losing families set on a particular camp, and to call while the enquiry was fresh, from a number families would recognise on later calls.
What did we build?
Before the call, a filter. An n8n workflow picks up new leads from the CRM and only calls those that pass six rules: a non-zero lead score and not an existing client; no contact from sales in the last 14 days; a summer football camp enquiry; from a parent, not a player; a budget above the minimum; and not a business proposal. It also pulls the programme's price, dates, weeks and level from the catalogue.
During the call. The agent introduces itself as an AI customer support agent, not sales, and works through the call one question at a time:
| Phase | What the agent does |
|---|---|
| 1. Introduction and verification | Confirms it's speaking to the parent, then the player's name and age, the parent's details and the programme, country, month and weeks |
| 2. Level and programme fit | Confirms the player's level and objective, and mentions higher-performance options if the player is stronger than the programme |
| 3. Consultant callback | If the family wants to hear about alternatives, takes the best times to call and ends the call |
| 4. Budget qualification | Only if the budget is well below the programme price: asks about a higher investment or shorter options |
| 5. Final intent | Summarises the choice and explains reservation details will follow by email |
| 6. Language routing | For Italian, German or Portuguese speakers, finds out whether English, Spanish or French suits for the consultant's follow-up |
After the call. A summary goes back to the CRM with a 1–10 lead score, the player's profile and level, programme interest and flexibility, budget, callback times, language and one of six next-step labels: call now, send proposal, send alternatives, nurture, disqualify or callback needed.
How does it run day to day?
Form submitted ─► CRM ─► n8n filters + programme data ─► VAPI call via Twilio (< 2 min) ─► summary, score, label ─► CRM updated
- Calls are short. The agent keeps calls under 8 minutes and ends once a consultant callback is arranged.
- No answer, one retry. The agent tries again 5–10 minutes later without leaving a voicemail. After a second miss, it creates a callback-needed task so sales makes one final attempt before switching to email.
- One consistent number. Calls go out from the same Twilio number, so families recognise it on later calls.
- English first. Phase 1 runs in English, with Spanish and French planned.
What were the results?
We don't have outcome figures to publish yet, such as pick-up or conversion rates. What we can state is how the agent works on every enquiry: the call starts within 2 minutes of the form, every answered call is scored 1–10 and given one of six next-step labels, and calls stay under 8 minutes.
- From form submission to the agent's callBuild spec, 2026
- 1–10Lead score on every answered callBuild spec, 2026
- Standard next-step labels for the sales teamBuild spec, 2026
What did we learn, and what are the limits?
- Map every branch before prompting. The order of the conversation mattered. Never ask "ready to book?" after a family has asked about alternatives. Mapping every path first avoided contradictory answers.
- End the call once it has done its job. Stopping once a callback is arranged prevents conflicting data later in the call.
- Handle budget with care. Raise it only when the gap is significant, and always offer alternatives rather than pressure.
- Get direct access routes early. Access requests routed through a busy founder slow everything down. Agree a named contact and turnaround times at kickoff.