OSCAR is the home-services app: more than 200 fixed-price services, from a leaking sink to a full apartment clean, booked in seconds, with a vetted professional at the door in as little as 30 minutes. It runs in more than 33 cities across Portugal, Spain, and the UK, completes more than 10,000 jobs a week, and holds a 4.7 rating across 15,000+ customer reviews. A promise that fast is held together by a human support team that follows every job in real time, around the clock.
That team serves two sides of the marketplace at once. Customers write in about a pro who arrived late or did not show, a service that fell short, a charge they did not expect, a refund that has not landed. Pros write in about registration steps, a payment they have not received, an account they need unblocked. Both arrive in Zendesk, by email and through the chat inside the app, and both need an answer in minutes.
Two sides, one queue, and the pattern nobody could see
Read one ticket at a time and every request looks like its own small emergency. What a support queue cannot tell you on its own is what keeps happening, and why. Is "service unconfirmed" a customer misunderstanding the app, or a step in the booking flow that fails? Is "refund not received" a payments issue, or an expectation the confirmation screen sets wrong? A team working the queue in real time has no time to find out, and reporting built for ticket counts tells you what was said, not what was needed.
So the same requests came back. Each one was handled well, and none of them went away.
Every conversation, named by what the person needed
Aide connected to OSCAR's Zendesk with one click and read the conversations OSCAR already had. The Customer Intent Engine trained a classifier on OSCAR's own data and produced its Customer Intent Map: 36 intents across six categories, nothing tagged by the team, in the language of OSCAR's operation rather than a templated approach.
From symptoms to root causes
The sixth category is the one that changed how OSCAR works. "Services not appearing." "Service unconfirmed." "Add payment method." Indicating product defects and friction points arriving disguised as tickets, from customers and pros who had no other way to say so. The same is true of intents filed under other headings: "wrongful cancelation" and "refund not received" describe an experience in the app before they describe a support conversation.
Aide reports each of them to OSCAR's product team by intent: how many, trending which way, with the actual conversations behind the number. This allows the product team to fix the flow. The intent's volume falls. The support team did not have to write a single report to make that happen.
"Aide has been incredibly helpful in detecting topics and reporting root causes of support requests over both email and our in-app chat, allowing us to optimize our product user experience and reduce the incoming support load. Our support team also starting to use Aide's AI agent, which is really amazing to see!"
Miguel Faria, OSCAR
This is the part of a queue that should shrink: not because a contact button was hidden or a bot absorbed the complaint, but because the reason to write in was removed. Every other request still gets a person, faster.
Drafts the agents actually send
For the conversations that do need a reply, Aide drafts it inside Zendesk, grounded in the detected intent and the conversation itself, and presents it to the agent. The agent reads, edits if needed, and sends. Nothing reaches a customer or a pro without a person deciding it should. That is the Agent Governance Engine at work: the AI acts inside the lines the team drew and hands everything else to a human. The edits, by OSCAR's own account, are getting smaller.
A map that keeps itself honest
A taxonomy built once goes stale the first time the business changes, and OSCAR changes constantly: new services, new cities, a new country. Aide's Continuous Learning Engine keeps the map current. It proposes new training examples from incoming conversations for the team to approve, hundreds queued at any time, flags examples that look like they belong under a different intent, and suggests which ones to drop. The team stays in command of every change, and the map gets sharper each month instead of drifting.
The result is not a smaller support team. It is a support team that can see the whole operation, both sides, by what people actually needed, and a product team that ships fixes from evidence instead of anecdotes.
The queue is not the goal. The goal is fewer reasons to write in, and a better answer for everyone who still does.
Transform your support operations with Aide
For more information or to schedule a demo, visit aide.app/demo or contact sales@aide.app.