This is the case that shaped the product. Transfeero did not have a volume problem; it had an audience problem. The same people, from the same desk, were answering a passenger asking "will the driver wait if my flight is late?" and a partner asking "when is September’s commission paid out?". Those are two different companies inside one company, and treating them as one is exactly how a partner-only rate card ends up in front of a customer.
Where it started
An airport transfer involves two people writing in for opposite reasons. The passenger wants to know where to find the driver, what happens if the flight is delayed, how to change the time. The partner wants to know how they get paid, which documents are needed to add a vehicle to the fleet, how the rate card works.
Before, all of it landed in the same inbox: WhatsApp on one side, email on the other, the website widget in between. Agents had to work out who they were talking to before they could even start answering, and a single body of knowledge had to serve two worlds that must not see each other.
How it was set up
The workspace is configured with two first-class audiences, "customers" and "partners". It is not a label on a contact: it is the dimension along which help centers, AI knowledge and answering behaviour are separated.
- Two distinct public help centers, each on its own subdomain, with their own articles and their own languages.
- Two AI agents, each grounded only in its own audience’s content: a document about partner commissions is not reachable from a conversation with a passenger.
- WhatsApp Business connected with one precise rule: the number the customer writes to decides that thread’s audience. The same person can be both a partner and a customer, and the two stay apart.
- Inbound and outbound email that re-attaches to the right thread, so a reply by mail does not open a new conversation.
- The widget on the site and the chat inside the app, with exactly the same experience.
An AI that reads the actual booking
The agent does not answer from articles alone: it is connected to Transfeero’s booking system. When a passenger asks about driver waiting time, the agent knows which booking is theirs, what kind of pickup point it is, and therefore how many minutes the driver really waits — the rule differs between an airport and a city address. The same connection shows a partner a different set of facts.
The details you only find by living in it
- On WhatsApp exactly one conversation per number stays live, because on the customer’s phone there is only one: when an agent opens a new one the previous is closed, and the new one carries a link back to the thread it continues.
- When an agent corrects a contact’s audience, that choice is stamped and no automatic classifier overwrites it again.
- The agent reads the photos customers send — the receipt, the pickup point, the damage — instead of replying that it cannot see images.
- Hand-off rules are written per audience: a booking change always goes to a person, a question about airport waiting time does not.
What changed
Agents no longer have to work out who they are talking to: the conversation arrives already labelled, with the contact’s history and, for passengers, the booking alongside it. The repetitive questions — driver waiting time, what to do on arrival, how to change a time — are handled by the agent on its own help center, and anything needing a decision reaches a person with the context already assembled.
The guarantee holding it all together is a negative one, and it is the most important: partner-only content is not "hidden" from a passenger. For that conversation it does not exist, because the separation lives in the query that retrieves content and in the database’s own rules, not in a checkbox in the interface.