Intermediate Agent

We built an Intermediate Agent prototype for a fictitious Marketplace, automating request handling between clients and vendors. It bridges language barriers, manages tasks such as date changes, and answers queries. This showcases the potential to streamline operations, reduce back-and-forth communication, and enhance efficiency.

Description

We’ve had the idea of an Intermediate Agent for marketplaces in our backlog for a very long time. So, when Cesc was invited to present at Phocuswright Europe in June, we took the opportunity to build it as an experiment and a demo!

Here is the use case we wanted to address:

• A marketplace sells travel packages with flights and hotels (leisure travel)
• The marketplace connects the client (travelers) and the vendor (hotel, airline)
• The team working at the marketplace receives and handles requests coming from both sides.
• In many cases, this brings a lot of added value, but in some, they are only qualifying and verifying the request, and forwarding them between both sides.

We saw that this last piece would be something an agent could help with.

In our prototype, we embody a marketplace called ACME Great Tours. It communicates with these 2 stakeholders:

1. The client speaks with the Marketplace via Telegram, in her native language, in this case English
2. Hotel Luis representatives interact with the Marketplace directly on their ticket manager (Freshdesk) with the agent in Spanish.

Our Intermediate Agent communicates with both the client and the hotel on their own communication channel - no need for them to use any specific tool. It is able to:

• Handle simple requests such as date changes.
• Answer generic and traveler-specific questions.

In the following demo, you will see how:

1. The Agent knows about the traveler and her bookings
2. It understands the traveler questions and requests in their language
3. It follows up with the client until the request is confirmed.
4. It translates, formats and sends the request to the hotel
5. It translates, formats and sends the request back to the client.

Here is a high-level overview of how it works

Potential

This prototype was only built to demonstrate the potential of such Agents. Obviously, there are a lot of things that need to be handled in a real world situation that we didn’t consider here. For example, validating the information is reliable and messages are sent in the right language, increasing the security, etc. It’s still in its infancy, but it’s widely promising.

Indeed, it’s becoming easier and easier to build this kind of agent. You could start with a simple case and iterate on it. Anytime this case happens, the agent can jump on it (supervised by a human?), and the rest is handled normally.

The potential is endless, when you imagine that the Agent could in the future:

• Know everything about a traveller - their bookings, preferences, constraints
• Create, modify, and cancel requests
• Answer any question you have previously answered or documented.
• Validate the information and documents provided

Those new skills could be introduced slowly, as you discover opportunities to reduce a high number of back-and-forths between the vendor and the client

👉 Do you find yourself in this situation of intermediate? Maybe there is a way to streamline the process!