The support team was good at the difficult conversations. Most of its time, though, went on the easy ones.
The challenge
Questions about where an order was, when it would arrive and how to return it made up most of the inbox. They arrived through email, website chat and social messages, and each channel had its own saved replies, which had drifted apart over time. Customers sometimes got different answers to the same question depending on where they asked.
What we built
A support system of ten agents, each responsible for one part of the journey:
- Routing. One agent reads every incoming message, identifies the channel, the customer and the intent, and sends it to the right place.
- Orders and delivery. Agents look up live order and tracking data and answer status questions directly.
- Returns. Agents check eligibility against the published returns policy and start the return when it qualifies.
- Product questions. Agents answer from product information and the approved knowledge base.
- Drafting. For anything unusual, an agent drafts a reply for a person to review.
- Handover. When a conversation needs a person, an agent passes it over with a summary and the full history attached.
All agents answer from a single knowledge base that the support lead owns and approves. When a question falls outside it, the agent says so and hands over rather than guessing.
How we delivered
The Blueprint listed the questions the agents would handle, the sources they could use and the conditions for handing over. We built in stages, starting with routing and order status running in shadow mode alongside the team, then returns and product questions once the answers matched what the team would have said. Every conversation is recorded, so the support lead can review exactly what each agent did.
What changed
Routine questions get a consistent answer from one approved source, whichever channel they arrive through. The team spends its time on the conversations that need a person, and those arrive with the context already gathered.
