A global experience-gifting platformExperience economy, curated leisure and gifting
Global Experience Concierge
A multi-agent concierge that takes a customer from a vague gift idea to a specific, bookable experience in the right city, across five countries and more than seventy cities, without adding support headcount.
Before
5 countriesCatalogues, pricing and availability that differ by region, so enquiries either tied up staff or bounced off vague answers and walked
After
2 to 3 sFor the full retrieval-and-response path from a single conversational entry point, down from 4 to 5 seconds
Overview
From a vague idea to a bookable experience.
A global experience-gifting platform operating across five countries and more than seventy cities was fielding a high volume of pre-booking enquiries where product catalogues, pricing and availability differ meaningfully by region. Seasonal spikes routinely outstripped support capacity.
NFI built a multi-agent orchestrated concierge that routes each enquiry by detected region and topic to a specialist, retrieves matching experiences from a large curated knowledge base, and guides the customer toward a booking link.
The problem
The right experience, in the right city, for the right person.
Something memorable for my parents' anniversary is not a search query, and a generic chatbot could not be trusted to match it to a specific experience in a specific region.
01
Regional variation
Products, pricing and availability differed across five countries and seventy cities.
02
Trust in the match
A generic assistant either tied up human staff or gave vague answers that customers walked away from.
03
Seasonal spikes
Peak periods outstripped support capacity, and headcount was not the answer.
The solution
Triage, specialists, and retrieval built for speed.
An orchestrator reads each enquiry and hands it to a regional specialist or a topic expert trained on the client's highest-volume enquiry categories, each speaking in a locally appropriate tone.
Specialists retrieve matching experiences from a large curated knowledge base using a hybrid semantic-plus-keyword search pipeline, generate qualifying questions, and guide customers toward booking links.
Query generation runs in parallel with retrieval to cut latency on the full response path, long conversations summarise themselves so memory stays coherent, and a multi-model fallback chain keeps the service available when individual providers degrade.
Multi-agent triage orchestratorRegional and topic specialist agentsHybrid semantic and keyword retrievalCurated experience knowledge baseParallel query generationAutomatic conversation summarisationMulti-model fallback chain
One entry point, seventy cities, seconds to answer.
70+Cities across five countries covered by regional and topic-expert specialists from one conversational entry point
HybridSemantic plus keyword retrieval over a large curated experience knowledge base
2 to 3 sFull retrieval-and-response path after parallelising query generation, down from 4 to 5 seconds
99.9%Uptime target supported by a multi-model fallback chain with graceful degradation
Questions
About this project.
A triage orchestrator detects the region and topic of each enquiry and routes it to a regional specialist or a topic expert, each speaking in a locally appropriate tone and retrieving matching experiences from a curated knowledge base.
The full retrieval-and-response path measures 2 to 3 seconds, down from 4 to 5 seconds, because query generation runs in parallel with hybrid semantic and keyword retrieval.
The concierge absorbs enquiry volume without adding support headcount, and a multi-model fallback chain supports a 99.9% uptime target with graceful degradation.
