Wingbuddy: AI Sales Email Assistant

Two Projects, One Problem

Sales teams spend hours writing emails, and the quality varies with experience. The communication skills a top performer has refined over years take years to acquire.

Wingbuddy is a travel agency whose sales team answers questions about a wide range of trip packages. We worked on that problem twice: first with a Chrome extension that improved what an agent had already written, then with a pipeline that writes the draft before the agent opens the email.

Project 1: A Chrome Extension for Reformulation

The extension sits inside the existing email workflow. A sales rep finishes a draft, wants it sharper, and clicks one button.

The AI applies company-specific best practices and, most usefully, has learned the writing style and personality of Wingbuddy’s best-performing sales representative. It reformulates on demand, keeping the original intent while improving clarity and persuasiveness.

What it changed: composition time dropped from 15-20 minutes to 5, across the sales team’s daily flow of emails. New reps got the codified best practices immediately instead of acquiring them over years. Prospects got consistent, professional writing.

Project 2: Real-Time Drafting on the Unified Platform

Reformulation still requires someone to write the first version. The second project removed that step.

Wingbuddy moved onto Flowful’s unified platform and adopted the mail app, a visual pipeline builder that automates email workflows with AI, connects to external tools, and acts on emails at every stage. The pipeline now writes a polished reply as the email arrives. Agents review and validate rather than compose.

The migration also mutualises processing across clients, which cuts cost, and gives Wingbuddy a base to adopt further apps without rebuilding.

How It Works: Classification Then Targeted Context

Accurate drafts need the right context and nothing more. Each node in the pipeline is modular, and the platform supports custom tool integrations, so the flow is shaped to the client:

1. Trigger and email filter: incoming mail is captured and filtered on allow lists, exclusion rules and labels, so only relevant messages enter the pipeline.
2. Classify: the AI reads the email, detects the language, whether it is a question, and which trip package it concerns.
3. FAQ filler: the node retrieves only the FAQ document for that package, which keeps the context lean and the cost down.
4. Gate: a condition decides whether the email warrants an automated draft at all.
5. Execute prompt: the selected FAQ content, the email body and the resolved prompt go to the language model, which writes the reply.
6. Create reply draft and label: the draft is created in the mailbox as a reply, and the email is labelled “AutoReplied” for tracking.

The Outcome

The draft is already there. It is written in real time as the mail arrives, so an agent opening a message has nothing to wait for and no context to rebuild.

Quality does not vary. Every draft uses the same structured knowledge base and the same model, so the tone is consistent across the whole team.

Cost stays down. Injecting one relevant FAQ rather than the entire knowledge base keeps token consumption and processing time low.

The team owns the knowledge. Wingbuddy updates the FAQ documents themselves through the platform, so the AI stays current without engineering time.

Built to Grow

The mail app is the first of several. The platform mutualises infrastructure across AI applications, and the same FAQ content and classification logic can carry future use cases, from chatbots to internal tools, without starting over.