AI Assistant for Sales

CASE STUDY 2

ROLES

Interaction Designer, Conversation Designer, Facilitator

RESPONSIBILITIES

Competitive Analysis, Brainstorming Workshops, Early Concepts, Wireframes, Prototypes, User stories

COLLABORATION

Sales & Marketing Stakeholders, Sales Agents, Product Management, Engineering, QA, Data Science, Analytics, Content Strategy, Program Management, Salesforce Designer, Researcher, Visual Designer

TIME LINE

5 months

Context

Prospects and customers expect seamless, fast, friction-free experiences when engaging with a company like Autodesk. Today, the inbound sales phone strategy competes with the digital-first strategies of the support and education teams. Having phone numbers prominently placed on the Autodesk website drives around 20,000 inbound calls per quarter into the digital sales teams, however 45% of those calls are looking for product, order or education support. We need to align Autodesk's contact strategies so prospects and customers have consistent experiences when proactively contacting Autodesk throughout their lifecycle. 

Problem

How might we make it easy for customers with sales inquiries to connect with sales agents via chat and guide support & education inquiries out of the sales path and to the right place

Strategy

Autodesk Assistant will transform pre-sales phone based interactions into a conversational dialog with a chatbot. Whether the site visitor is new to Autodesk exploring our offerings or a seasoned customer looking for a new solution, the Autodesk Assistant will deliver a digital-first experience that was designed to meet market expectations.

  • New AI models for distinguishing sales inquiries from non-sales

  • Existing solutions for support & education

  • Robust contact platform

Discovery

I facilitated workshops with stakeholders from sales and e-store organizations to align on top customer intents. I also helped label 200 sample inputs for the data science team to train the AI routing models.

Design

Actions

  • Identify top sales intents

  • Lead with topic selection

  • Pass context during agent handoff

  • Build UX Library of components

  • Collect user feedback

Fallback options

  • Human agent always an option

  • Guide support/education inquiries out of sales path

  • Fallback to phone when chat is unavailable

Scope Change

  • New LLM + RAG for self-service

  • Assumed no UX needed

  • Conversational FAQs added to flow

  • Designed guardrails to minimize risk

  • Designed appropriate responses for conversation repair (error handling)

  • Limited number of turns

Early Results

Next Steps

  • Deep-dive into conversation logs

  • Include generative answers

  • Conversational memory

  • Improve formatting of answers

  • Topics or open-input or both

  • Revisit turn counts

  • Replace phone number with call-scheduling