Response time
Previously about 20 minutes; sometimes hours or days.
For a social media marketing agency, we built an AI agent that drafts replies in each brand’s voice, follows the conversation and suggests the next step of the sales script. The agency decides what it may send.
Results from the delivered project.
Previously about 20 minutes; sometimes hours or days.
Across the agency’s social and messaging channels.
Twice the share of conversations that end in an order.
The agency managed messages for several brands. A faster answer still needed the brand’s voice, the conversation’s context and the right sales step.
We connected the agency’s channels and brand accounts in one queue.
A fine-tuned model and script stages connect what the customer said to the next reply.
Drafts wait for a person. The agent sends automatically only where the agency permits it.
Review a reply, adjust the brand’s voice or follow the sales script. Explore each part of the same workspace.
The suggested reply stays beside the conversation. A manager can send, edit or discard it.
Drafts are the default. Automatic replies go out only in cases the agency has allowed.
Fine-tuning carried the agency’s phrasing and rhythm into generated drafts.
The script engine uses the conversation to choose a stage and its next sales step.
Channel integrations, model work and script stages connect to the team’s inbox.
Cursopic took the agent from channel integrations to a working inbox, then stayed with the team as it moved into production use.
Channel integrations and model fine-tuning came first. The inbox, voice settings, script stages and sending rules brought the workflow together for launch.
Delivered as a connected Python application for the agency’s team.
We monitored replies, adjusted rules and fixed the issues the team reported while managers got used to reviewing and editing drafts.
The team’s feedback shaped adjustments during the support month.